fix ssh_box error parsing

.github/.codecov.yml

@@ -1,19 +1,14 @@
 codecov:
   notify:
     wait_for_ci: true
-    # our project is large, so 6 builds are typically uploaded. this waits till 5/6
-    # See https://docs.codecov.com/docs/notifications#section-preventing-notifications-until-after-n-builds
-    after_n_builds: 5
 
 coverage:
   status:
     patch:
       default:
-        threshold: 100% # allow patch coverage to be lower than project coverage by any amount
+        threshold: 10% # allow patch coverage to be lower than project coverage by at most 10%
     project:
       default:
         threshold: 5% # allow project coverage to drop at most 5%
 
 comment: false
-github_checks:
-    annotations: false

.github/ISSUE_TEMPLATE/bug_template.yml

@@ -12,7 +12,7 @@ body:
       label: Is there an existing issue for the same bug?
       description: Please check if an issue already exists for the bug you encountered.
       options:
-      - label: I have checked the troubleshooting document at https://docs.all-hands.dev/modules/usage/troubleshooting
+      - label: I have checked the troubleshooting document at https://opendevin.github.io/OpenDevin/modules/usage/troubleshooting
         required: true
       - label: I have checked the existing issues.
         required: true
@@ -28,8 +28,8 @@ body:
   - type: textarea
     id: current-version
     attributes:
-      label: Current OpenDevin version
-      description: What version of OpenDevin are you using? If you're running in docker, tell us the tag you're using (e.g. ghcr.io/opendevin/opendevin:0.3.1).
+      label: Current Version
+      description: What version are you using? If you're running in docker, tell us the tag you're using (e.g. ghcr.io/opendevin/opendevin:0.3.1).
       render: bash
     validations:
       required: true
@@ -52,12 +52,6 @@ body:
         - Model:
         - Agent:
 
-  - type: textarea
-    id: os-version
-    attributes:
-      label: Operating System
-      description: What Operating System are you using? Linux, Mac OS, WSL on Windows
-
   - type: textarea
     id: repro-steps
     attributes:
@@ -72,4 +66,4 @@ body:
     id: additional-context
     attributes:
       label: Logs, Errors, Screenshots, and Additional Context
-      description: If you want to share the chat history you can click the thumbs-down (👎) button above the input field and you will get a shareable link (you can also click thumbs up when things are going well of course!). LLM logs will be stored in the `logs/llm/default` folder. Please add any additional context about the problem here.
+      description: LLM logs will be stored in the `logs/llm/default` folder. Please add any additional context about the problem here.

.github/ISSUE_TEMPLATE/question.md

@@ -0,0 +1,16 @@
+---
+name: Question
+about: Use this template to ask a question regarding the project.
+title: ''
+labels: question
+assignees: ''
+
+---
+
+## Describe your question
+
+<!--A clear and concise description of what you want to know.-->
+
+## Additional context
+
+<!--Add any other context about the question here, like what you've tried so far.-->

.github/dependabot.yml

@@ -1,22 +0,0 @@
-# To get started with Dependabot version updates, you'll need to specify which
-# package ecosystems to update and where the package manifests are located.
-# Please see the documentation for all configuration options:
-# https://docs.github.com/code-security/dependabot/dependabot-version-updates/configuration-options-for-the-dependabot.yml-file
-
-version: 2
-updates:
-  - package-ecosystem: "pip" # See documentation for possible values
-    directory: "/" # Location of package manifests
-    schedule:
-      interval: "daily"
-    open-pull-requests-limit: 20
-  - package-ecosystem: "npm" # See documentation for possible values
-    directory: "/frontend" # Location of package manifests
-    schedule:
-      interval: "daily"
-    open-pull-requests-limit: 20
-  - package-ecosystem: "npm" # See documentation for possible values
-    directory: "/docs" # Location of package manifests
-    schedule:
-      interval: "daily"
-    open-pull-requests-limit: 20

.github/pull_request_template.md

@@ -1,8 +0,0 @@
-**What is the problem that this fixes or functionality that this introduces? Does it fix any open issues?**
-
----
-
-**Give a summary of what the PR does, explaining any non-trivial design decisions**
-
----
-**Other references**

.github/workflows/dummy-agent-test.yml

@@ -1,4 +1,4 @@
-name: Run E2E test with dummy agent
+name: Run e2e test with dummy agent
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
@@ -10,9 +10,6 @@ on:
     - main
   pull_request:
 
-env:
-  PERSIST_SANDBOX : "false"
-
 jobs:
   test:
     runs-on: ubuntu-latest
@@ -30,13 +27,4 @@ jobs:
           wget https://huggingface.co/BAAI/bge-small-en-v1.5/raw/main/1_Pooling/config.json -P /tmp/llama_index/models--BAAI--bge-small-en-v1.5/snapshots/5c38ec7c405ec4b44b94cc5a9bb96e735b38267a/1_Pooling/
       - name: Run tests
         run: |
-          set -e
-          poetry run python opendevin/core/main.py -t "do a flip" -d ./workspace/ -c DummyAgent
-      - name: Check exit code
-        run: |
-          if [ $? -ne 0 ]; then
-            echo "Test failed"
-            exit 1
-          else
-            echo "Test passed"
-          fi
+          poetry run python opendevin/core/main.py -t "do a flip" -m ollama/not-a-model -d ./workspace/ -c DummyAgent

.github/workflows/ghcr-runtime.yml

@@ -1,265 +0,0 @@
-name: Build Publish and Test Runtime Image
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
-  cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
-
-on:
-  push:
-    branches:
-      - main
-    tags:
-      - '*'
-  pull_request:
-  workflow_dispatch:
-    inputs:
-      reason:
-        description: 'Reason for manual trigger'
-        required: true
-        default: ''
-
-jobs:
-  ghcr_build_runtime:
-    runs-on: ubuntu-latest
-
-    outputs:
-      tags: ${{ steps.capture-tags.outputs.tags }}
-
-    permissions:
-      contents: read
-      packages: write
-
-    strategy:
-      matrix:
-        image: ["od_runtime"]
-        base_image: ["ubuntu:22.04"]
-        platform: ["amd64", "arm64"]
-
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-
-      - name: Free Disk Space (Ubuntu)
-        uses: jlumbroso/free-disk-space@main
-        with:
-          # this might remove tools that are actually needed,
-          # if set to "true" but frees about 6 GB
-          tool-cache: true
-          # all of these default to true, but feel free to set to
-          # "false" if necessary for your workflow
-          android: true
-          dotnet: true
-          haskell: true
-          large-packages: true
-          docker-images: false
-          swap-storage: true
-
-      - name: Set up QEMU
-        uses: docker/setup-qemu-action@v3
-
-      - name: Set up Docker Buildx
-        id: buildx
-        uses: docker/setup-buildx-action@v3
-
-      - name: Install poetry via pipx
-        run: pipx install poetry
-
-      - name: Set up Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.11"
-          cache: "poetry"
-
-      - name: Install Python dependencies using Poetry
-        run: make install-python-dependencies
-
-      - name: Create source distribution and Dockerfile
-        run: poetry run python3 opendevin/runtime/utils/runtime_build.py --base_image ${{ matrix.base_image }} --build_folder containers/runtime
-
-      - name: Build and export image
-        id: build
-        run: ./containers/build.sh ${{ matrix.image }} ${{ github.repository_owner }} ${{ matrix.platform }}
-
-      - name: Capture tags
-        id: capture-tags
-        run: |
-          tags=$(cat tags.txt)
-          echo "tags=$tags"
-          echo "tags=$tags" >> $GITHUB_OUTPUT
-
-      - name: Upload Docker image as artifact
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.image }}-docker-image-${{ matrix.platform }}
-          path: /tmp/${{ matrix.image }}_image_${{ matrix.platform }}.tar
-
-  test-for-runtime:
-    name: Test for Runtime
-    runs-on: ubuntu-latest
-    needs: ghcr_build_runtime
-    env:
-      PERSIST_SANDBOX: "false"
-    steps:
-      - uses: actions/checkout@v4
-
-      - name: Free Disk Space (Ubuntu)
-        uses: jlumbroso/free-disk-space@main
-        with:
-          # this might remove tools that are actually needed,
-          # when set to "true" but frees about 6 GB
-          tool-cache: true
-
-          # all of these default to true, but feel free to set to
-          # "false" if necessary for your workflow
-          android: true
-          dotnet: true
-          haskell: true
-          large-packages: true
-          swap-storage: true
-
-      - name: Install poetry via pipx
-        run: pipx install poetry
-
-      - name: Set up Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.11"
-          cache: "poetry"
-
-      - name: Install Python dependencies using Poetry
-        run: make install-python-dependencies
-
-      - name: Download Runtime Docker image
-        uses: actions/download-artifact@v4
-        with:
-          name: od_runtime-docker-image-amd64
-          path: /tmp/
-
-      - name: Load Runtime image and run runtime tests
-        run: |
-          # Load the Docker image and capture the output
-          output=$(docker load -i /tmp/od_runtime_image_amd64.tar)
-
-          # Extract the first image name from the output
-          image_name=$(echo "$output" | grep -oP 'Loaded image: \K.*' | head -n 1)
-
-          # Print the full name of the image
-          echo "Loaded Docker image: $image_name"
-
-          SANDBOX_CONTAINER_IMAGE=$image_name TEST_IN_CI=true poetry run pytest --cov=agenthub --cov=opendevin --cov-report=xml -s ./tests/unit/test_runtime.py
-
-      - name: Upload coverage to Codecov
-        uses: codecov/codecov-action@v4
-        env:
-          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
-
-  ghcr_push:
-    runs-on: ubuntu-latest
-    # don't push if runtime tests fail
-    needs: [ghcr_build_runtime, test-for-runtime]
-    if: github.ref == 'refs/heads/main' || startsWith(github.ref, 'refs/tags/')
-
-    env:
-      tags: ${{ needs.ghcr_build_runtime.outputs.tags }}
-
-    permissions:
-      contents: read
-      packages: write
-
-    strategy:
-      matrix:
-        image: ["od_runtime"]
-        platform: ["amd64", "arm64"]
-
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-
-      - name: Free Disk Space (Ubuntu)
-        uses: jlumbroso/free-disk-space@main
-        with:
-          tool-cache: true
-          android: true
-          dotnet: true
-          haskell: true
-          large-packages: true
-          docker-images: false
-          swap-storage: true
-
-      - name: Login to GHCR
-        uses: docker/login-action@v2
-        with:
-          registry: ghcr.io
-          username: ${{ github.repository_owner }}
-          password: ${{ secrets.GITHUB_TOKEN }}
-
-      - name: Download Docker images
-        uses: actions/download-artifact@v4
-        with:
-          name: ${{ matrix.image }}-docker-image-${{ matrix.platform }}
-          path: /tmp/${{ matrix.platform }}
-
-      - name: List downloaded files
-        run: |
-          ls -la /tmp/${{ matrix.platform }}
-          file /tmp/${{ matrix.platform }}/*
-
-      - name: Load images and push to registry
-        run: |
-          mv /tmp/${{ matrix.platform }}/${{ matrix.image }}_image_${{ matrix.platform }}.tar ./${{ matrix.image }}_image_${{ matrix.platform }}.tar
-          if ! loaded_image=$(docker load -i ${{ matrix.image }}_image_${{ matrix.platform }}.tar | grep "Loaded image:" | head -n 1 | awk '{print $3}'); then
-            echo "Failed to load Docker image"
-            exit 1
-          fi
-          echo "loaded image = $loaded_image"
-          tags=$(echo ${tags} | tr ' ' '\n')
-          image_name=$(echo "ghcr.io/${{ github.repository_owner }}/${{ matrix.image }}" | tr '[:upper:]' '[:lower:]')
-          echo "image name = $image_name"
-          for tag in $tags; do
-            echo "tag = $tag"
-            if [ -n "$image_name" ]; then
-              docker tag $loaded_image $image_name:${tag}_${{ matrix.platform }}
-              docker push $image_name:${tag}_${{ matrix.platform }}
-            else
-              echo "Skipping tag and push due to empty image_name"
-            fi
-          done
-
-  create_manifest:
-    runs-on: ubuntu-latest
-    needs: [ghcr_build_runtime, ghcr_push]
-    if: github.ref == 'refs/heads/main' || startsWith(github.ref, 'refs/tags/')
-
-    env:
-      tags: ${{ needs.ghcr_build_runtime.outputs.tags }}
-
-    strategy:
-      matrix:
-        image: ["od_runtime"]
-
-    permissions:
-      contents: read
-      packages: write
-
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-
-      - name: Login to GHCR
-        uses: docker/login-action@v2
-        with:
-          registry: ghcr.io
-          username: ${{ github.repository_owner }}
-          password: ${{ secrets.GITHUB_TOKEN }}
-
-      - name: Create and push multi-platform manifest
-        run: |
-          image_name=$(echo "ghcr.io/${{ github.repository_owner }}/${{ matrix.image }}" | tr '[:upper:]' '[:lower:]')
-          echo "image name = $image_name"
-          tags=$(echo ${tags} | tr ' ' '\n')
-          for tag in $tags; do
-            echo 'tag = $tag'
-            docker buildx imagetools create --tag $image_name:$tag \
-              $image_name:${tag}_amd64 \
-              $image_name:${tag}_arm64
-          done

.github/workflows/ghcr.yml

@@ -1,4 +1,4 @@
-name: Build Publish and Test Docker Image
+name: Publish Docker Image
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
@@ -7,7 +7,7 @@ concurrency:
 on:
   push:
     branches:
-      - main
+    - main
     tags:
       - '*'
   pull_request:
@@ -19,23 +19,19 @@ on:
         default: ''
 
 jobs:
-  ghcr_build:
+  ghcr_build_and_push:
     runs-on: ubuntu-latest
 
-    outputs:
-      tags: ${{ steps.capture-tags.outputs.tags }}
-
     permissions:
       contents: read
       packages: write
 
     strategy:
       matrix:
-        image: ["sandbox", "opendevin"]
-        platform: ["amd64", "arm64"]
+        image: ["app", "sandbox"]
 
     steps:
-      - name: Checkout
+      - name: checkout
         uses: actions/checkout@v4
 
       - name: Free Disk Space (Ubuntu)
@@ -44,6 +40,7 @@ jobs:
           # this might remove tools that are actually needed,
           # if set to "true" but frees about 6 GB
           tool-cache: true
+
           # all of these default to true, but feel free to set to
           # "false" if necessary for your workflow
           android: true
@@ -60,204 +57,26 @@ jobs:
         id: buildx
         uses: docker/setup-buildx-action@v3
 
-      - name: Build and export image
-        id: build
-        run: ./containers/build.sh ${{ matrix.image }} ${{ github.repository_owner }} ${{ matrix.platform }}
-
-      - name: Capture tags
-        id: capture-tags
-        run: |
-          tags=$(cat tags.txt)
-          echo "tags=$tags"
-          echo "tags=$tags" >> $GITHUB_OUTPUT
-
-      - name: Upload Docker image as artifact
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ matrix.image }}-docker-image-${{ matrix.platform }}
-          path: /tmp/${{ matrix.image }}_image_${{ matrix.platform }}.tar
-
-  test-for-sandbox:
-    name: Test for Sandbox
-    runs-on: ubuntu-latest
-    needs: ghcr_build
-    env:
-      PERSIST_SANDBOX: "false"
-    steps:
-      - uses: actions/checkout@v4
-
-      - name: Install poetry via pipx
-        run: pipx install poetry
-
-      - name: Set up Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.11"
-          cache: "poetry"
-
-      - name: Install Python dependencies using Poetry
-        run: make install-python-dependencies
-
-      - name: Download sandbox Docker image
-        uses: actions/download-artifact@v4
-        with:
-          name: sandbox-docker-image-amd64
-          path: /tmp/
-
-      - name: Load sandbox image and run sandbox tests
-        run: |
-          # Load the Docker image and capture the output
-          output=$(docker load -i /tmp/sandbox_image_amd64.tar)
-
-          # Extract the first image name from the output
-          image_name=$(echo "$output" | grep -oP 'Loaded image: \K.*' | head -n 1)
-
-          # Print the full name of the image
-          echo "Loaded Docker image: $image_name"
-
-          SANDBOX_CONTAINER_IMAGE=$image_name TEST_IN_CI=true poetry run pytest --cov=agenthub --cov=opendevin --cov-report=xml -s ./tests/unit/test_sandbox.py
-
-      - name: Upload coverage to Codecov
-        uses: codecov/codecov-action@v4
-        env:
-          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
-
-  integration-tests-on-linux:
-    name: Integration Tests on Linux
-    runs-on: ubuntu-latest
-    needs: ghcr_build
-    env:
-      PERSIST_SANDBOX: "false"
-    strategy:
-      fail-fast: false
-      matrix:
-        python-version: ["3.11"]
-        sandbox: ["ssh", "local"]
-    steps:
-      - uses: actions/checkout@v4
-
-      - name: Install poetry via pipx
-        run: pipx install poetry
-
-      - name: Set up Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: ${{ matrix.python-version }}
-          cache: 'poetry'
-
-      - name: Install Python dependencies using Poetry
-        run: make install-python-dependencies
-
-      - name: Download sandbox Docker image
-        uses: actions/download-artifact@v4
-        with:
-          name: sandbox-docker-image-amd64
-          path: /tmp/
-
-      - name: Load sandbox image and run integration tests
-        env:
-          SANDBOX_BOX_TYPE: ${{ matrix.sandbox }}
-        run: |
-          # Load the Docker image and capture the output
-          output=$(docker load -i /tmp/sandbox_image_amd64.tar)
-
-          # Extract the first image name from the output
-          image_name=$(echo "$output" | grep -oP 'Loaded image: \K.*' | head -n 1)
-
-          # Print the full name of the image
-          echo "Loaded Docker image: $image_name"
-
-          SANDBOX_CONTAINER_IMAGE=$image_name TEST_IN_CI=true TEST_ONLY=true ./tests/integration/regenerate.sh
-
-      - name: Upload coverage to Codecov
-        uses: codecov/codecov-action@v4
-        env:
-          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
-
-  ghcr_push:
-    runs-on: ubuntu-latest
-    # don't push if integration tests or sandbox tests fail
-    needs: [ghcr_build, integration-tests-on-linux, test-for-sandbox]
-    if: github.ref == 'refs/heads/main' || startsWith(github.ref, 'refs/tags/')
-
-    env:
-      tags: ${{ needs.ghcr_build.outputs.tags }}
-
-    permissions:
-      contents: read
-      packages: write
-
-    strategy:
-      matrix:
-        image: ["sandbox", "opendevin"]
-        platform: ["amd64", "arm64"]
-
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-
-      - name: Login to GHCR
-        uses: docker/login-action@v2
+      - name: Login to ghcr
+        uses: docker/login-action@v1
         with:
           registry: ghcr.io
           username: ${{ github.repository_owner }}
           password: ${{ secrets.GITHUB_TOKEN }}
 
-      - name: Download Docker images
-        uses: actions/download-artifact@v4
-        with:
-          name: ${{ matrix.image }}-docker-image-${{ matrix.platform }}
-          path: /tmp/${{ matrix.platform }}
-
-      - name: Load images and push to registry
+      - name: Build and push ${{ matrix.image }}
+        if: "!github.event.pull_request.head.repo.fork"
         run: |
-          mv /tmp/${{ matrix.platform }}/${{ matrix.image }}_image_${{ matrix.platform }}.tar .
-          loaded_image=$(docker load -i ${{ matrix.image }}_image_${{ matrix.platform }}.tar | grep "Loaded image:" | head -n 1 | awk '{print $3}')
-          echo "loaded image = $loaded_image"
-          tags=$(echo ${tags} | tr ' ' '\n')
-          image_name=$(echo "ghcr.io/${{ github.repository_owner }}/${{ matrix.image }}" | tr '[:upper:]' '[:lower:]')
-          echo "image name = $image_name"
-          for tag in $tags; do
-            echo "tag = $tag"
-            docker tag $loaded_image $image_name:${tag}_${{ matrix.platform }}
-            docker push $image_name:${tag}_${{ matrix.platform }}
-          done
+          ./containers/build.sh ${{ matrix.image }} ${{ github.repository_owner }} --push
 
-  create_manifest:
+      - name: Build ${{ matrix.image }}
+        if: "github.event.pull_request.head.repo.fork"
+        run: |
+          ./containers/build.sh ${{ matrix.image }} ${{ github.repository_owner }}
+
+  docker_build_success:
+    name: Docker Build Success
     runs-on: ubuntu-latest
-    needs: [ghcr_build, ghcr_push]
-    if: github.ref == 'refs/heads/main' || startsWith(github.ref, 'refs/tags/')
-
-    env:
-      tags: ${{ needs.ghcr_build.outputs.tags }}
-
-    strategy:
-      matrix:
-        image: ["sandbox", "opendevin"]
-
-    permissions:
-      contents: read
-      packages: write
-
+    needs: ghcr_build_and_push
     steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-
-      - name: Login to GHCR
-        uses: docker/login-action@v2
-        with:
-          registry: ghcr.io
-          username: ${{ github.repository_owner }}
-          password: ${{ secrets.GITHUB_TOKEN }}
-
-      - name: Create and push multi-platform manifest
-        run: |
-          image_name=$(echo "ghcr.io/${{ github.repository_owner }}/${{ matrix.image }}" | tr '[:upper:]' '[:lower:]')
-          echo "image name = $image_name"
-          tags=$(echo ${tags} | tr ' ' '\n')
-          for tag in $tags; do
-            echo 'tag = $tag'
-            docker buildx imagetools create --tag $image_name:$tag \
-              $image_name:${tag}_amd64 \
-              $image_name:${tag}_arm64
-          done
+    - run: echo Done!

.github/workflows/lint.yml

@@ -47,4 +47,11 @@ jobs:
       - name: Install pre-commit
         run: pip install pre-commit==3.7.0
       - name: Run pre-commit hooks
-        run: pre-commit run --files opendevin/**/* agenthub/**/* evaluation/**/* tests/**/* --show-diff-on-failure --config ./dev_config/python/.pre-commit-config.yaml
+        if: github.ref != 'refs/heads/main'
+        run: |
+          git fetch https://github.com/OpenDevin/OpenDevin.git main:main && \
+          pre-commit run \
+            --files \
+            $(git diff --name-only $(git merge-base main $(git branch --show-current)) $(git branch --show-current) | tr '\n' ' ') \
+            --show-diff-on-failure \
+            --config ./dev_config/python/.pre-commit-config.yaml

.github/workflows/review-pr.yml

@@ -12,15 +12,15 @@ jobs:
   dogfood:
     if: contains(github.event.pull_request.labels.*.name, 'review-this')
     runs-on: ubuntu-latest
+    container:
+      image: ghcr.io/opendevin/opendevin
+      volumes:
+        - /var/run/docker.sock:/var/run/docker.sock
+
     steps:
-    - uses: actions/checkout@v4
-    - name: Set up Python
-      uses: actions/setup-python@v5
-      with:
-        python-version: '3.11'
     - name: install git, github cli
       run: |
-        sudo apt-get install -y git gh
+        apt-get install -y git gh
         git config --global --add safe.directory $PWD
 
     - name: Checkout Repository
@@ -34,9 +34,7 @@ jobs:
 
     - name: Write Task File
       run: |
-        echo "Your coworker wants to apply a pull request to this project." > task.txt
-        echo "Read and review ${{ github.event.pull_request.number }}.diff file. Create a review-${{ github.event.pull_request.number }}.txt and write your concise comments and suggestions there." >> task.txt
-        echo "Do not ask me for confirmation at any point." >> task.txt
+        echo "Your coworker wants to apply a pull request to this project. Read and review ${{ github.event.pull_request.number }}.diff file. Create a review-${{ github.event.pull_request.number }}.txt and write your concise comments and suggestions there." > task.txt
         echo "" >> task.txt
         echo "Title" >> task.txt
         echo "${{ github.event.pull_request.title }}" >> task.txt
@@ -46,26 +44,12 @@ jobs:
         echo "" >> task.txt
         echo "Diff file is: ${{ github.event.pull_request.number }}.diff" >> task.txt
 
-    - name: Set up environment
-      run: |
-        curl -sSL https://install.python-poetry.org | python3 -
-        export PATH="/github/home/.local/bin:$PATH"
-        poetry install --without evaluation
-        poetry run playwright install --with-deps chromium
-
     - name: Run OpenDevin
       env:
-        LLM_API_KEY: ${{ secrets.LLM_API_KEY }}
-        LLM_MODEL: ${{ vars.LLM_MODEL }}
-        SANDBOX_BOX_TYPE: ssh
+        LLM_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+        SANDBOX_TYPE: exec
       run: |
-        # Append path to launch poetry
-        export PATH="/github/home/.local/bin:$PATH"
-        # Append path to correctly import package, note: must set pwd at first
-        export PYTHONPATH=$(pwd):$PYTHONPATH
-        export WORKSPACE_MOUNT_PATH=$GITHUB_WORKSPACE
-        export WORKSPACE_BASE=$GITHUB_WORKSPACE
-        echo -e "/exit\n" | poetry run python opendevin/core/main.py -i 50 -f task.txt
+        WORKSPACE_MOUNT_PATH=$GITHUB_WORKSPACE python ./opendevin/core/main.py -i 50 -f task.txt -d $GITHUB_WORKSPACE
         rm task.txt
 
     - name: Check if review file is non-empty

.github/workflows/run-integration-tests.yml

@@ -0,0 +1,76 @@
+name: Run Integration Tests
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+
+on:
+  push:
+    branches:
+    - main
+  pull_request:
+
+jobs:
+  integration-tests:
+    name: Integration Tests
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ["3.11"]
+        agent: ["SWEAgent", "PlannerAgent", "MonologueAgent", "CodeActAgent"]
+        sandbox: ["ssh", "exec"]
+        include:
+          - agent: "MonologueAgent"
+            embedding-model: "local"
+          - agent: "MonologueAgent"
+            # sufficient to have one agent testing against local sandbox
+            sandbox: "local"
+            embedding-model: "local"
+          - agent: "SWEAgent"
+            embedding-model: "none"
+          - agent: "PlannerAgent"
+            embedding-model: "none"
+          - agent: "CodeActAgent"
+            embedding-model: "none"
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install poetry via pipx
+        run: pipx install poetry
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.11'
+          cache: 'poetry'
+
+      - name: Install Python dependencies using Poetry
+        run: poetry install
+
+      - name: Build Environment
+        run: make build
+
+      - name: Run Integration Tests
+        env:
+          SANDBOX_TYPE: ${{ matrix.sandbox }}
+          AGENT: ${{ matrix.agent }}
+          MAX_ITERATIONS: 10
+          LLM_EMBEDDING_MODEL: ${{ matrix.embedding-model }}
+        run: |
+          rm -rf workspace
+          mkdir workspace
+          WORKSPACE_BASE="$GITHUB_WORKSPACE/workspace" \
+            WORKSPACE_MOUNT_PATH="$GITHUB_WORKSPACE/workspace" \
+            poetry run pytest --cov=agenthub --cov=opendevin --cov-report=xml \
+            -s ./tests/integration
+      - name: Upload coverage to Codecov
+        uses: codecov/codecov-action@v4
+        env:
+          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
+  test_matrix_success:
+    name: All Integration Tests Passed
+    runs-on: ubuntu-latest
+    needs: [integration-tests]
+    steps:
+    - run: echo Done!

.github/workflows/run-unit-tests.yml

@@ -7,52 +7,15 @@ concurrency:
 on:
   push:
     branches:
-      - main
-    paths-ignore:
-      - '**/*.md'
-      - 'frontend/**'
-      - 'docs/**'
-      - 'evaluation/**'
+    - main
   pull_request:
 
-env:
-  PERSIST_SANDBOX : "false"
-
 jobs:
-  fe-test:
-    runs-on: ubuntu-latest
-
-    strategy:
-      matrix:
-        node-version: [20]
-
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-
-      - name: Set up Node.js
-        uses: actions/setup-node@v4
-        with:
-          node-version: ${{ matrix.node-version }}
-
-      - name: Install dependencies
-        working-directory: ./frontend
-        run: npm ci
-
-      - name: Run tests and collect coverage
-        working-directory: ./frontend
-        run: npm run test:coverage
-
-      - name: Upload coverage to Codecov
-        uses: codecov/codecov-action@v4
-        env:
-          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
-
   test-on-macos:
     name: Test on macOS
-    runs-on: macos-12
+    runs-on: macos-13
     env:
-      INSTALL_DOCKER: "1" # Set to '0' to skip Docker installation
+      INSTALL_DOCKER: '0'  # Set to '0' to skip Docker installation
     strategy:
       matrix:
         python-version: ["3.11"]
@@ -67,7 +30,7 @@ jobs:
         uses: actions/setup-python@v5
         with:
           python-version: ${{ matrix.python-version }}
-          cache: "poetry"
+          cache: 'poetry'
 
       - name: Install Python dependencies using Poetry
         run: poetry install
@@ -75,47 +38,8 @@ jobs:
       - name: Install & Start Docker
         if: env.INSTALL_DOCKER == '1'
         run: |
-          INSTANCE_NAME="colima-${GITHUB_RUN_ID}"
-
-          # Uninstall colima to upgrade to the latest version
-          if brew list colima &>/dev/null; then
-            brew uninstall colima
-            # unlinking colima dependency: go
-            brew uninstall go@1.21
-          fi
-          rm -rf ~/.colima ~/.lima
-          brew install --HEAD colima
-          brew install docker
-
-          start_colima() {
-            # Find a free port in the range 10000-20000
-            RANDOM_PORT=$((RANDOM % 10001 + 10000))
-
-            # Original line:
-            if ! colima start --network-address --arch x86_64 --cpu=1 --memory=1 --verbose --ssh-port $RANDOM_PORT; then
-              echo "Failed to start Colima."
-              return 1
-            fi
-            return 0
-          }
-
-          # Attempt to start Colima for 5 total attempts:
-          ATTEMPT_LIMIT=5
-          for ((i=1; i<=ATTEMPT_LIMIT; i++)); do
-
-            if start_colima; then
-              echo "Colima started successfully."
-              break
-            else
-              colima stop -f
-              sleep 10
-              colima delete -f
-              if [ $i -eq $ATTEMPT_LIMIT ]; then
-                exit 1
-              fi
-              sleep 10
-            fi
-          done
+          brew install colima docker
+          colima start
 
           # For testcontainers to find the Colima socket
           # https://github.com/abiosoft/colima/blob/main/docs/FAQ.md#cannot-connect-to-the-docker-daemon-at-unixvarrundockersock-is-the-docker-daemon-running
@@ -125,7 +49,7 @@ jobs:
         run: make build
 
       - name: Run Tests
-        run: poetry run pytest --forked --cov=agenthub --cov=opendevin --cov-report=xml ./tests/unit -k "not test_sandbox.py and not test_runtime.py"
+        run: poetry run pytest --cov=agenthub --cov=opendevin --cov-report=xml ./tests/unit -k "not test_sandbox"
 
       - name: Upload coverage to Codecov
         uses: codecov/codecov-action@v4
@@ -135,7 +59,7 @@ jobs:
     name: Test on Linux
     runs-on: ubuntu-latest
     env:
-      INSTALL_DOCKER: "0" # Set to '0' to skip Docker installation
+      INSTALL_DOCKER: '0'  # Set to '0' to skip Docker installation
     strategy:
       matrix:
         python-version: ["3.11"]
@@ -150,7 +74,7 @@ jobs:
         uses: actions/setup-python@v5
         with:
           python-version: ${{ matrix.python-version }}
-          cache: "poetry"
+          cache: 'poetry'
 
       - name: Install Python dependencies using Poetry
         run: poetry install --without evaluation
@@ -159,9 +83,45 @@ jobs:
         run: make build
 
       - name: Run Tests
-        run: poetry run pytest --forked --cov=agenthub --cov=opendevin --cov-report=xml ./tests/unit -k "not test_sandbox.py and not test_runtime.py"
+        run: poetry run pytest --cov=agenthub --cov=opendevin --cov-report=xml ./tests/unit -k "not test_sandbox"
 
       - name: Upload coverage to Codecov
         uses: codecov/codecov-action@v4
         env:
           CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
+
+  test-for-sandbox:
+    name: Test for Sandbox
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install poetry via pipx
+        run: pipx install poetry
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.11'
+          cache: 'poetry'
+
+      - name: Install Python dependencies using Poetry
+        run: poetry install
+
+      - name: Build Environment
+        run: make build
+
+      - name: Run Integration Test for Sandbox
+        run: |
+          poetry run pytest --cov=agenthub --cov=opendevin --cov-report=xml -s ./tests/unit/test_sandbox.py
+
+      - name: Upload coverage to Codecov
+        uses: codecov/codecov-action@v4
+        env:
+          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
+  test_matrix_success:
+    name: All Mac/Linux Tests Passed
+    runs-on: ubuntu-latest
+    needs: [test-on-macos, test-on-linux, test-for-sandbox]
+    steps:
+    - run: echo Done!

.github/workflows/solve-issue.yml

@@ -36,27 +36,14 @@ jobs:
         echo "BODY:" >> task.txt
         echo "${ISSUE_BODY}" >> task.txt
 
-    - name: Set up environment
-      run: |
-        curl -sSL https://install.python-poetry.org | python3 -
-        export PATH="/github/home/.local/bin:$PATH"
-        poetry install --without evaluation
-        poetry run playwright install --with-deps chromium
-
-
     - name: Run OpenDevin
       env:
         ISSUE_TITLE: ${{ github.event.issue.title }}
         ISSUE_BODY: ${{ github.event.issue.body }}
         LLM_API_KEY: ${{ secrets.OPENAI_API_KEY }}
-        OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
-        SANDBOX_BOX_TYPE: ssh
+        SANDBOX_TYPE: exec
       run: |
-        # Append path to launch poetry
-        export PATH="/github/home/.local/bin:$PATH"
-        # Append path to correctly import package, note: must set pwd at first
-        export PYTHONPATH=$(pwd):$PYTHONPATH
-        WORKSPACE_MOUNT_PATH=$GITHUB_WORKSPACE poetry run python ./opendevin/core/main.py -i 50 -f task.txt -d $GITHUB_WORKSPACE
+        WORKSPACE_MOUNT_PATH=$GITHUB_WORKSPACE python ./opendevin/core/main.py -i 50 -f task.txt -d $GITHUB_WORKSPACE
         rm task.txt
 
     - name: Setup Git, Create Branch, and Commit Changes

.github/workflows/update-pyproject-version.yml

@@ -1,48 +0,0 @@
-name: Update pyproject.toml Version and Tags
-
-on:
-  release:
-    types:
-      - published
-
-jobs:
-  update-pyproject-and-tags:
-    runs-on: ubuntu-latest
-
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0  # Fetch all history for all branches and tags
-
-      - name: Set up Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.11"
-
-      - name: Install dependencies
-        run: |
-          python -m pip install --upgrade pip
-          pip install toml
-
-      - name: Get release tag
-        id: get_release_tag
-        run: echo "RELEASE_TAG=${GITHUB_REF#refs/tags/}" >> $GITHUB_ENV
-
-      - name: Update pyproject.toml with release tag
-        run: |
-          python -c "
-          import toml
-          with open('pyproject.toml', 'r') as f:
-              data = toml.load(f)
-          data['tool']['poetry']['version'] = '${{ env.RELEASE_TAG }}'
-          with open('pyproject.toml', 'w') as f:
-              toml.dump(data, f)
-          "
-
-      - name: Commit and push pyproject.toml changes
-        uses: stefanzweifel/git-auto-commit-action@v4
-        with:
-          commit_message: "Update pyproject.toml version to ${{ env.RELEASE_TAG }}"
-          branch: main
-          file_pattern: pyproject.toml

.gitignore

@@ -126,7 +126,6 @@ env/
 venv/
 ENV/
 env.bak/
-.env.bak
 venv.bak/
 *venv/
 
@@ -161,14 +160,9 @@ cython_debug/
 #  option (not recommended) you can uncomment the following to ignore the entire idea folder.
 .idea/
 .vscode/
-.cursorignore
 
 # evaluation
-evaluation/evaluation_outputs
-evaluation/outputs
-evaluation/swe_bench/eval_workspace*
 evaluation/SWE-bench/data
-evaluation/webarena/scripts/webarena_env.sh
 
 # frontend
 
@@ -181,8 +175,6 @@ frontend/yarn.lock
 
 # testing
 frontend/coverage
-test_results*
-/_test_files_tmp/
 
 # production
 frontend/build
@@ -210,17 +202,8 @@ cache
 
 # configuration
 config.toml
-config.toml_
-config.toml.bak
-
-containers/agnostic_sandbox
-
-# swe-bench-eval
-image_build_logs
-run_instance_logs
-
-od_runtime_*.tar
-
-# docker build
-containers/runtime/Dockerfile
-containers/runtime/project.tar.gz
+evaluation/swe_bench/eval_workspace
+evaluation/outputs
+evaluation/evaluation_outputs
+test_results*
+/_test_files_tmp/

CODE_OF_CONDUCT.md

@@ -61,7 +61,7 @@ representative at an online or offline event.
 
 Instances of abusive, harassing, or otherwise unacceptable behavior may be
 reported to the community leaders responsible for enforcement at
-contact@all-hands.dev
+contact@rbren.io
 All complaints will be reviewed and investigated promptly and fairly.
 
 All community leaders are obligated to respect the privacy and security of the

CONTRIBUTING.md

@@ -1,32 +1,13 @@
 # Contributing
 
 Thanks for your interest in contributing to OpenDevin! We welcome and appreciate contributions.
+To report bugs, create a [GitHub issue](https://github.com/OpenDevin/OpenDevin/issues/new/choose).
 
-## How Can I Contribute?
-
-There are many ways that you can contribute:
-
-1. **Download and use** OpenDevin, and send [issues](https://github.com/OpenDevin/OpenDevin/issues) when you encounter something that isn't working or a feature that you'd like to see.
-2. **Send feedback** after each session by [clicking the thumbs-up thumbs-down buttons](https://docs.all-hands.dev/modules/usage/feedback), so we can see where things are working and failing, and also build an open dataset for training code agents.
-3. **Improve the Codebase** by sending PRs (see details below). In particular, we have some [good first issue](https://github.com/OpenDevin/OpenDevin/labels/good%20first%20issue) issues that may be ones to start on.
-
-## Understanding OpenDevin's CodeBase
-
-To understand the codebase, please refer to the README in each module:
-- [frontend](./frontend/README.md)
-- [agenthub](./agenthub/README.md)
-- [evaluation](./evaluation/README.md)
-- [opendevin](./opendevin/README.md)
-    - [server](./opendevin/server/README.md)
-
-When you write code, it is also good to write tests. Please navigate to the `tests` folder to see existing test suites.
-At the moment, we have two kinds of tests: `unit` and `integration`. Please refer to the README for each test suite. These tests also run on GitHub's continuous integration to ensure quality of the project.
-
-## Sending Pull Requests to OpenDevin
-
+## Contribution Guide
 ### 1. Fork the Official Repository
-Fork the [OpenDevin repository](https://github.com/OpenDevin/OpenDevin) into your own account.
-Clone your own forked repository into your local environment:
+
+Fork [OpenDevin repository](https://github.com/OpenDevin/OpenDevin) into your own account.
+Clone your own forked repository into your local environment.
 
 ```shell
 git clone git@github.com:<YOUR-USERNAME>/OpenDevin.git
@@ -35,23 +16,21 @@ git clone git@github.com:<YOUR-USERNAME>/OpenDevin.git
 ### 2. Configure Git
 
 Set the official repository as your [upstream](https://www.atlassian.com/git/tutorials/git-forks-and-upstreams) to synchronize with the latest update in the official repository.
-Add the original repository as upstream:
+Add the original repository as upstream
 
 ```shell
 cd OpenDevin
 git remote add upstream git@github.com:OpenDevin/OpenDevin.git
 ```
 
-Verify that the remote is set:
-
+Verify that the remote is set.
 ```shell
 git remote -v
 ```
-
 You should see both `origin` and `upstream` in the output.
 
 ### 3. Synchronize with Official Repository
-Synchronize latest commit with official repository before coding:
+Synchronize latest commit with official repository before coding.
 
 ```shell
 git fetch upstream
@@ -60,48 +39,25 @@ git merge upstream/main
 git push origin main
 ```
 
-### 4. Set up the Development Environment
+### 4. Create a New Branch And Open a Pull Request
+After you finish implementation, open forked repository. The source branch is your new branch, and the target branch is `OpenDevin/OpenDevin` `main` branch. Then PR should appears in [OpenDevin PRs](https://github.com/OpenDevin/OpenDevin/pulls).
 
-We have a separate doc [Development.md](https://github.com/OpenDevin/OpenDevin/blob/main/Development.md) that tells you how to set up a development workflow.
-
-### 5. Write Code and Commit It
-
-Once you have done this, you can write code, test it, and commit it to a branch (replace `my_branch` with an appropriate name):
-
-```shell
-git checkout -b my_branch
-git add .
-git commit
-git push origin my_branch
-```
-
-### 6. Open a Pull Request
-
-* On GitHub, go to the page of your forked repository, and create a Pull Request:
-   - Click on `Branches`
-   - Click on the `...` beside your branch and click on `New pull request`
-   - Set `base repository` to `OpenDevin/OpenDevin`
-   - Set `base` to `main`
-   - Click `Create pull request`
-
-The PR should appear in [OpenDevin PRs](https://github.com/OpenDevin/OpenDevin/pulls).
-
-Then the OpenDevin team will review your code.
+Then OpenDevin team will review your code.
 
 ## PR Rules
 
 ### 1. Pull Request title
-As described [here](https://github.com/commitizen/conventional-commit-types/blob/master/index.json), a valid PR title should begin with one of the following prefixes:
+
+As described in [here](https://github.com/commitizen/conventional-commit-types/blob/master/index.json), a valid PR title should begin with one of the following prefixes:
 
 - `feat`: A new feature
 - `fix`: A bug fix
-- `docs`: Documentation only changes
-- `style`: Changes that do not affect the meaning of the code (white space, formatting, missing semicolons, etc.)
+- `doc`: Documentation only changes
 - `refactor`: A code change that neither fixes a bug nor adds a feature
+- `style`: A refactoring that improves code style
 - `perf`: A code change that improves performance
 - `test`: Adding missing tests or correcting existing tests
-- `build`: Changes that affect the build system or external dependencies (example scopes: gulp, broccoli, npm)
-- `ci`: Changes to our CI configuration files and scripts (example scopes: Travis, Circle, BrowserStack, SauceLabs)
+- `ci`: Changes to CI configuration files and scripts (example scopes: `.github`, `ci` (Buildkite))
 - `chore`: Other changes that don't modify src or test files
 - `revert`: Reverts a previous commit
 
@@ -111,6 +67,24 @@ For example, a PR title could be:
 
 You may also check out previous PRs in the [PR list](https://github.com/OpenDevin/OpenDevin/pulls).
 
+As described in [here](https://github.com/OpenDevin/OpenDevin/labels), we create several labels. Every PR should be tagged with the corresponding labels.
+
 ### 2. Pull Request description
+
 - If your PR is small (such as a typo fix), you can go brief.
-- If it contains a lot of changes, it's better to write more details.
+- If it is large and you have changed a lot, it's better to write more details.
+
+
+## How to begin
+Please refer to the README in each module:
+- [frontend](./frontend/README.md)
+- [agenthub](./agenthub/README.md)
+- [evaluation](./evaluation/README.md)
+- [opendevin](./opendevin/README.md)
+    - [server](./opendevin/server/README.md)
+    - [mock server](./opendevin/mock/README.md)
+
+## Tests
+Please navigate to `tests` folder to see existing test suites.
+At the moment, we have two kinds of tests: `unit` and `integration`. Please refer to the README for each test suite. These tests also run on CI to ensure quality of
+the project.

Development.md

@@ -1,16 +1,14 @@
 # Development Guide
 This guide is for people working on OpenDevin and editing the source code.
-If you wish to contribute your changes, check out the [CONTRIBUTING.md](https://github.com/OpenDevin/OpenDevin/blob/main/CONTRIBUTING.md) on how to clone and setup the project initially before moving on.
-Otherwise, you can clone the OpenDevin project directly.
 
 ## Start the server for development
+
 ### 1. Requirements
-* Linux, Mac OS, or [WSL on Windows](https://learn.microsoft.com/en-us/windows/wsl/install)  [ Ubuntu <= 22.04]
-* [Docker](https://docs.docker.com/engine/install/) (For those on MacOS, make sure to allow the default Docker socket to be used from advanced settings!)
-* [Python](https://www.python.org/downloads/) = 3.11
+* Linux, Mac OS, or [WSL on Windows](https://learn.microsoft.com/en-us/windows/wsl/install)
+* [Docker](https://docs.docker.com/engine/install/)(For those on MacOS, make sure to allow the default Docker socket to be used from advanced settings!)
+* [Python](https://www.python.org/downloads/) >= 3.11
 * [NodeJS](https://nodejs.org/en/download/package-manager) >= 18.17.1
 * [Poetry](https://python-poetry.org/docs/#installing-with-the-official-installer) >= 1.8
-* netcat => sudo apt-get install netcat
 
 Make sure you have all these dependencies installed before moving on to `make build`.
 
@@ -29,40 +27,41 @@ mamba install conda-forge::poetry
 ```
 
 ### 2. Build and Setup The Environment
-Begin by building the project which includes setting up the environment and installing dependencies. This step ensures that OpenDevin is ready to run on your system:
 
-```bash
-make build
-```
+- **Build the Project:** Begin by building the project, which includes setting up the environment and installing dependencies. This step ensures that OpenDevin is ready to run smoothly on your system.
+    ```bash
+    make build
+    ```
 
 ### 3. Configuring the Language Model
+
 OpenDevin supports a diverse array of Language Models (LMs) through the powerful [litellm](https://docs.litellm.ai) library. By default, we've chosen the mighty GPT-4 from OpenAI as our go-to model, but the world is your oyster! You can unleash the potential of Anthropic's suave Claude, the enigmatic Llama, or any other LM that piques your interest.
 
-To configure the LM of your choice, run:
+To configure the LM of your choice, follow these steps:
 
+1. **Using the Makefile: The Effortless Approach**
+   With a single command, you can have a smooth LM setup for your OpenDevin experience. Simply run:
    ```bash
    make setup-config
    ```
-
    This command will prompt you to enter the LLM API key, model name, and other variables ensuring that OpenDevin is tailored to your specific needs. Note that the model name will apply only when you run headless. If you use the UI, please set the model in the UI.
 
-   Note: If you have previously run OpenDevin using the docker command, you may have already set some environmental variables in your terminal. The final configurations are set from highest to lowest priority:
-   Environment variables > config.toml variables > default variables
-
 **Note on Alternative Models:**
-Some alternative models may prove more challenging to tame than others. Fear not, brave adventurer! We shall soon unveil LLM-specific documentation to guide you on your quest.
-And if you've already mastered the art of wielding a model other than OpenAI's GPT, we encourage you to share your setup instructions with us by creating instructions and adding it [to our documentation](https://github.com/OpenDevin/OpenDevin/tree/main/docs/modules/usage/llms).
+Some alternative models may prove more challenging to tame than others. Fear not, brave adventurer! We shall soon unveil LLM-specific documentation to guide you on your quest. And if you've already mastered the art of wielding a model other than OpenAI's GPT, we encourage you to [share your setup instructions with us](https://github.com/OpenDevin/OpenDevin/issues/417).
 
 For a full list of the LM providers and models available, please consult the [litellm documentation](https://docs.litellm.ai/docs/providers).
 
-### 4. Running the application
-#### Option A: Run the Full Application
-Once the setup is complete, launching OpenDevin is as simple as running a single command. This command starts both the backend and frontend servers seamlessly, allowing you to interact with OpenDevin:
-```bash
-make run
-```
+There is also [documentation for running with local models using ollama](./docs/documentation/LOCAL_LLM_GUIDE.md).
+
+### 4. Run the Application
+
+- **Run the Application:** Once the setup is complete, launching OpenDevin is as simple as running a single command. This command starts both the backend and frontend servers seamlessly, allowing you to interact with OpenDevin without any hassle.
+    ```bash
+    make run
+    ```
+
+### 5. Individual Server Startup
 
-#### Option B: Individual Server Startup
 - **Start the Backend Server:** If you prefer, you can start the backend server independently to focus on backend-related tasks or configurations.
     ```bash
     make start-backend
@@ -74,26 +73,24 @@ make run
     ```
 
 ### 6. LLM Debugging
-If you encounter any issues with the Language Model (LM) or you're simply curious, you can inspect the actual LLM prompts and responses. To do so, export DEBUG=1 in the environment and restart the backend.
-OpenDevin will then log the prompts and responses in the logs/llm/CURRENT_DATE directory, allowing you to identify the causes.
+
+If you encounter any issues with the Language Model (LM) or you're simply curious, you can inspect the actual LLM prompts and responses. To do so, export DEBUG=1 in the environment and restart the backend. OpenDevin will then log the prompts and responses in the logs/llm/CURRENT_DATE directory, allowing you to identify the causes.
 
 ### 7. Help
-Need assistance or information on available targets and commands? The help command provides all the necessary guidance to ensure a smooth experience with OpenDevin.
-```bash
-make help
- ```
+
+- **Get Some Help:** Need assistance or information on available targets and commands? The help command provides all the necessary guidance to ensure a smooth experience with OpenDevin.
+    ```bash
+    make help
+    ```
 
 ### 8. Testing
-To run tests, refer to the following:
+
 #### Unit tests
 
 ```bash
-poetry run pytest ./tests/unit/test_*.py
+poetry run pytest ./tests/unit/test_sandbox.py
 ```
 
 #### Integration tests
-Please refer to [this README](./tests/integration/README.md) for details.
 
-### 9. Add or update dependency
-1. Add your dependency in `pyproject.toml` or use `poetry add xxx`
-2. Update the poetry.lock file via `poetry lock --no-update`
+Please refer to [this README](./tests/integration/README.md) for details.

ISSUE_TRIAGE.md

@@ -1,25 +0,0 @@
-# Issue Triage
-These are the procedures and guidelines on how issues are triaged in this repo by the maintainers.
-
-## General
-* Most issues must be tagged with **enhancement** or **bug**
-* Issues may be tagged with what it relates to (**backend**, **frontend**, **agent quality**, etc.)
-
-## Severity
-* **Low**: Minor issues, single user report
-* **Medium**: Affecting multiple users
-* **Critical**: Affecting all users or potential security issues
-
-## Effort
-* Issues may be estimated with effort required (**small effort**, **medium effort**, **large effort**)
-
-## Difficulty
-* Issues with low implementation difficulty may be tagged with **good first issue**
-
-## Not Enough Information
-* User is asked to provide more information (logs, how to reproduce, etc.) when the issue is not clear
-* If an issue is unclear and the author does not provide more information or respond to a request, the issue may be closed as **not planned** (Usually after a week)
-
-## Multiple Requests/Fixes in One Issue
-* These issues will be narrowed down to one request/fix so the issue is more easily tracked and fixed
-* Issues may be broken down into multiple issues if required

Makefile

@@ -2,15 +2,14 @@ SHELL=/bin/bash
 # Makefile for OpenDevin project
 
 # Variables
-DOCKER_IMAGE = ghcr.io/opendevin/sandbox:main
+DOCKER_IMAGE = ghcr.io/opendevin/sandbox
 BACKEND_PORT = 3000
 BACKEND_HOST = "127.0.0.1:$(BACKEND_PORT)"
 FRONTEND_PORT = 3001
 DEFAULT_WORKSPACE_DIR = "./workspace"
-DEFAULT_MODEL = "gpt-4o"
+DEFAULT_MODEL = "gpt-3.5-turbo"
 CONFIG_FILE = config.toml
-PRE_COMMIT_CONFIG_PATH = "./dev_config/python/.pre-commit-config.yaml"
-PYTHON_VERSION = 3.11
+PRECOMMIT_CONFIG_PATH = "./dev_config/python/.pre-commit-config.yaml"
 
 # ANSI color codes
 GREEN=$(shell tput -Txterm setaf 2)
@@ -28,7 +27,7 @@ ifeq ($(INSTALL_DOCKER),)
 endif
 	@$(MAKE) -s install-python-dependencies
 	@$(MAKE) -s install-frontend-dependencies
-	@$(MAKE) -s install-pre-commit-hooks
+	@$(MAKE) -s install-precommit-hooks
 	@$(MAKE) -s build-frontend
 	@echo "$(GREEN)Build completed successfully.$(RESET)"
 
@@ -63,10 +62,10 @@ check-system:
 
 check-python:
 	@echo "$(YELLOW)Checking Python installation...$(RESET)"
-	@if command -v python$(PYTHON_VERSION) > /dev/null; then \
-		echo "$(BLUE)$(shell python$(PYTHON_VERSION) --version) is already installed.$(RESET)"; \
+	@if command -v python3.11 > /dev/null; then \
+		echo "$(BLUE)$(shell python3.11 --version) is already installed.$(RESET)"; \
 	else \
-		echo "$(RED)Python $(PYTHON_VERSION) is not installed. Please install Python $(PYTHON_VERSION) to continue.$(RESET)"; \
+		echo "$(RED)Python 3.11 is not installed. Please install Python 3.11 to continue.$(RESET)"; \
 		exit 1; \
 	fi
 
@@ -113,13 +112,13 @@ check-poetry:
 			echo "$(BLUE)$(shell poetry --version) is already installed.$(RESET)"; \
 		else \
 			echo "$(RED)Poetry 1.8 or later is required. You can install poetry by running the following command, then adding Poetry to your PATH:"; \
-			echo "$(RED) curl -sSL https://install.python-poetry.org | python$(PYTHON_VERSION) -$(RESET)"; \
+			echo "$(RED) curl -sSL https://install.python-poetry.org | python3 -$(RESET)"; \
 			echo "$(RED)More detail here: https://python-poetry.org/docs/#installing-with-the-official-installer$(RESET)"; \
 			exit 1; \
 		fi; \
 	else \
 		echo "$(RED)Poetry is not installed. You can install poetry by running the following command, then adding Poetry to your PATH:"; \
-		echo "$(RED) curl -sSL https://install.python-poetry.org | python$(PYTHON_VERSION) -$(RESET)"; \
+		echo "$(RED) curl -sSL https://install.python-poetry.org | python3.11 -$(RESET)"; \
 		echo "$(RED)More detail here: https://python-poetry.org/docs/#installing-with-the-official-installer$(RESET)"; \
 		exit 1; \
 	fi
@@ -131,13 +130,8 @@ pull-docker-image:
 
 install-python-dependencies:
 	@echo "$(GREEN)Installing Python dependencies...$(RESET)"
-	@if [ -z "${TZ}" ]; then \
-		echo "Defaulting TZ (timezone) to UTC"; \
-		export TZ="UTC"; \
-	fi
-	poetry env use python$(PYTHON_VERSION)
 	@if [ "$(shell uname)" = "Darwin" ]; then \
-		echo "$(BLUE)Installing chroma-hnswlib...$(RESET)"; \
+		echo "$(BLUE)Installing `chroma-hnswlib`...$(RESET)"; \
 		export HNSWLIB_NO_NATIVE=1; \
 		poetry run pip install chroma-hnswlib; \
 	fi
@@ -147,14 +141,7 @@ install-python-dependencies:
 		poetry run pip install playwright; \
 		poetry run playwright install chromium; \
 	else \
-		if [ ! -f cache/playwright_chromium_is_installed.txt ]; then \
-			echo "Running playwright install --with-deps chromium..."; \
-			poetry run playwright install --with-deps chromium; \
-			mkdir -p cache; \
-			touch cache/playwright_chromium_is_installed.txt; \
-		else \
-			echo "Setup already done. Skipping playwright installation."; \
-		fi \
+		poetry run playwright install --with-deps chromium; \
 	fi
 	@echo "$(GREEN)Python dependencies installed successfully.$(RESET)"
 
@@ -162,19 +149,22 @@ install-frontend-dependencies:
 	@echo "$(YELLOW)Setting up frontend environment...$(RESET)"
 	@echo "$(YELLOW)Detect Node.js version...$(RESET)"
 	@cd frontend && node ./scripts/detect-node-version.js
-	echo "$(BLUE)Installing frontend dependencies with npm...$(RESET)"
-	@cd frontend && npm install
+	@cd frontend && \
+		echo "$(BLUE)Installing frontend dependencies with npm...$(RESET)" && \
+		npm install && \
+		echo "$(BLUE)Running make-i18n with npm...$(RESET)" && \
+		npm run make-i18n
 	@echo "$(GREEN)Frontend dependencies installed successfully.$(RESET)"
 
-install-pre-commit-hooks:
+install-precommit-hooks:
 	@echo "$(YELLOW)Installing pre-commit hooks...$(RESET)"
 	@git config --unset-all core.hooksPath || true
-	@poetry run pre-commit install --config $(PRE_COMMIT_CONFIG_PATH)
+	@poetry run pre-commit install --config $(PRECOMMIT_CONFIG_PATH)
 	@echo "$(GREEN)Pre-commit hooks installed successfully.$(RESET)"
 
 lint-backend:
 	@echo "$(YELLOW)Running linters...$(RESET)"
-	@poetry run pre-commit run --files opendevin/**/* agenthub/**/* evaluation/**/* --show-diff-on-failure --config $(PRE_COMMIT_CONFIG_PATH)
+	@poetry run pre-commit run --files $$(git diff --name-only $$(git merge-base main $$(git branch --show-current)) $$(git branch --show-current) | tr '\n' ' ') --show-diff-on-failure --config $(PRECOMMIT_CONFIG_PATH)
 
 lint-frontend:
 	@echo "$(YELLOW)Running linters for frontend...$(RESET)"
@@ -205,10 +195,11 @@ start-frontend:
 	@echo "$(YELLOW)Starting frontend...$(RESET)"
 	@cd frontend && VITE_BACKEND_HOST=$(BACKEND_HOST) VITE_FRONTEND_PORT=$(FRONTEND_PORT) npm run start
 
-# Common setup for running the app (non-callable)
-_run_setup:
+# Run the app
+run:
+	@echo "$(YELLOW)Running the app...$(RESET)"
 	@if [ "$(OS)" = "Windows_NT" ]; then \
-		echo "$(RED) Windows is not supported, use WSL instead!$(RESET)"; \
+		echo "$(RED)`make run` is not supported on Windows. Please run `make start-frontend` and `make start-backend` separately.$(RESET)"; \
 		exit 1; \
 	fi
 	@mkdir -p logs
@@ -217,21 +208,9 @@ _run_setup:
 	@echo "$(YELLOW)Waiting for the backend to start...$(RESET)"
 	@until nc -z localhost $(BACKEND_PORT); do sleep 0.1; done
 	@echo "$(GREEN)Backend started successfully.$(RESET)"
-
-# Run the app (standard mode)
-run:
-	@echo "$(YELLOW)Running the app...$(RESET)"
-	@$(MAKE) -s _run_setup
 	@cd frontend && echo "$(BLUE)Starting frontend with npm...$(RESET)" && npm run start -- --port $(FRONTEND_PORT)
 	@echo "$(GREEN)Application started successfully.$(RESET)"
 
-# Run the app (WSL mode)
-run-wsl:
-	@echo "$(YELLOW)Running the app in WSL mode...$(RESET)"
-	@$(MAKE) -s _run_setup
-	@cd frontend && echo "$(BLUE)Starting frontend with npm (WSL mode)...$(RESET)" && npm run dev_wsl -- --port $(FRONTEND_PORT)
-	@echo "$(GREEN)Application started successfully in WSL mode.$(RESET)"
-
 # Setup config.toml
 setup-config:
 	@echo "$(YELLOW)Setting up config.toml...$(RESET)"
@@ -242,20 +221,10 @@ setup-config:
 setup-config-prompts:
 	@echo "[core]" > $(CONFIG_FILE).tmp
 
-	@read -p "Enter your workspace directory (as absolute path) [default: $(DEFAULT_WORKSPACE_DIR)]: " workspace_dir; \
+	@read -p "Enter your workspace directory [default: $(DEFAULT_WORKSPACE_DIR)]: " workspace_dir; \
 	 workspace_dir=$${workspace_dir:-$(DEFAULT_WORKSPACE_DIR)}; \
 	 echo "workspace_base=\"$$workspace_dir\"" >> $(CONFIG_FILE).tmp
 
-	@read -p "Do you want to persist the sandbox container? [true/false] [default: false]: " persist_sandbox; \
-	 persist_sandbox=$${persist_sandbox:-false}; \
-	 if [ "$$persist_sandbox" = "true" ]; then \
-		 read -p "Enter a password for the sandbox container: " ssh_password; \
-		 echo "ssh_password=\"$$ssh_password\"" >> $(CONFIG_FILE).tmp; \
-		 echo "persist_sandbox=$$persist_sandbox" >> $(CONFIG_FILE).tmp; \
-	 else \
-		echo "persist_sandbox=$$persist_sandbox" >> $(CONFIG_FILE).tmp; \
-	 fi
-
 	@echo "" >> $(CONFIG_FILE).tmp
 
 	@echo "[llm]" >> $(CONFIG_FILE).tmp
@@ -316,4 +285,4 @@ help:
 	@echo "  $(GREEN)help$(RESET)                - Display this help message, providing information on available targets."
 
 # Phony targets
-.PHONY: build check-dependencies check-python check-npm check-docker check-poetry pull-docker-image install-python-dependencies install-frontend-dependencies install-pre-commit-hooks lint start-backend start-frontend run run-wsl setup-config setup-config-prompts help
+.PHONY: build check-dependencies check-python check-npm check-docker check-poetry pull-docker-image install-python-dependencies install-frontend-dependencies install-precommit-hooks lint start-backend start-frontend run setup-config setup-config-prompts help

README.md

@@ -26,17 +26,15 @@
   <br/>
   <a href="https://join.slack.com/t/opendevin/shared_invite/zt-2i1iqdag6-bVmvamiPA9EZUu7oCO6KhA"><img src="https://img.shields.io/badge/Slack-Join%20Us-red?logo=slack&logoColor=white&style=for-the-badge" alt="Join our Slack community"></a>
   <a href="https://discord.gg/ESHStjSjD4"><img src="https://img.shields.io/badge/Discord-Join%20Us-purple?logo=discord&logoColor=white&style=for-the-badge" alt="Join our Discord community"></a>
-  <a href="https://codecov.io/github/opendevin/opendevin?branch=main"><img alt="CodeCov" src="https://img.shields.io/codecov/c/github/opendevin/opendevin?style=for-the-badge"></a>
+  <br/>
+  <a href="https://xwang.dev/blog/2024/opendevin-codeact-1.0-swebench/"><img src="https://img.shields.io/badge/SWE--bench%20Lite-21.0%25-green?style=for-the-badge" alt="SWE-bench "></a>
 </div>
 
 <!-- PROJECT LOGO -->
 <div align="center">
   <img src="./docs/static/img/logo.png" alt="Logo" width="200" height="200">
   <h1 align="center">OpenDevin: Code Less, Make More</h1>
-  <a href="https://docs.all-hands.dev/modules/usage/intro"><img src="https://img.shields.io/badge/Documentation-OpenDevin-blue?logo=googledocs&logoColor=white&style=for-the-badge" alt="Check out the documentation"></a>
-  <a href="https://arxiv.org/abs/2407.16741"><img src="https://img.shields.io/badge/Paper-%20on%20Arxiv-red?logo=arxiv&style=for-the-badge" alt="Paper on Arxiv"></a>
-  <br/>
-  <a href="https://huggingface.co/spaces/OpenDevin/evaluation"><img src="https://img.shields.io/badge/Evaluation-Benchmark%20on%20HF%20Space-green?logo=huggingface&style=for-the-badge" alt="Evaluation Benchmark"></a>
+  <a href="https://opendevin.github.io/OpenDevin/"><img src="https://img.shields.io/badge/Documenation-OpenDevin-blue?logo=googledocs&logoColor=white&style=for-the-badge" alt="Check out the documentation"></a>
 </div>
 <hr>
 
@@ -46,18 +44,16 @@ OpenDevin agents collaborate with human developers to write code, fix bugs, and
 
 ![App screenshot](./docs/static/img/screenshot.png)
 
-## ⚡ Getting Started
-OpenDevin works best with Docker version 26.0.0+ (Docker Desktop 4.31.0+).
-You must be using Linux, Mac OS, or WSL on Windows.
-
-To start OpenDevin in a docker container, run the following commands in your terminal:
-
-> [!WARNING]
-> When you run the following command, files in `./workspace` may be modified or deleted.
+## ⚡ Quick Start
+You can run OpenDevin with Docker. It works best with the most recent
+version of Docker, `26.0.0`.
 
 ```bash
-WORKSPACE_BASE=$(pwd)/workspace
-docker run -it \
+#The directory you want OpenDevin to modify. MUST be an absolute path!
+export WORKSPACE_BASE=$(pwd)/workspace;
+
+docker run \
+    -it \
     --pull=always \
     -e SANDBOX_USER_ID=$(id -u) \
     -e WORKSPACE_MOUNT_PATH=$WORKSPACE_BASE \
@@ -65,31 +61,13 @@ docker run -it \
     -v /var/run/docker.sock:/var/run/docker.sock \
     -p 3000:3000 \
     --add-host host.docker.internal:host-gateway \
-    --name opendevin-app-$(date +%Y%m%d%H%M%S) \
-    ghcr.io/opendevin/opendevin
+    ghcr.io/opendevin/opendevin:0.5
 ```
 
-> [!NOTE]
-> By default, this command pulls the `latest` tag, which represents the most recent release of OpenDevin. You have other options as well:
-> - For a specific release version, use `ghcr.io/opendevin/opendevin:<OpenDevin_version>` (replace <OpenDevin_version> with the desired version number).
-> - For the most up-to-date development version, use `ghcr.io/opendevin/opendevin:main`. This version may be **(unstable!)** and is recommended for testing or development purposes only.
->
-> Choose the tag that best suits your needs based on stability requirements and desired features.
-
-You'll find OpenDevin running at [http://localhost:3000](http://localhost:3000) with access to `./workspace`. To have OpenDevin operate on your code, place it in `./workspace`.
-OpenDevin will only have access to this workspace folder. The rest of your system will not be affected as it runs in a secured docker sandbox.
-
-Upon opening OpenDevin, you must select the appropriate `Model` and enter the `API Key` within the settings that should pop up automatically. These can be set at any time by selecting
-the `Settings` button (gear icon) in the UI. If the required `Model` does not exist in the list, you can manually enter it in the text box.
-
-For the development workflow, see [Development.md](https://github.com/OpenDevin/OpenDevin/blob/main/Development.md).
-
-Are you having trouble? Check out our [Troubleshooting Guide](https://docs.all-hands.dev/modules/usage/troubleshooting).
-
 ## 🚀 Documentation
 
 To learn more about the project, and for tips on using OpenDevin,
-**check out our [documentation](https://docs.all-hands.dev/modules/usage/intro)**.
+**check out our [documentation](https://opendevin.github.io/OpenDevin/)**.
 
 There you'll find resources on how to use different LLM providers (like ollama and Anthropic's Claude),
 troubleshooting resources, and advanced configuration options.
@@ -111,10 +89,15 @@ For details, please check [CONTRIBUTING.md](./CONTRIBUTING.md).
 Whether you're a developer, a researcher, or simply enthusiastic about OpenDevin, we'd love to have you in our community.
 Let's make software engineering better together!
 
-- [Slack workspace](https://join.slack.com/t/opendevin/shared_invite/zt-2jsrl32uf-fTeeFjNyNYxqSZt5NPY3fA) - Here we talk about research, architecture, and future development.
+- [Slack workspace](https://join.slack.com/t/opendevin/shared_invite/zt-2ggtwn3k5-PvAA2LUmqGHVZ~XzGq~ILw) - Here we talk about research, architecture, and future development.
 - [Discord server](https://discord.gg/ESHStjSjD4) - This is a community-run server for general discussion, questions, and feedback.
 
 ## 📈 Progress
+<p align="center">
+    <a href="https://www.swebench.com/lite.html">
+        <img src="/docs/static/img/results.png" alt="SWE-Bench Lite Score" width="500" height="auto">
+    </a>
+</p>
 
 <p align="center">
   <a href="https://star-history.com/#OpenDevin/OpenDevin&Date">
@@ -136,17 +119,3 @@ Distributed under the MIT License. See [`LICENSE`](./LICENSE) for more informati
 [issues-url]: https://github.com/OpenDevin/OpenDevin/issues
 [license-shield]: https://img.shields.io/github/license/opendevin/opendevin?style=for-the-badge
 [license-url]: https://github.com/OpenDevin/OpenDevin/blob/main/LICENSE
-
-## 📚 Cite
-
-```
-@misc{opendevin,
-      title={{OpenDevin: An Open Platform for AI Software Developers as Generalist Agents}}, 
-      author={Xingyao Wang and Boxuan Li and Yufan Song and Frank F. Xu and Xiangru Tang and Mingchen Zhuge and Jiayi Pan and Yueqi Song and Bowen Li and Jaskirat Singh and Hoang H. Tran and Fuqiang Li and Ren Ma and Mingzhang Zheng and Bill Qian and Yanjun Shao and Niklas Muennighoff and Yizhe Zhang and Binyuan Hui and Junyang Lin and Robert Brennan and Hao Peng and Heng Ji and Graham Neubig},
-      year={2024},
-      eprint={2407.16741},
-      archivePrefix={arXiv},
-      primaryClass={cs.SE},
-      url={https://arxiv.org/abs/2407.16741}, 
-}
-```

agenthub/README.md

@@ -1,16 +1,16 @@
-# Agent Hub
+# Agent Framework Research
 
 In this folder, there may exist multiple implementations of `Agent` that will be used by the framework.
 
-For example, `agenthub/codeact_agent`, etc.
+For example, `agenthub/monologue_agent`, `agenthub/metagpt_agent`, `agenthub/codeact_agent`, etc.
 Contributors from different backgrounds and interests can choose to contribute to any (or all!) of these directions.
 
 ## Constructing an Agent
 
-The abstraction for an agent can be found [here](../opendevin/controller/agent.py).
+The abstraction for an agent can be found [here](../opendevin/agent.py).
 
 Agents are run inside of a loop. At each iteration, `agent.step()` is called with a
-[State](../opendevin/controller/state/state.py) input, and the agent must output an [Action](../opendevin/events/action).
+[State](../opendevin/state.py) input, and the agent must output an [Action](../opendevin/action).
 
 Every agent also has a `self.llm` which it can use to interact with the LLM configured by the user.
 See the [LiteLLM docs for `self.llm.completion`](https://docs.litellm.ai/docs/completion).
@@ -28,33 +28,39 @@ The `state` contains:
 
 Here is a list of available Actions, which can be returned by `agent.step()`:
 
-- [`CmdRunAction`](../opendevin/events/action/commands.py) - Runs a command inside a sandboxed terminal
-- [`IPythonRunCellAction`](../opendevin/events/action/commands.py) - Execute a block of Python code interactively (in Jupyter notebook) and receives `CmdOutputObservation`. Requires setting up `jupyter` [plugin](../opendevin/runtime/plugins) as a requirement.
-- [`FileReadAction`](../opendevin/events/action/files.py) - Reads the content of a file
-- [`FileWriteAction`](../opendevin/events/action/files.py) - Writes new content to a file
-- [`BrowseURLAction`](../opendevin/events/action/browse.py) - Gets the content of a URL
-- [`AddTaskAction`](../opendevin/events/action/tasks.py) - Adds a subtask to the plan
-- [`ModifyTaskAction`](../opendevin/events/action/tasks.py) - Changes the state of a subtask.
-- [`AgentFinishAction`](../opendevin/events/action/agent.py) - Stops the control loop, allowing the user/delegator agent to enter a new task
-- [`AgentRejectAction`](../opendevin/events/action/agent.py) - Stops the control loop, allowing the user/delegator agent to enter a new task
-- [`AgentFinishAction`](../opendevin/events/action/agent.py) - Stops the control loop, allowing the user to enter a new task
-- [`MessageAction`](../opendevin/events/action/message.py) - Represents a message from an agent or the user
+- [`CmdRunAction`](../opendevin/action/bash.py) - Runs a command inside a sandboxed terminal
+- [`CmdKillAction`](../opendevin/action/bash.py) - Kills a background command
+- [`IPythonRunCellAction`](../opendevin/action/bash.py) - Execute a block of Python code interactively (in Jupyter notebook) and receives `CmdOutputObservation`. Requires setting up `jupyter` [plugin](../opendevin/sandbox/plugins) as a requirement.
+- [`FileReadAction`](../opendevin/action/fileop.py) - Reads the content of a file
+- [`FileWriteAction`](../opendevin/action/fileop.py) - Writes new content to a file
+- [`BrowseURLAction`](../opendevin/action/browse.py) - Gets the content of a URL
+- [`AgentRecallAction`](../opendevin/action/agent.py) - Searches memory (e.g. a vector database)
+- [`AddTaskAction`](../opendevin/action/tasks.py) - Adds a subtask to the plan
+- [`ModifyTaskAction`](../opendevin/action/tasks.py) - Changes the state of a subtask
+- [`AgentThinkAction`](../opendevin/action/agent.py) - A no-op that allows the agent to add plaintext to the history (as well as the chat log)
+- [`AgentTalkAction`](../opendevin/action/agent.py) - A no-op that allows the agent to add plaintext to the history and talk to the user.
+- [`AgentFinishAction`](../opendevin/action/agent.py) - Stops the control loop, allowing the user/delegator agent to enter a new task
+- [`AgentRejectAction`](../opendevin/action/agent.py) - Stops the control loop, allowing the user/delegator agent to enter a new task
+- [`AgentFinishAction`](../opendevin/action/agent.py) - Stops the control loop, allowing the user to enter a new task
+- [`MessageAction`](../opendevin/action/message.py) - Represents a message from an agent or the user
 
 You can use `action.to_dict()` and `action_from_dict` to serialize and deserialize actions.
 
 ## Observations
 
 There are also several types of Observations. These are typically available in the step following the corresponding Action.
-But they may also appear as a result of asynchronous events (e.g. a message from the user).
+But they may also appear as a result of asynchronous events (e.g. a message from the user, logs from a command running
+in the background).
 
 Here is a list of available Observations:
 
-- [`CmdOutputObservation`](../opendevin/events/observation/commands.py)
-- [`BrowserOutputObservation`](../opendevin/events/observation/browse.py)
-- [`FileReadObservation`](../opendevin/events/observation/files.py)
-- [`FileWriteObservation`](../opendevin/events/observation/files.py)
-- [`ErrorObservation`](../opendevin/events/observation/error.py)
-- [`SuccessObservation`](../opendevin/events/observation/success.py)
+- [`CmdOutputObservation`](../opendevin/observation/run.py)
+- [`BrowserOutputObservation`](../opendevin/observation/browse.py)
+- [`FileReadObservation`](../opendevin/observation/files.py)
+- [`FileWriteObservation`](../opendevin/observation/files.py)
+- [`AgentRecallObservation`](../opendevin/observation/recall.py)
+- [`ErrorObservation`](../opendevin/observation/error.py)
+- [`SuccessObservation`](../opendevin/observation/success.py)
 
 You can use `observation.to_dict()` and `observation_from_dict` to serialize and deserialize observations.
 
@@ -70,3 +76,14 @@ def step(self, state: "State") -> "Action"
 
 `step` moves the agent forward one step towards its goal. This probably means
 sending a prompt to the LLM, then parsing the response into an `Action`.
+
+### `search_memory`
+
+```
+def search_memory(self, query: str) -> list[str]:
+```
+
+`search_memory` should return a list of events that match the query. This will be used
+for the `recall` action.
+
+You can optionally just return `[]` for this method, meaning the agent has no long-term memory.

agenthub/SWE_agent/Examples.md

@@ -0,0 +1,405 @@
+# GPT-4 Examples:
+
+### Notes:
+- The final result is not completed as I only gave it 10 steps, but shows very promising use of the tools.
+- This whole run costs: $0.35 using gpt-4-turbo
+  - Input Tokens: 29,734
+  - Completion Tokens: 1,636
+
+Interesting things:
+1. Using comments as an initial placeholder for the next move and then having that output in memory is a very cool behavior.
+2. It makes very small iterative changes due to the prompting.
+3. A lot of the logic for the game is present but not put together well
+4. There are a lot of odd quirks with the code that it has written.
+5. Some odd agent behavior with putting random lines of code in random places?
+6. The structure of the feedback loop needs some improvement so the agent can see the changes it has made better in context.
+7. Might be beneficial to add in some line numbers so the model knows what lines it is looking at.
+
+
+```
+==============
+STEP 0
+
+PLAN:
+Please make conway's game of life in python using the pygame library. Be sure to use the tools in your workspace to help you out.
+Usage(completion_tokens=91, prompt_tokens=2047, total_tokens=2138)
+
+==== RAW OUTPUT ====
+Thoughts:
+To start implementing Conway's Game of Life in Python using the pygame library, I first need to set up a Python script file where I can write the code. I'll create a new Python file named `game_of_life.py`. This file will serve as the base for writing the game logic and using pygame functionalities.
+
+Action:
+"""
+write game_of_life.py # This is a placeholder for the game of life implementation using pygame.
+"""
+
+==== END RAW ====
+
+
+ACTION:
+FileWriteAction(path='game_of_life.py', content='# This is a placeholder for the game of life implementation using pygame.', start=0, end=-1, thoughts="Thoughts:\nTo start implementing Conway's Game of Life in Python using the pygame library, I first need to set up a Python script file where I can write the code. I'll create a new Python file named `game_of_life.py`. This file will serve as the base for writing the game logic and using pygame functionalities.\n\nAction:", action=<ActionType.WRITE: 'write'>)
+
+OBSERVATION:
+WRITE OPERATION:
+You have written to "game_of_life.py" on these lines: 0:-1.
+
+
+==============
+STEP 1
+
+PLAN:
+Please make conway's game of life in python using the pygame library. Be sure to use the tools in your workspace to help you out.
+Usage(completion_tokens=101, prompt_tokens=2351, total_tokens=2452)
+
+==== RAW OUTPUT ====
+Thoughts:
+The placeholder for the game of life implementation using pygame has been set up in the file `game_of_life.py`. The next step is to start implementing the basic structure of the game. This includes importing necessary libraries, initializing the pygame module, setting up the display window, and defining the main game loop. I will begin by importing pygame and initializing it, which are the foundational steps for any pygame application.
+
+Action:
+"""
+edit 0 0 import pygame
+
+pygame.init()
+"""
+
+==== END RAW ====
+
+
+ACTION:
+FileWriteAction(path='game_of_life.py', content='import pygame\n\npygame.init()', start=0, end=0, thoughts='Thoughts:\nThe placeholder for the game of life implementation using pygame has been set up in the file `game_of_life.py`. The next step is to start implementing the basic structure of the game. This includes importing necessary libraries, initializing the pygame module, setting up the display window, and defining the main game loop. I will begin by importing pygame and initializing it, which are the foundational steps for any pygame application.\n\nAction:', action=<ActionType.WRITE: 'write'>)
+
+OBSERVATION:
+WRITE OPERATION:
+You have written to "game_of_life.py" on these lines: 0:0.
+
+
+==============
+STEP 2
+
+PLAN:
+Please make conway's game of life in python using the pygame library. Be sure to use the tools in your workspace to help you out.
+Usage(completion_tokens=123, prompt_tokens=2552, total_tokens=2675)
+
+==== RAW OUTPUT ====
+Thoughts:
+Previously, I initialized pygame in the `game_of_life.py` file. The next step is to set up the display window for the game. This involves defining the size of the window and setting the title. This is crucial as it provides the visual interface for the game. I will also add a comment to indicate that this is the setup for the display window.
+
+Action:
+"""
+edit 2 2
+# Setup the display window
+screen_size = (800, 600)
+screen = pygame.display.set_mode(screen_size)
+pygame.display.set_caption("Conway's Game of Life")
+"""
+==== END RAW ====
+
+
+ACTION:
+FileWriteAction(path='game_of_life.py', content='# Setup the display window\nscreen_size = (800, 600)\nscreen = pygame.display.set_mode(screen_size)\npygame.display.set_caption("Conway\'s Game of Life")', start=2, end=2, thoughts='Thoughts:\nPreviously, I initialized pygame in the `game_of_life.py` file. The next step is to set up the display window for the game. This involves defining the size of the window and setting the title. This is crucial as it provides the visual interface for the game. I will also add a comment to indicate that this is the setup for the display window.\n\nAction:', action=<ActionType.WRITE: 'write'>)
+
+OBSERVATION:
+WRITE OPERATION:
+You have written to "game_of_life.py" on these lines: 2:2.
+
+
+==============
+STEP 3
+
+PLAN:
+Please make conway's game of life in python using the pygame library. Be sure to use the tools in your workspace to help you out.
+Usage(completion_tokens=144, prompt_tokens=2777, total_tokens=2921)
+
+==== RAW OUTPUT ====
+Thoughts:
+So far, I have set up the basic structure for Conway's Game of Life using pygame, including initializing pygame and setting up the display window. The next step is to define the main game loop where the game's logic will be continuously updated and rendered. This loop will handle events, update the game state, and redraw the screen. I will start by defining the main loop structure and a basic event handling mechanism to allow for quitting the game when the user presses the close button.
+
+Action:
+"""
+edit 5 5 # Main game loop
+running = True
+while running:
+    for event in pygame.event.get():
+        if event.type == pygame.QUIT:
+            running = False
+"""
+
+==== END RAW ====
+
+
+ACTION:
+FileWriteAction(path='game_of_life.py', content='# Main game loop\nrunning = True\nwhile running:\n    for event in pygame.event.get():\n        if event.type == pygame.QUIT:\n            running = False', start=5, end=5, thoughts="Thoughts:\nSo far, I have set up the basic structure for Conway's Game of Life using pygame, including initializing pygame and setting up the display window. The next step is to define the main game loop where the game's logic will be continuously updated and rendered. This loop will handle events, update the game state, and redraw the screen. I will start by defining the main loop structure and a basic event handling mechanism to allow for quitting the game when the user presses the close button.\n\nAction:", action=<ActionType.WRITE: 'write'>)
+
+OBSERVATION:
+WRITE OPERATION:
+You have written to "game_of_life.py" on these lines: 5:5.
+
+
+==============
+STEP 4
+
+PLAN:
+Please make conway's game of life in python using the pygame library. Be sure to use the tools in your workspace to help you out.
+Usage(completion_tokens=222, prompt_tokens=3022, total_tokens=3244)
+
+==== RAW OUTPUT ====
+Thoughts:
+So far, I have set up the basic structure for Conway's Game of Life using pygame, including initializing pygame, setting up the display window, and defining the main game loop. The next step is to implement the logic for Conway's Game of Life. This involves creating a grid to represent the game state, initializing it with a random pattern, and defining the rules for cell survival, birth, and death. I will start by defining the grid size and initializing the grid with random states.
+
+Action:
+"""
+edit 7 7 # Define grid size and initialize grid
+grid_size = (50, 50)  # Grid size as a tuple (rows, columns)
+grid = [[0 for x in range(grid_size[1])] for y in range(grid_size[0])]  # Initialize grid with zeros
+import random
+for y in range(grid_size[0]):
+    for x in range(grid_size[1]):
+        grid[y][x] = random.randint(0, 1)  # Randomly assign cells as alive (1) or dead (0)
+"""
+==== END RAW ====
+
+
+ACTION:
+FileWriteAction(path='game_of_life.py', content='# Define grid size and initialize grid\ngrid_size = (50, 50)  # Grid size as a tuple (rows, columns)\ngrid = [[0 for x in range(grid_size[1])] for y in range(grid_size[0])]  # Initialize grid with zeros\nimport random\nfor y in range(grid_size[0]):\n    for x in range(grid_size[1]):\n        grid[y][x] = random.randint(0, 1)  # Randomly assign cells as alive (1) or dead (0)', start=7, end=7, thoughts="Thoughts:\nSo far, I have set up the basic structure for Conway's Game of Life using pygame, including initializing pygame, setting up the display window, and defining the main game loop. The next step is to implement the logic for Conway's Game of Life. This involves creating a grid to represent the game state, initializing it with a random pattern, and defining the rules for cell survival, birth, and death. I will start by defining the grid size and initializing the grid with random states.\n\nAction:", action=<ActionType.WRITE: 'write'>)
+
+OBSERVATION:
+WRITE OPERATION:
+You have written to "game_of_life.py" on these lines: 7:7.
+
+
+==============
+STEP 5
+
+PLAN:
+Please make conway's game of life in python using the pygame library. Be sure to use the tools in your workspace to help you out.
+Usage(completion_tokens=234, prompt_tokens=3158, total_tokens=3392)
+
+==== RAW OUTPUT ====
+Thoughts:
+So far, I have set up the basic structure for Conway's Game of Life using pygame, including initializing pygame, setting up the display window, defining the main game loop, and initializing the grid with random states. The next step is to implement the rules for Conway's Game of Life. This involves checking each cell's neighbors to determine its next state based on the number of alive neighbors. I will start by defining a function to count the alive neighbors for a given cell.
+
+Action:
+"""
+edit 9 9 def count_alive_neighbors(grid, x, y):
+    directions = [(-1, -1), (-1, 0), (-1, 1), (0, -1), (0, 1), (1, -1), (1, 0), (1, 1)]
+    count = 0
+    for dx, dy in directions:
+        nx, ny = x + dx, y + dy
+        if 0 <= nx < grid_size[1] and 0 <= ny < grid_size[0]:
+            count += grid[ny][nx]
+    return count
+"""
+
+==== END RAW ====
+
+
+ACTION:
+FileWriteAction(path='game_of_life.py', content='def count_alive_neighbors(grid, x, y):\n    directions = [(-1, -1), (-1, 0), (-1, 1), (0, -1), (0, 1), (1, -1), (1, 0), (1, 1)]\n    count = 0\n    for dx, dy in directions:\n        nx, ny = x + dx, y + dy\n        if 0 <= nx < grid_size[1] and 0 <= ny < grid_size[0]:\n            count += grid[ny][nx]\n    return count', start=9, end=9, thoughts="Thoughts:\nSo far, I have set up the basic structure for Conway's Game of Life using pygame, including initializing pygame, setting up the display window, defining the main game loop, and initializing the grid with random states. The next step is to implement the rules for Conway's Game of Life. This involves checking each cell's neighbors to determine its next state based on the number of alive neighbors. I will start by defining a function to count the alive neighbors for a given cell.\n\nAction:", action=<ActionType.WRITE: 'write'>)
+
+OBSERVATION:
+WRITE OPERATION:
+You have written to "game_of_life.py" on these lines: 9:9.
+
+
+==============
+STEP 6
+
+PLAN:
+Please make conway's game of life in python using the pygame library. Be sure to use the tools in your workspace to help you out.
+Usage(completion_tokens=257, prompt_tokens=3290, total_tokens=3547)
+
+==== RAW OUTPUT ====
+Thoughts:
+So far, I have set up the basic structure for Conway's Game of Life using pygame, including initializing pygame, setting up the display window, defining the main game loop, initializing the grid with random states, and creating a function to count alive neighbors. The next step is to implement the rules for Conway's Game of Life. This involves updating the grid state based on the number of alive neighbors each cell has. I will define a function `update_grid` that applies the rules of the game to update the grid for the next generation.
+
+Action:
+"""
+write game_of_life.py def update_grid(grid):
+    new_grid = [[0 for x in range(grid_size[1])] for y in range(grid_size[0])]
+    for y in range(grid_size[0]):
+        for x in range(grid_size[1]):
+            alive_neighbors = count_alive_neighbors(grid, x, y)
+            if grid[y][x] == 1 and (alive_neighbors == 2 or alive_neighbors == 3):
+                new_grid[y][x] = 1
+            elif grid[y][x] == 0 and alive_neighbors == 3:
+                new_grid[y][x] = 1
+    return new_grid
+ 11
+ 11
+"""
+==== END RAW ====
+
+
+ACTION:
+FileWriteAction(path='game_of_life.py', content='def update_grid(grid):\n    new_grid = [[0 for x in range(grid_size[1])] for y in range(grid_size[0])]\n    for y in range(grid_size[0]):\n        for x in range(grid_size[1]):\n            alive_neighbors = count_alive_neighbors(grid, x, y)\n            if grid[y][x] == 1 and (alive_neighbors == 2 or alive_neighbors == 3):\n                new_grid[y][x] = 1\n            elif grid[y][x] == 0 and alive_neighbors == 3:\n                new_grid[y][x] = 1\n    return new_grid', start=11, end=11, thoughts="Thoughts:\nSo far, I have set up the basic structure for Conway's Game of Life using pygame, including initializing pygame, setting up the display window, defining the main game loop, initializing the grid with random states, and creating a function to count alive neighbors. The next step is to implement the rules for Conway's Game of Life. This involves updating the grid state based on the number of alive neighbors each cell has. I will define a function `update_grid` that applies the rules of the game to update the grid for the next generation.\n\nAction:", action=<ActionType.WRITE: 'write'>)
+
+OBSERVATION:
+WRITE OPERATION:
+You have written to "game_of_life.py" on these lines: 11:11.
+
+
+==============
+STEP 7
+
+PLAN:
+Please make conway's game of life in python using the pygame library. Be sure to use the tools in your workspace to help you out.
+Usage(completion_tokens=124, prompt_tokens=3416, total_tokens=3540)
+
+==== RAW OUTPUT ====
+Thoughts:
+So far, I have set up the basic structure for Conway's Game of Life using pygame, including initializing pygame, setting up the display window, defining the main game loop, initializing the grid with random states, and creating functions to count alive neighbors and update the grid. The next step is to integrate the `update_grid` function into the main game loop to continuously update the state of the grid based on the game rules. This will allow the game to evolve over time, showing different generations of the cell population.
+
+Action:
+"""
+edit 6 6
+    grid = update_grid(grid)
+"""
+==== END RAW ====
+
+
+ACTION:
+FileWriteAction(path='game_of_life.py', content='grid = update_grid(grid)', start=6, end=6, thoughts="Thoughts:\nSo far, I have set up the basic structure for Conway's Game of Life using pygame, including initializing pygame, setting up the display window, defining the main game loop, initializing the grid with random states, and creating functions to count alive neighbors and update the grid. The next step is to integrate the `update_grid` function into the main game loop to continuously update the state of the grid based on the game rules. This will allow the game to evolve over time, showing different generations of the cell population.\n\nAction:", action=<ActionType.WRITE: 'write'>)
+
+OBSERVATION:
+WRITE OPERATION:
+You have written to "game_of_life.py" on these lines: 6:6.
+
+
+==============
+STEP 8
+
+PLAN:
+Please make conway's game of life in python using the pygame library. Be sure to use the tools in your workspace to help you out.
+Usage(completion_tokens=166, prompt_tokens=3393, total_tokens=3559)
+
+==== RAW OUTPUT ====
+Thoughts:
+So far, I have set up the basic structure for Conway's Game of Life using pygame, including initializing pygame, setting up the display window, defining the main game loop, initializing the grid with random states, and creating functions to count alive neighbors and update the grid. The next step is to integrate the `update_grid` function into the main game loop to continuously update the state of the grid based on the game rules. This will allow the game to evolve over time, showing different generations of the cell population. I need to ensure that the `update_grid` function is called within the main game loop and that the display is updated accordingly. I will start by checking the current content of the main game loop to see where to integrate the `update_grid` function.
+
+Action:
+"""
+goto 6
+"""
+==== END RAW ====
+
+
+ACTION:
+FileReadAction(path='game_of_life.py', start=6, end=106, thoughts="Thoughts:\nSo far, I have set up the basic structure for Conway's Game of Life using pygame, including initializing pygame, setting up the display window, defining the main game loop, initializing the grid with random states, and creating functions to count alive neighbors and update the grid. The next step is to integrate the `update_grid` function into the main game loop to continuously update the state of the grid based on the game rules. This will allow the game to evolve over time, showing different generations of the cell population. I need to ensure that the `update_grid` function is called within the main game loop and that the display is updated accordingly. I will start by checking the current content of the main game loop to see where to integrate the `update_grid` function.\n\nAction:", action=<ActionType.READ: 'read'>)
+
+OBSERVATION:
+grid = update_grid(grid)
+running = True
+# Define grid size and initialize grid
+grid_size = (50, 50)  # Grid size as a tuple (rows, columns)
+def count_alive_neighbors(grid, x, y):
+    directions = [(-1, -1), (-1, 0), (-1, 1), (0, -1), (0, 1), (1, -1), (1, 0), (1, 1)]
+def update_grid(grid):
+    new_grid = [[0 for x in range(grid_size[1])] for y in range(grid_size[0])]
+    for y in range(grid_size[0]):
+        for x in range(grid_size[1]):
+            alive_neighbors = count_alive_neighbors(grid, x, y)
+            if grid[y][x] == 1 and (alive_neighbors == 2 or alive_neighbors == 3):
+                new_grid[y][x] = 1
+            elif grid[y][x] == 0 and alive_neighbors == 3:
+                new_grid[y][x] = 1
+    return new_grid
+    count = 0
+    for dx, dy in directions:
+        nx, ny = x + dx, y + dy
+        if 0 <= nx < grid_size[1] and 0 <= ny < grid_size[0]:
+            count += grid[ny][nx]
+    return count
+grid = [[0 for x in range(grid_size[1])] for y in range(grid_size[0])]  # Initialize grid with zeros
+import random
+for y in range(grid_size[0]):
+    for x in range(grid_size[1]):
+        grid[y][x] = random.randint(0, 1)  # Randomly assign cells as alive (1) or dead (0)
+while running:
+    for event in pygame.event.get():
+        if event.type == pygame.QUIT:
+            running = False
+pygame.display.set_caption("Conway's Game of Life")
+pygame.init()
+
+
+
+==============
+STEP 9
+
+PLAN:
+Please make conway's game of life in python using the pygame library. Be sure to use the tools in your workspace to help you out.
+Usage(completion_tokens=174, prompt_tokens=3728, total_tokens=3902)
+
+==== RAW OUTPUT ====
+Thoughts:
+From the previous actions, it seems that the basic structure for Conway's Game of Life using pygame has been set up, including initializing pygame, setting up the display window, defining the main game loop, initializing the grid with random states, and creating functions to count alive neighbors and update the grid. The next step is to ensure that the game loop properly updates the grid and refreshes the display to show the new state of the grid in each iteration. This involves calling `pygame.display.update()` to update the entire screen and using `pygame.time.Clock()` to control the frame rate. I will start by adding these necessary updates to the game loop.
+
+Action:
+"""
+edit 95 95
+    pygame.display.update()
+    clock = pygame.time.Clock()
+    clock.tick(10)  # Control the frame rate (10 frames per second)
+"""
+
+==== END RAW ====
+
+
+ACTION:
+FileWriteAction(path='game_of_life.py', content='pygame.display.update()\n    clock = pygame.time.Clock()\n    clock.tick(10)  # Control the frame rate (10 frames per second)', start=95, end=95, thoughts="Thoughts:\nFrom the previous actions, it seems that the basic structure for Conway's Game of Life using pygame has been set up, including initializing pygame, setting up the display window, defining the main game loop, initializing the grid with random states, and creating functions to count alive neighbors and update the grid. The next step is to ensure that the game loop properly updates the grid and refreshes the display to show the new state of the grid in each iteration. This involves calling `pygame.display.update()` to update the entire screen and using `pygame.time.Clock()` to control the frame rate. I will start by adding these necessary updates to the game loop.\n\nAction:", action=<ActionType.WRITE: 'write'>)
+
+OBSERVATION:
+WRITE OPERATION:
+You have written to "game_of_life.py" on these lines: 95:95.
+16:43:55 - opendevin:INFO: agent_controller.py:125 - Exited before finishing the task.
+
+
+```
+
+## Result:
+
+```
+import pygame
+
+# Setup the display window
+screen_size = (800, 600)
+screen = pygame.display.set_mode(screen_size)
+# Main game loop
+grid = update_grid(grid)
+running = True
+# Define grid size and initialize grid
+grid_size = (50, 50)  # Grid size as a tuple (rows, columns)
+def count_alive_neighbors(grid, x, y):
+    directions = [(-1, -1), (-1, 0), (-1, 1), (0, -1), (0, 1), (1, -1), (1, 0), (1, 1)]
+def update_grid(grid):
+    new_grid = [[0 for x in range(grid_size[1])] for y in range(grid_size[0])]
+    for y in range(grid_size[0]):
+        for x in range(grid_size[1]):
+            alive_neighbors = count_alive_neighbors(grid, x, y)
+            if grid[y][x] == 1 and (alive_neighbors == 2 or alive_neighbors == 3):
+                new_grid[y][x] = 1
+            elif grid[y][x] == 0 and alive_neighbors == 3:
+                new_grid[y][x] = 1
+    return new_grid
+    count = 0
+    for dx, dy in directions:
+        nx, ny = x + dx, y + dy
+        if 0 <= nx < grid_size[1] and 0 <= ny < grid_size[0]:
+            count += grid[ny][nx]
+    return count
+grid = [[0 for x in range(grid_size[1])] for y in range(grid_size[0])]  # Initialize grid with zeros
+import random
+for y in range(grid_size[0]):
+    for x in range(grid_size[1]):
+        grid[y][x] = random.randint(0, 1)  # Randomly assign cells as alive (1) or dead (0)
+while running:
+    for event in pygame.event.get():
+        if event.type == pygame.QUIT:
+            running = False
+pygame.display.set_caption("Conway's Game of Life")
+pygame.init()
+# This is a placeholder for the game of life implementation using pygame.pygame.display.update()
+    clock = pygame.time.Clock()
+    clock.tick(10)  # Control the frame rate (10 frames per second)
+
+```

agenthub/SWE_agent/__init__.py

@@ -0,0 +1,5 @@
+from opendevin.controller.agent import Agent
+
+from .agent import SWEAgent
+
+Agent.register('SWEAgent', SWEAgent)

agenthub/SWE_agent/agent.py

@@ -0,0 +1,108 @@
+from opendevin.controller.agent import Agent
+from opendevin.controller.state.state import State
+from opendevin.events.action import (
+    Action,
+    FileReadAction,
+    FileWriteAction,
+    MessageAction,
+)
+from opendevin.events.observation import Observation
+from opendevin.events.serialization.event import event_to_memory
+from opendevin.llm.llm import LLM
+
+from .parser import parse_command
+from .prompts import (
+    CONTEXT_PROMPT,
+    MEMORY_FORMAT,
+    NO_ACTION,
+    STEP_PROMPT,
+    SYSTEM_MESSAGE,
+)
+
+
+class SWEAgent(Agent):
+    """
+    An attempt to recreate swe_agent with output parsing, prompting style, and Application Computer Interface (ACI).
+
+    SWE-agent includes ACI functions like 'goto', 'search_for', 'edit', 'scroll', 'run'
+    """
+
+    def __init__(self, llm: LLM):
+        super().__init__(llm)
+        self.memory_window = 4
+        self.max_retries = 2
+        self.running_memory: list[str] = []
+        self.cur_file: str = ''
+        self.cur_line: int = 0
+
+    def _remember(self, action: Action, observation: Observation) -> None:
+        """Agent has a limited memory of the few steps implemented as a queue"""
+        memory = MEMORY_FORMAT(event_to_memory(action), event_to_memory(observation))
+        self.running_memory.append(memory)
+
+    def _think_act(self, messages: list[dict]) -> tuple[Action, str]:
+        resp = self.llm.completion(
+            messages=messages,
+            temperature=0.05,
+        )
+        action_resp = resp['choices'][0]['message']['content']
+        print(f"\033[1m\033[91m{resp['usage']}\033[0m")
+        print(
+            '\n==== RAW OUTPUT ====',
+            f'\033[96m{action_resp}\033[0m',
+            '==== END RAW ====\n',
+            sep='\n',
+        )
+        return parse_command(action_resp, self.cur_file, self.cur_line)
+
+    def _update(self, action: Action) -> None:
+        if isinstance(action, (FileReadAction, FileWriteAction)):
+            self.cur_file = action.path
+            self.cur_line = action.start
+
+    def step(self, state: State) -> Action:
+        """
+        SWE-Agent step:
+            1. Get context - past actions, custom commands, current step
+            2. Perform think-act - prompt model for action and reasoning
+            3. Catch errors - ensure model takes action (5 attempts max)
+        """
+        for prev_action, obs in state.updated_info:
+            self._remember(prev_action, obs)
+
+        goal = state.get_current_user_intent()
+        prompt = STEP_PROMPT(goal, self.cur_file, self.cur_line)
+
+        msgs = [
+            {'content': SYSTEM_MESSAGE, 'role': 'system'},
+            {'content': prompt, 'role': 'user'},
+        ]
+
+        if len(self.running_memory) > 0:
+            context = CONTEXT_PROMPT(self.running_memory, self.memory_window)
+            msgs.insert(1, {'content': context, 'role': 'user'})
+        # clrs = [''] * (len(msgs)-2) + ['\033[0;36m', '\033[0;35m']
+        # print('\n\n'.join([c+m['content']+'\033[0m' for c, m in zip(clrs, msgs)]))
+        action, thought = self._think_act(messages=msgs)
+
+        start_msg_len = len(msgs)
+        while not action and len(msgs) < self.max_retries + start_msg_len:
+            error = NO_ACTION(thought)
+            error_msg = {'content': error, 'role': 'user'}
+            msgs.append(error_msg)
+            action, thought = self._think_act(messages=msgs)
+
+        if not action:
+            action = MessageAction(thought)
+
+        self._update(action)
+        self.latest_action = action
+        return action
+
+    def search_memory(self, query: str) -> list[str]:
+        return [item for item in self.running_memory if query in item]
+
+    def reset(self) -> None:
+        """Used to reset the agent"""
+        self.running_memory = []
+        super().reset()

agenthub/SWE_agent/parser.py

@@ -0,0 +1,167 @@
+import re
+
+from opendevin.events.action import (
+    Action,
+    AgentFinishAction,
+    BrowseURLAction,
+    CmdRunAction,
+    FileReadAction,
+    FileWriteAction,
+    MessageAction,
+)
+
+from .prompts import COMMAND_USAGE, CUSTOM_DOCS
+
+# commands: exit, read, write, browse, kill, search_file, search_dir
+
+no_open_file_error = MessageAction(
+    'You are not currently in a file. You can use the read command to open a file and then use goto to navigate through it.'
+)
+
+
+def invalid_error(cmd, docs):
+    return f"""ERROR:
+Invalid command structure for
+```
+{cmd}
+```
+You may have caused this error by having multiple commands within your command block.
+If so, try again by running only one of the commands:
+
+Try again using this format:
+{COMMAND_USAGE[docs]}
+"""
+
+
+def get_action_from_string(
+    command_string: str, path: str, line: int, thoughts: str = ''
+) -> Action | None:
+    """
+    Parses the command string to find which command the agent wants to run
+    Converts the command into a proper Action and returns
+    """
+    vars = command_string.split(' ')
+    cmd = vars[0]
+    args = [] if len(vars) == 1 else vars[1:]
+
+    if 'exit' == cmd:
+        return AgentFinishAction()
+
+    elif 'think' == cmd:
+        return MessageAction(' '.join(args))
+
+    elif 'scroll_up' == cmd:
+        if not path:
+            return no_open_file_error
+        return FileReadAction(path, line + 100, line + 200, thoughts)
+
+    elif 'scroll_down' == cmd:
+        if not path:
+            return no_open_file_error
+        return FileReadAction(path, line - 100, line, thoughts)
+
+    elif 'goto' == cmd:
+        if not path:
+            return no_open_file_error
+        rex = r'^goto\s+(\d+)$'
+        valid = re.match(rex, command_string)
+        if valid:
+            start = int(valid.group(1))
+            end = start + 100
+            return FileReadAction(path, start, end, thoughts)
+        else:
+            return MessageAction(invalid_error(command_string, 'goto'))
+
+    elif 'edit' == cmd:
+        if not path:
+            return no_open_file_error
+        rex = r'^edit\s+(\d+)\s+(-?\d+)\s+(\S.*)$'
+        valid = re.match(rex, command_string, re.DOTALL)
+        if valid:
+            start = int(valid.group(1))
+            end = int(valid.group(2))
+            change = valid.group(3)
+            if '"' == change[-1] and '"' == change[0]:
+                change = change[1:-1]
+            return FileWriteAction(path, change, start, end, thoughts)
+        else:
+            return MessageAction(invalid_error(command_string, 'edit'))
+
+    elif 'read' == cmd:
+        rex = r'^read\s+(\S+)(?:\s+(\d+))?(?:\s+(-?\d+))?$'
+        valid = re.match(rex, command_string, re.DOTALL)
+        if valid:
+            file = valid.group(1)
+            start_str = valid.group(2)
+            end_str = valid.group(3)
+
+            start = 0 if not start_str else int(start_str)
+            end = -1 if not end_str else int(end_str)
+
+            return FileReadAction(file, start, end, thoughts)
+        else:
+            return MessageAction(invalid_error(command_string, 'read'))
+
+    elif 'write' == cmd:
+        rex = r'^write\s+(\S+)\s+(.*?)\s*(\d+)?\s*(-?\d+)?$'
+        valid = re.match(rex, command_string, re.DOTALL)
+
+        if valid:
+            file = valid.group(1)
+            content = valid.group(2)
+            start_str = valid.group(3)
+            end_str = valid.group(4)
+
+            start = 0 if not start_str else int(start_str)
+            end = -1 if not end_str else int(end_str)
+
+            if '"' == content[-1] and '"' == content[0]:
+                content = content[1:-1]
+
+            return FileWriteAction(file, content, start, end, thoughts)
+        else:
+            return MessageAction(invalid_error(command_string, 'write'))
+
+    elif 'browse' == cmd:
+        return BrowseURLAction(args[0].strip())
+
+    elif cmd in ['search_file', 'search_dir', 'find_file']:
+        rex = r'^(search_file|search_dir|find_file)\s+(\S+)(?:\s+(\S+))?$'
+        valid = re.match(rex, command_string, re.DOTALL)
+        if valid:
+            return CmdRunAction(command_string)
+        else:
+            return MessageAction(
+                f'Invalid command structure for\n ```\n{command_string}\n```.\nTry again using this format:\n{CUSTOM_DOCS}'
+            )
+    else:
+        # check bash command
+        obs = str(CmdRunAction(f'type {cmd}'))
+        if obs.split(':')[-1].strip() == 'not found':
+            # echo not found error for llm
+            return MessageAction(content=obs)
+        else:
+            # run valid command
+            return CmdRunAction(command_string)
+
+
+def parse_command(input_str: str, path: str, line: int):
+    """
+    Parses a given string and separates the command (enclosed in triple backticks) from any accompanying text.
+
+    Args:
+        input_str (str): The input string to be parsed.
+
+    Returns:
+        tuple: A tuple containing the command and the accompanying text (if any).
+    """
+    input_str = input_str.strip()
+    if '```' in input_str:
+        parts = input_str.split('```')
+        command_str = parts[1].strip()
+        ind = 2 if len(parts) > 2 else 1
+        accompanying_text = ''.join(parts[:-ind]).strip()
+        action = get_action_from_string(command_str, path, line, accompanying_text)
+        if action:
+            return action, accompanying_text
+    return None, input_str  # used for retry

agenthub/SWE_agent/prompts.py

@@ -0,0 +1,207 @@
+DEFAULT_COMMANDS_DICT = {
+    'exit': 'Executed when task is complete',
+    'read <file_name> [<start_line>] [<end_line>]': "Shows a given file's contents starting from <start_line> up to <end_line>. Default: start_line = 0, end_line = -1. By default the whole file will be read.",
+    'write <file> <changes> [<start_line>] [<end_line>]': 'Modifies a <file> by replacing the current lines between <start_line> and <end_line> with <changes>. Default start_line = 0 and end_line = -1. Calling this with no line args will replace the whole file.',
+    'browse <url>': 'Returns the text version of any url, this can be useful to look up documentation or finding issues on github',
+    'scroll_up': 'Takes no arguments. This will scroll up and show you the 100 lines above your current lines',
+    'scroll_down': 'Takes no arguments. This will scroll down and show you the 100 lines below your current lines',
+    'edit <start_line> <end_line> <changes>': 'This will modify lines in the currently open file. use start_line and end_line to designate which lines to change and then write the multiline changes. Set end_line to -1 to denote the end of the file',
+    'goto <line_num>': 'This will take you directly to a line and show you the 100 lines below it.',
+    '<bash_command> <args>': 'You can use any bash command you need (cd, ls, rm, grep, dir, mv, wget, git, zip, etc.) with their arguments included',
+    'pip install <package>': 'You can use this to import python packages. Make sure you include the correct package name when using this command.',
+    'ls': 'Use the ls command to view all the files in your current directory, this is a good starting point.',
+    'NOT ALLOWED': 'You cannot use interactive commands like python or node',
+}
+
+COMMAND_USAGE = {
+    'exit': 'Usage:\n```\nexit\n```\nExecuted when task is complete',
+    'read': "Args:\n<file_name> [<start_line>] [<end_line>]\nUsage:\n```\nread file.py\n```\nor\n```\nread example.py <start_line> <end_line>\n```\nShows a given file's contents starting from <start_line> up to <end_line>. Default: start_line = 0, end_line = -1. by default the whole file will be read.",
+    'write': 'Args:\n<file> <changes> [<start_line>] [<end_line>]\nUsage:\n```\nwrite "def main():\n    print("This is line one")" 0 2\n```\nModifies a <file> by replacing the current lines between <start_line> and <end_line> with <changes>. Default start_line = 0 and end_line = -1. Calling this with no line args will replace the whole file.',
+    'edit': 'Args:\n<start_line> <end_line> <changes>\nUsage:\n```\nedit 0 1 import pandas as pd\n```\nThis will modify the current file you are in with the changes you make between the line numbers you designate',
+    'goto': 'Args:\n<line_num>\nUsage:\n```\ngoto <line_num>\n```\nThis will show you the 100 lines below and including the line you specify within your current file.',
+    'scroll_up': 'Usage:\n```\nscroll_up\n```\nThis will return the 100 lines above where you are currently at',
+    'scroll_down': 'Usage:\n```\nscroll_down\n```\nThis will return the 100 line below where you are currently at',
+    'browse': 'Args:\n<url>\nUsage:\n```\nbrowse https://github.com/OpenDevin/OpenDevin\n```\nThis will fetch the Text elements from the given url and show them to you.',
+}
+
+DEFAULT_COMMANDS = '\n'.join([k + ' - ' + v for k, v in DEFAULT_COMMANDS_DICT.items()])
+
+# from opendevin.parse_commands import parse_command_file
+# USE parse_command_file(filepath) to get the custom commands
+CUSTOM_DOCS = None
+
+CUSTOM_COMMANDS = f"""Custom bash commands:
+{CUSTOM_DOCS}
+"""
+
+DOCUMENTATION = f"""DOCUMENTATION:
+It is recommend that you use the commands provided for interacting with files and your directory because they have been specially built for you.
+They will make it much easier for you to look at files and make changes. Using these commands will help you be better at your task.
+You can open an file by using either the read or write operations.
+- If a file already exists you should read it before making any changes. Use the `edit` command to make changes once you have read it.
+- If you are creating a new file use the write command. Use the `edit` command to make changes once you have created the new file.
+
+Commands:
+{DEFAULT_COMMANDS}
+{CUSTOM_COMMANDS}
+
+The following commands require an open file to be used: edit, scroll_up, scroll_down, goto
+To modify the current file use 'edit'. To move through the current file use 'goto' or 'scroll_up'/'scroll_down'
+when using write and edit do not surround the code with any "" just write the code.
+"""
+
+GENERAL_GUIDELINES = """INSTRUCTIONS:
+Now, you're going to solve this issue on your own. You can use any bash commands or custom commands you wish to complete your task. Edit all the files you need to and run any checks or tests that you want.
+Remember, YOU CAN ONLY ENTER ONE COMMAND AT A TIME. You should always wait for feedback after every command.
+When you're satisfied with all of the changes you've made, you can indicate that you are done by running the exit command.
+Note however that you cannot use any interactive session commands (e.g. python, vim, node) in this environment, but you can write scripts and run them. E.g. you can write a python script and then run it with `python <script_name>.py`.
+
+NOTE ABOUT THE write COMMAND: Indentation really matters! When editing a file, make sure to insert appropriate indentation before each line!
+
+IMPORTANT TIPS:
+1. Reproduce the bug: Always start by trying to replicate the bug that the issue discusses. If the issue includes code for reproducing the bug, we recommend that you re-implement that in your environment and run it to ensure you can reproduce the bug. Then, start trying to fix it. When you think you've fixed the bug, re-run the bug reproduction script to make sure that the issue has indeed been resolved.
+   If the bug reproduction script does not print anything when it successfully runs, we recommend adding a print("Script completed successfully, no errors.") command at the end of the file, so that you can be sure the script ran fine all the way through.
+2. Try different commands: If you run a command and it doesn't work, try running a different command. A command that did not work once will not work the second time unless you modify it.
+3. Navigate large files: If you open a file and need to get to an area around a specific line that is not in the first 100 lines, say line 583, you would use the 'read' command like this: 'read <file> 583'. This is a much faster way to read through the file.
+4. Handle input files: If the bug reproduction script requires inputting/reading a specific file, such as 'buggy-input.png', and you'd like to understand how to input that file, conduct a search in the existing repository code to see whether someone else has already done that. Do this by running the command: 'search_dir "buggy-input.png"'. If that doesn't work, use the Linux 'find' command.
+5. Understand your context: Always make sure to look at the currently open file and the current working directory. The currently open file might be in a different directory than the working directory.
+6. Verify your edits: When editing files, it is easy to accidentally specify a wrong line number or to write code with incorrect indentation. Always check the code after you issue an edit to make sure that it reflects what you wanted to accomplish. If it didn't, issue another command to fix it.
+7. Thoroughly test your solution: After making any changes to fix a bug, be sure to thoroughly test your solution to ensure the bug has been resolved. Re-run the bug reproduction script and verify that the issue has been addressed.
+"""
+
+RESPONSE_FORMAT = """RESPONSE FORMAT:
+This is the format of the response you will make in order to solve the current issue.
+You will be given multiple iterations to complete this task so break it into steps and solve them one by one.
+
+Your output must contain the following:
+- First, thoughts about what your next action should be and plan it out.
+    - You will have a memory of your thoughts so you can use this to remember things for the next step.
+    - Use your thoughts to think about what you are currently doing, what you have done on prior steps and how that relates to solving the problem.
+- Second, create a piece of code that will execute your next action based on the thoughts you have.
+    - Remember that you can only have one action for each thought, do not include multiple actions.
+
+Your code MUST be surrounded in triple back ticks EXACTLY like this:
+```
+<code>
+```
+
+Notes:
+- Adhere to the format so that the program loop continues smoothly, it is very important to only give one command per output.
+- DO NOT give more than one command within the triple backticks. This will just throw an error and nothing will happen as a result.
+- Do not give multiple code blocks, if you do only the second one will be captured and run, this might give an error if the first one was necessary.
+- To execute multiple commands you should write them down in your thoughts section so you can remember it on the next step and execute them then.
+- The only commands you are not capable of executing are interactive commands like `python` or `node` by themselves.
+- If you think that you have completed the task that has been given to you based on your previous actions and outputs then use ``` exit ``` as the command to let the system know that you are done.
+- DO NOT make any copies of your previous memories those will be provided to you at each step, making copies just wastes time and energy. Think smarter not harder.
+- The write and edit commands requires proper indentation in the content section ex. `write hw.py def hello():\n    print(\'Hello World\')` this is how you would have to format your write command.
+    - The white spaces matter as the code changes will be added to the code so they must have proper syntax.
+
+This is a template using the format described above
+Items in <> are suggestions for you, fill them out based on the context of the problem you are solving.
+
+[ FORMAT ]
+Thoughts:
+<Provide clear and concise thoughts on the next step to take, highlighting any important details or context that should be remembered.>
+<You can use multiple lines to express your thoughts>
+
+Action:
+```
+<command> <params>
+```
+[ END FORMAT ]
+
+Do not provide anything extra just your thought and action.
+"""
+
+SYSTEM_MESSAGE = f"""SYSTEM INFO:
+You am an autonomous coding agent, here to provide solutions for coding issues.
+You have been designed to assist you with a wide range of programming tasks, from code editing and debugging to testing and deployment.
+You have access to a variety of tools and commands that you can use to help you solve problems efficiently.
+
+{GENERAL_GUIDELINES}
+
+{DOCUMENTATION}
+""".strip()
+
+
+def NO_ACTION(latest):
+    return f"""
+You did not include any action to take in your most recent output:
+
+===== Output ======
+{latest}
+==== End Output ===
+
+Remember these are the custom commands you can use:
+{DOCUMENTATION}
+
+Lets try that again, it is very important that you adhere to the output format
+This time, be sure to use the exact format below, replacing anything in <> with the appropriate value(s):
+{RESPONSE_FORMAT}
+
+It is crucial you use the format provided as the output will be parsed automatically.
+"""
+
+
+def file_info(file: str, line: int):
+    if file:
+        return f"""CURRENT WORKSPACE:
+    Open File: {file} on line {line}
+    You can use these commands with the current file:
+    Navigation: `scroll_up`, `scroll_down`, and `goto <line>`
+    Modification: `edit <start_line> <end_line> <changes>`
+    """
+
+
+def STEP_PROMPT(task, file, line_num):
+    return f"""
+{RESPONSE_FORMAT}
+You are currently trying to complete this task:
+{task}
+
+{file_info(file, line_num)}
+
+Keep all of the guidelines above in mind when you are thinking and making code.
+Please come up with a thought and action based on your current task and latest steps.
+Make sure that you do not repeat the same actions, there will not be any changes in result if you do not changes anything.
+Be very strict about the formatting that you use and make sure you follow the guidelines.
+NEVER output multiple commands. ONLY take ONE STEP at a time.
+When you have completed your task run the "exit" command.
+Begin with your thought about the next step and then come up with an action to perform your thought.
+""".strip()
+
+
+def unpack_dict(data: dict, restrict: list[str] = []):
+    lines = []
+    for key, value in data.items():
+        if key in restrict:
+            continue
+        elif isinstance(value, dict):
+            nested_str = unpack_dict(value, restrict).replace('\n', '\n  ')
+            val = f'{key}:' + '\n  ' + f'{nested_str}'
+            lines.append(val)
+        else:
+            lines.append(f'{key}: {value}')
+    return '\n'.join(lines)
+
+
+def MEMORY_FORMAT(act, obs):
+    return f"""
+Previous Action:
+{unpack_dict(act, ["content"])}
+
+Output from Action:
+{unpack_dict(obs)}
+""".strip()
+
+
+def CONTEXT_PROMPT(memory, window):
+    res = f'These are your past {window} actions:\n'
+    window_size = window if len(memory) > window else len(memory)
+    cur_mems = memory[-window_size:]
+    res += '===== Previous Actions =====\n'
+    for idx, mem in enumerate(cur_mems):
+        res += f'\nMemory {idx}:\n{mem}\n'
+    res += '======= End Actions =======\n'
+    res += 'Use these memories to provide additional context to the problem you are solving.\nRemember that you have already completed these steps so you do not need to perform them again.'
+    return res

agenthub/__init__.py

@@ -9,21 +9,21 @@ load_dotenv()
 
 
 from . import (  # noqa: E402
-    browsing_agent,
+    SWE_agent,
     codeact_agent,
-    codeact_swe_agent,
     delegator_agent,
     dummy_agent,
+    monologue_agent,
     planner_agent,
 )
 
 __all__ = [
+    'monologue_agent',
     'codeact_agent',
-    'codeact_swe_agent',
     'planner_agent',
+    'SWE_agent',
     'delegator_agent',
     'dummy_agent',
-    'browsing_agent',
 ]
 
 for agent in all_microagents.values():

agenthub/browsing_agent/README.md

@@ -1,16 +0,0 @@
-# Browsing Agent Framework
-
-This folder implements the basic BrowserGym [demo agent](https://github.com/ServiceNow/BrowserGym/tree/main/demo_agent) that enables full-featured web browsing.
-
-
-## Test run
-
-Note that for browsing tasks, GPT-4 is usually a requirement to get reasonable results, due to the complexity of the web page structures.
-
-```
-poetry run python ./opendevin/core/main.py \
-           -i 10 \
-           -t "tell me the usa's president using google search" \
-           -c BrowsingAgent \
-           -m gpt-4o-2024-05-13
-```

agenthub/browsing_agent/__init__.py

@@ -1,5 +0,0 @@
-from opendevin.controller.agent import Agent
-
-from .browsing_agent import BrowsingAgent
-
-Agent.register('BrowsingAgent', BrowsingAgent)

agenthub/browsing_agent/browsing_agent.py

@@ -1,211 +0,0 @@
-import os
-
-from browsergym.core.action.highlevel import HighLevelActionSet
-from browsergym.utils.obs import flatten_axtree_to_str
-
-from agenthub.browsing_agent.response_parser import BrowsingResponseParser
-from opendevin.controller.agent import Agent
-from opendevin.controller.state.state import State
-from opendevin.core.logger import opendevin_logger as logger
-from opendevin.events.action import (
-    Action,
-    AgentFinishAction,
-    BrowseInteractiveAction,
-    MessageAction,
-)
-from opendevin.events.event import EventSource
-from opendevin.events.observation import BrowserOutputObservation
-from opendevin.events.observation.observation import Observation
-from opendevin.llm.llm import LLM
-from opendevin.runtime.plugins import (
-    PluginRequirement,
-)
-from opendevin.runtime.tools import RuntimeTool
-
-USE_NAV = (
-    os.environ.get('USE_NAV', 'true') == 'true'
-)  # only disable NAV actions when running webarena and miniwob benchmarks
-USE_CONCISE_ANSWER = (
-    os.environ.get('USE_CONCISE_ANSWER', 'false') == 'true'
-)  # only return concise answer when running webarena and miniwob benchmarks
-
-if not USE_NAV and USE_CONCISE_ANSWER:
-    EVAL_MODE = True  # disabled NAV actions and only return concise answer, for webarena and miniwob benchmarks\
-else:
-    EVAL_MODE = False
-
-
-def get_error_prefix(last_browser_action: str) -> str:
-    return f'IMPORTANT! Last action is incorrect:\n{last_browser_action}\nThink again with the current observation of the page.\n'
-
-
-def get_system_message(goal: str, action_space: str) -> str:
-    return f"""\
-# Instructions
-Review the current state of the page and all other information to find the best
-possible next action to accomplish your goal. Your answer will be interpreted
-and executed by a program, make sure to follow the formatting instructions.
-
-# Goal:
-{goal}
-
-# Action Space
-{action_space}
-"""
-
-
-CONCISE_INSTRUCTION = """\
-
-Here is another example with chain of thought of a valid action when providing a concise answer to user:
-"
-In order to accomplish my goal I need to send the information asked back to the user. This page list the information of HP Inkjet Fax Machine, which is the product identified in the objective. Its price is $279.49. I will send a message back to user with the answer.
-```send_msg_to_user("$279.49")```
-"
-"""
-
-
-def get_prompt(error_prefix: str, cur_axtree_txt: str, prev_action_str: str) -> str:
-    prompt = f"""\
-{error_prefix}
-
-# Current Accessibility Tree:
-{cur_axtree_txt}
-
-# Previous Actions
-{prev_action_str}
-
-Here is an example with chain of thought of a valid action when clicking on a button:
-"
-In order to accomplish my goal I need to click on the button with bid 12
-```click("12")```
-"
-""".strip()
-    if USE_CONCISE_ANSWER:
-        prompt += CONCISE_INSTRUCTION
-    return prompt
-
-
-class BrowsingAgent(Agent):
-    VERSION = '1.0'
-    """
-    An agent that interacts with the browser.
-    """
-
-    sandbox_plugins: list[PluginRequirement] = []
-    runtime_tools: list[RuntimeTool] = [RuntimeTool.BROWSER]
-    response_parser = BrowsingResponseParser()
-
-    def __init__(
-        self,
-        llm: LLM,
-    ) -> None:
-        """Initializes a new instance of the BrowsingAgent class.
-
-        Parameters:
-        - llm (LLM): The llm to be used by this agent
-        """
-        super().__init__(llm)
-        # define a configurable action space, with chat functionality, web navigation, and webpage grounding using accessibility tree and HTML.
-        # see https://github.com/ServiceNow/BrowserGym/blob/main/core/src/browsergym/core/action/highlevel.py for more details
-        action_subsets = ['chat', 'bid']
-        if USE_NAV:
-            action_subsets.append('nav')
-        self.action_space = HighLevelActionSet(
-            subsets=action_subsets,
-            strict=False,  # less strict on the parsing of the actions
-            multiaction=True,  # enable to agent to take multiple actions at once
-        )
-
-        self.reset()
-
-    def reset(self) -> None:
-        """Resets the Browsing Agent."""
-        super().reset()
-        self.cost_accumulator = 0
-        self.error_accumulator = 0
-
-    def step(self, state: State) -> Action:
-        """Performs one step using the Browsing Agent.
-        This includes gathering information on previous steps and prompting the model to make a browsing command to execute.
-
-        Parameters:
-        - state (State): used to get updated info
-
-        Returns:
-        - BrowseInteractiveAction(browsergym_command) - BrowserGym commands to run
-        - MessageAction(content) - Message action to run (e.g. ask for clarification)
-        - AgentFinishAction() - end the interaction
-        """
-        messages = []
-        prev_actions = []
-        cur_axtree_txt = ''
-        error_prefix = ''
-        last_obs = None
-        last_action = None
-
-        if EVAL_MODE and len(state.history.get_events_as_list()) == 1:
-            # for webarena and miniwob++ eval, we need to retrieve the initial observation already in browser env
-            # initialize and retrieve the first observation by issuing an noop OP
-            # For non-benchmark browsing, the browser env starts with a blank page, and the agent is expected to first navigate to desired websites
-            return BrowseInteractiveAction(browser_actions='noop()')
-
-        for event in state.history.get_events():
-            if isinstance(event, BrowseInteractiveAction):
-                prev_actions.append(event.browser_actions)
-                last_action = event
-            elif isinstance(event, MessageAction) and event.source == EventSource.AGENT:
-                # agent has responded, task finished.
-                return AgentFinishAction(outputs={'content': event.content})
-            elif isinstance(event, Observation):
-                last_obs = event
-
-        if EVAL_MODE:
-            prev_actions = prev_actions[1:]  # remove the first noop action
-
-        prev_action_str = '\n'.join(prev_actions)
-        # if the final BrowserInteractiveAction exec BrowserGym's send_msg_to_user,
-        # we should also send a message back to the user in OpenDevin and call it a day
-        if (
-            isinstance(last_action, BrowseInteractiveAction)
-            and last_action.browsergym_send_msg_to_user
-        ):
-            return MessageAction(last_action.browsergym_send_msg_to_user)
-
-        if isinstance(last_obs, BrowserOutputObservation):
-            if last_obs.error:
-                # add error recovery prompt prefix
-                error_prefix = get_error_prefix(last_obs.last_browser_action)
-                self.error_accumulator += 1
-                if self.error_accumulator > 5:
-                    return MessageAction('Too many errors encountered. Task failed.')
-            try:
-                cur_axtree_txt = flatten_axtree_to_str(
-                    last_obs.axtree_object,
-                    extra_properties=last_obs.extra_element_properties,
-                    with_clickable=True,
-                    filter_visible_only=True,
-                )
-            except Exception as e:
-                logger.error(
-                    'Error when trying to process the accessibility tree: %s', e
-                )
-                return MessageAction('Error encountered when browsing.')
-
-        if (goal := state.get_current_user_intent()) is None:
-            goal = state.inputs['task']
-        system_msg = get_system_message(
-            goal,
-            self.action_space.describe(with_long_description=False, with_examples=True),
-        )
-
-        messages.append({'role': 'system', 'content': system_msg})
-
-        prompt = get_prompt(error_prefix, cur_axtree_txt, prev_action_str)
-        messages.append({'role': 'user', 'content': prompt})
-        logger.debug(prompt)
-        response = self.llm.completion(
-            messages=messages,
-            temperature=0.0,
-            stop=[')```', ')\n```'],
-        )
-        return self.response_parser.parse(response)

agenthub/browsing_agent/prompt.py

@@ -1,787 +0,0 @@
-import abc
-import difflib
-import logging
-import platform
-from copy import deepcopy
-from dataclasses import asdict, dataclass
-from textwrap import dedent
-from typing import Literal, Union
-from warnings import warn
-
-from browsergym.core.action.base import AbstractActionSet
-from browsergym.core.action.highlevel import HighLevelActionSet
-from browsergym.core.action.python import PythonActionSet
-
-from opendevin.runtime.browser.browser_env import BrowserEnv
-
-from .utils import (
-    ParseError,
-    parse_html_tags_raise,
-)
-
-
-@dataclass
-class Flags:
-    use_html: bool = True
-    use_ax_tree: bool = False
-    drop_ax_tree_first: bool = True  # This flag is no longer active TODO delete
-    use_thinking: bool = False
-    use_error_logs: bool = False
-    use_past_error_logs: bool = False
-    use_history: bool = False
-    use_action_history: bool = False
-    use_memory: bool = False
-    use_diff: bool = False
-    html_type: str = 'pruned_html'
-    use_concrete_example: bool = True
-    use_abstract_example: bool = False
-    multi_actions: bool = False
-    action_space: Literal[
-        'python', 'bid', 'coord', 'bid+coord', 'bid+nav', 'coord+nav', 'bid+coord+nav'
-    ] = 'bid'
-    is_strict: bool = False
-    # This flag will be automatically disabled `if not chat_model_args.has_vision()`
-    use_screenshot: bool = True
-    enable_chat: bool = False
-    max_prompt_tokens: int = 100_000
-    extract_visible_tag: bool = False
-    extract_coords: Literal['False', 'center', 'box'] = 'False'
-    extract_visible_elements_only: bool = False
-    demo_mode: Literal['off', 'default', 'only_visible_elements'] = 'off'
-
-    def copy(self):
-        return deepcopy(self)
-
-    def asdict(self):
-        """Helper for JSON serializble requirement."""
-        return asdict(self)
-
-    @classmethod
-    def from_dict(self, flags_dict):
-        """Helper for JSON serializble requirement."""
-        if isinstance(flags_dict, Flags):
-            return flags_dict
-
-        if not isinstance(flags_dict, dict):
-            raise ValueError(
-                f'Unregcognized type for flags_dict of type {type(flags_dict)}.'
-            )
-        return Flags(**flags_dict)
-
-
-class PromptElement:
-    """Base class for all prompt elements. Prompt elements can be hidden.
-
-    Prompt elements are used to build the prompt. Use flags to control which
-    prompt elements are visible. We use class attributes as a convenient way
-    to implement static prompts, but feel free to override them with instance
-    attributes or @property decorator.
-    """
-
-    _prompt = ''
-    _abstract_ex = ''
-    _concrete_ex = ''
-
-    def __init__(self, visible: bool = True) -> None:
-        """Prompt element that can be hidden.
-
-        Parameters
-        ----------
-        visible : bool, optional
-            Whether the prompt element should be visible, by default True. Can
-            be a callable that returns a bool. This is useful when a specific
-            flag changes during a shrink iteration.
-        """
-        self._visible = visible
-
-    @property
-    def prompt(self):
-        """Avoid overriding this method. Override _prompt instead."""
-        return self._hide(self._prompt)
-
-    @property
-    def abstract_ex(self):
-        """Useful when this prompt element is requesting an answer from the llm.
-        Provide an abstract example of the answer here. See Memory for an
-        example.
-
-        Avoid overriding this method. Override _abstract_ex instead
-        """
-        return self._hide(self._abstract_ex)
-
-    @property
-    def concrete_ex(self):
-        """Useful when this prompt element is requesting an answer from the llm.
-        Provide a concrete example of the answer here. See Memory for an
-        example.
-
-        Avoid overriding this method. Override _concrete_ex instead
-        """
-        return self._hide(self._concrete_ex)
-
-    @property
-    def is_visible(self):
-        """Handle the case where visible is a callable."""
-        visible = self._visible
-        if callable(visible):
-            visible = visible()
-        return visible
-
-    def _hide(self, value):
-        """Return value if visible is True, else return empty string."""
-        if self.is_visible:
-            return value
-        else:
-            return ''
-
-    def _parse_answer(self, text_answer) -> dict:
-        if self.is_visible:
-            return self._parse_answer(text_answer)
-        else:
-            return {}
-
-
-class Shrinkable(PromptElement, abc.ABC):
-    @abc.abstractmethod
-    def shrink(self) -> None:
-        """Implement shrinking of this prompt element.
-
-        You need to recursively call all shrinkable elements that are part of
-        this prompt. You can also implement a shrinking strategy for this prompt.
-        Shrinking is can be called multiple times to progressively shrink the
-        prompt until it fits max_tokens. Default max shrink iterations is 20.
-        """
-        pass
-
-
-class Truncater(Shrinkable):
-    """A prompt element that can be truncated to fit the context length of the LLM.
-    Of course, it will be great that we never have to use the functionality here to `shrink()` the prompt.
-    Extend this class for prompt elements that can be truncated. Usually long observations such as AxTree or HTML.
-    """
-
-    def __init__(self, visible, shrink_speed=0.3, start_truncate_iteration=10):
-        super().__init__(visible=visible)
-        self.shrink_speed = shrink_speed  # the percentage shrunk in each iteration
-        self.start_truncate_iteration = (
-            start_truncate_iteration  # the iteration to start truncating
-        )
-        self.shrink_calls = 0
-        self.deleted_lines = 0
-
-    def shrink(self) -> None:
-        if self.is_visible and self.shrink_calls >= self.start_truncate_iteration:
-            # remove the fraction of _prompt
-            lines = self._prompt.splitlines()
-            new_line_count = int(len(lines) * (1 - self.shrink_speed))
-            self.deleted_lines += len(lines) - new_line_count
-            self._prompt = '\n'.join(lines[:new_line_count])
-            self._prompt += (
-                f'\n... Deleted {self.deleted_lines} lines to reduce prompt size.'
-            )
-
-        self.shrink_calls += 1
-
-
-def fit_tokens(
-    shrinkable: Shrinkable,
-    max_prompt_chars=None,
-    max_iterations=20,
-):
-    """Shrink a prompt element until it fits max_tokens.
-
-    Parameters
-    ----------
-    shrinkable : Shrinkable
-        The prompt element to shrink.
-    max_prompt_chars : int
-        The maximum number of chars allowed.
-    max_iterations : int, optional
-        The maximum number of shrink iterations, by default 20.
-    model_name : str, optional
-        The name of the model used when tokenizing.
-
-    Returns:
-    -------
-    str : the prompt after shrinking.
-    """
-    if max_prompt_chars is None:
-        return shrinkable.prompt
-
-    for _ in range(max_iterations):
-        prompt = shrinkable.prompt
-        if isinstance(prompt, str):
-            prompt_str = prompt
-        elif isinstance(prompt, list):
-            prompt_str = '\n'.join([p['text'] for p in prompt if p['type'] == 'text'])
-        else:
-            raise ValueError(f'Unrecognized type for prompt: {type(prompt)}')
-        n_chars = len(prompt_str)
-        if n_chars <= max_prompt_chars:
-            return prompt
-        shrinkable.shrink()
-
-    logging.info(
-        dedent(
-            f"""\
-            After {max_iterations} shrink iterations, the prompt is still
-            {len(prompt_str)} chars (greater than {max_prompt_chars}). Returning the prompt as is."""
-        )
-    )
-    return prompt
-
-
-class HTML(Truncater):
-    def __init__(self, html, visible: bool = True, prefix='') -> None:
-        super().__init__(visible=visible, start_truncate_iteration=5)
-        self._prompt = f'\n{prefix}HTML:\n{html}\n'
-
-
-class AXTree(Truncater):
-    def __init__(
-        self, ax_tree, visible: bool = True, coord_type=None, prefix=''
-    ) -> None:
-        super().__init__(visible=visible, start_truncate_iteration=10)
-        if coord_type == 'center':
-            coord_note = """\
-Note: center coordinates are provided in parenthesis and are
-  relative to the top left corner of the page.\n\n"""
-        elif coord_type == 'box':
-            coord_note = """\
-Note: bounding box of each object are provided in parenthesis and are
-  relative to the top left corner of the page.\n\n"""
-        else:
-            coord_note = ''
-        self._prompt = f'\n{prefix}AXTree:\n{coord_note}{ax_tree}\n'
-
-
-class Error(PromptElement):
-    def __init__(self, error, visible: bool = True, prefix='') -> None:
-        super().__init__(visible=visible)
-        self._prompt = f'\n{prefix}Error from previous action:\n{error}\n'
-
-
-class Observation(Shrinkable):
-    """Observation of the current step.
-
-    Contains the html, the accessibility tree and the error logs.
-    """
-
-    def __init__(self, obs, flags: Flags) -> None:
-        super().__init__()
-        self.flags = flags
-        self.obs = obs
-        self.html = HTML(obs[flags.html_type], visible=flags.use_html, prefix='## ')
-        self.ax_tree = AXTree(
-            obs['axtree_txt'],
-            visible=flags.use_ax_tree,
-            coord_type=flags.extract_coords,
-            prefix='## ',
-        )
-        self.error = Error(
-            obs['last_action_error'],
-            visible=flags.use_error_logs and obs['last_action_error'],
-            prefix='## ',
-        )
-
-    def shrink(self):
-        self.ax_tree.shrink()
-        self.html.shrink()
-
-    @property
-    def _prompt(self) -> str:  # type: ignore
-        return f'\n# Observation of current step:\n{self.html.prompt}{self.ax_tree.prompt}{self.error.prompt}\n\n'
-
-    def add_screenshot(self, prompt):
-        if self.flags.use_screenshot:
-            if isinstance(prompt, str):
-                prompt = [{'type': 'text', 'text': prompt}]
-            img_url = BrowserEnv.image_to_jpg_base64_url(
-                self.obs['screenshot'], add_data_prefix=True
-            )
-            prompt.append({'type': 'image_url', 'image_url': img_url})
-
-        return prompt
-
-
-class MacNote(PromptElement):
-    def __init__(self) -> None:
-        super().__init__(visible=platform.system() == 'Darwin')
-        self._prompt = '\nNote: you are on mac so you should use Meta instead of Control for Control+C etc.\n'
-
-
-class BeCautious(PromptElement):
-    def __init__(self, visible: bool = True) -> None:
-        super().__init__(visible=visible)
-        self._prompt = """\
-\nBe very cautious. Avoid submitting anything before verifying the effect of your
-actions. Take the time to explore the effect of safe actions first. For example
-you can fill a few elements of a form, but don't click submit before verifying
-that everything was filled correctly.\n"""
-
-
-class GoalInstructions(PromptElement):
-    def __init__(self, goal, visible: bool = True) -> None:
-        super().__init__(visible)
-        self._prompt = f"""\
-# Instructions
-Review the current state of the page and all other information to find the best
-possible next action to accomplish your goal. Your answer will be interpreted
-and executed by a program, make sure to follow the formatting instructions.
-
-## Goal:
-{goal}
-"""
-
-
-class ChatInstructions(PromptElement):
-    def __init__(self, chat_messages, visible: bool = True) -> None:
-        super().__init__(visible)
-        self._prompt = """\
-# Instructions
-
-You are a UI Assistant, your goal is to help the user perform tasks using a web browser. You can
-communicate with the user via a chat, in which the user gives you instructions and in which you
-can send back messages. You have access to a web browser that both you and the user can see,
-and with which only you can interact via specific commands.
-
-Review the instructions from the user, the current state of the page and all other information
-to find the best possible next action to accomplish your goal. Your answer will be interpreted
-and executed by a program, make sure to follow the formatting instructions.
-
-## Chat messages:
-
-"""
-        self._prompt += '\n'.join(
-            [
-                f"""\
- - [{msg['role']}] {msg['message']}"""
-                for msg in chat_messages
-            ]
-        )
-
-
-class SystemPrompt(PromptElement):
-    _prompt = """\
-You are an agent trying to solve a web task based on the content of the page and
-a user instructions. You can interact with the page and explore. Each time you
-submit an action it will be sent to the browser and you will receive a new page."""
-
-
-class MainPrompt(Shrinkable):
-    def __init__(
-        self,
-        obs_history,
-        actions,
-        memories,
-        thoughts,
-        flags: Flags,
-    ) -> None:
-        super().__init__()
-        self.flags = flags
-        self.history = History(obs_history, actions, memories, thoughts, flags)
-        if self.flags.enable_chat:
-            self.instructions: Union[ChatInstructions, GoalInstructions] = (
-                ChatInstructions(obs_history[-1]['chat_messages'])
-            )
-        else:
-            if (
-                'chat_messages' in obs_history[-1]
-                and sum(
-                    [msg['role'] == 'user' for msg in obs_history[-1]['chat_messages']]
-                )
-                > 1
-            ):
-                logging.warning(
-                    'Agent is in goal mode, but multiple user messages are present in the chat. Consider switching to `enable_chat=True`.'
-                )
-            self.instructions = GoalInstructions(obs_history[-1]['goal'])
-
-        self.obs = Observation(obs_history[-1], self.flags)
-        self.action_space = ActionSpace(self.flags)
-
-        self.think = Think(visible=flags.use_thinking)
-        self.memory = Memory(visible=flags.use_memory)
-
-    @property
-    def _prompt(self) -> str:  # type: ignore
-        prompt = f"""\
-{self.instructions.prompt}\
-{self.obs.prompt}\
-{self.history.prompt}\
-{self.action_space.prompt}\
-{self.think.prompt}\
-{self.memory.prompt}\
-"""
-
-        if self.flags.use_abstract_example:
-            prompt += f"""
-# Abstract Example
-
-Here is an abstract version of the answer with description of the content of
-each tag. Make sure you follow this structure, but replace the content with your
-answer:
-{self.think.abstract_ex}\
-{self.memory.abstract_ex}\
-{self.action_space.abstract_ex}\
-"""
-
-        if self.flags.use_concrete_example:
-            prompt += f"""
-# Concrete Example
-
-Here is a concrete example of how to format your answer.
-Make sure to follow the template with proper tags:
-{self.think.concrete_ex}\
-{self.memory.concrete_ex}\
-{self.action_space.concrete_ex}\
-"""
-        return self.obs.add_screenshot(prompt)
-
-    def shrink(self):
-        self.history.shrink()
-        self.obs.shrink()
-
-    def _parse_answer(self, text_answer):
-        ans_dict = {}
-        ans_dict.update(self.think._parse_answer(text_answer))
-        ans_dict.update(self.memory._parse_answer(text_answer))
-        ans_dict.update(self.action_space._parse_answer(text_answer))
-        return ans_dict
-
-
-class ActionSpace(PromptElement):
-    def __init__(self, flags: Flags) -> None:
-        super().__init__()
-        self.flags = flags
-        self.action_space = _get_action_space(flags)
-
-        self._prompt = (
-            f'# Action space:\n{self.action_space.describe()}{MacNote().prompt}\n'
-        )
-        self._abstract_ex = f"""
-<action>
-{self.action_space.example_action(abstract=True)}
-</action>
-"""
-        self._concrete_ex = f"""
-<action>
-{self.action_space.example_action(abstract=False)}
-</action>
-"""
-
-    def _parse_answer(self, text_answer):
-        ans_dict = parse_html_tags_raise(
-            text_answer, keys=['action'], merge_multiple=True
-        )
-
-        try:
-            # just check if action can be mapped to python code but keep action as is
-            # the environment will be responsible for mapping it to python
-            self.action_space.to_python_code(ans_dict['action'])
-        except Exception as e:
-            raise ParseError(
-                f'Error while parsing action\n: {e}\n'
-                'Make sure your answer is restricted to the allowed actions.'
-            )
-
-        return ans_dict
-
-
-def _get_action_space(flags: Flags) -> AbstractActionSet:
-    match flags.action_space:
-        case 'python':
-            action_space = PythonActionSet(strict=flags.is_strict)
-            if flags.multi_actions:
-                warn(
-                    f'Flag action_space={repr(flags.action_space)} incompatible with multi_actions={repr(flags.multi_actions)}.',
-                    stacklevel=2,
-                )
-            if flags.demo_mode != 'off':
-                warn(
-                    f'Flag action_space={repr(flags.action_space)} incompatible with demo_mode={repr(flags.demo_mode)}.',
-                    stacklevel=2,
-                )
-            return action_space
-        case 'bid':
-            action_subsets = ['chat', 'bid']
-        case 'coord':
-            action_subsets = ['chat', 'coord']
-        case 'bid+coord':
-            action_subsets = ['chat', 'bid', 'coord']
-        case 'bid+nav':
-            action_subsets = ['chat', 'bid', 'nav']
-        case 'coord+nav':
-            action_subsets = ['chat', 'coord', 'nav']
-        case 'bid+coord+nav':
-            action_subsets = ['chat', 'bid', 'coord', 'nav']
-        case _:
-            raise NotImplementedError(
-                f'Unknown action_space {repr(flags.action_space)}'
-            )
-
-    action_space = HighLevelActionSet(
-        subsets=action_subsets,
-        multiaction=flags.multi_actions,
-        strict=flags.is_strict,
-        demo_mode=flags.demo_mode,
-    )
-
-    return action_space
-
-
-class Memory(PromptElement):
-    _prompt = ''  # provided in the abstract and concrete examples
-
-    _abstract_ex = """
-<memory>
-Write down anything you need to remember for next steps. You will be presented
-with the list of previous memories and past actions.
-</memory>
-"""
-
-    _concrete_ex = """
-<memory>
-I clicked on bid 32 to activate tab 2. The accessibility tree should mention
-focusable for elements of the form at next step.
-</memory>
-"""
-
-    def _parse_answer(self, text_answer):
-        return parse_html_tags_raise(
-            text_answer, optional_keys=['memory'], merge_multiple=True
-        )
-
-
-class Think(PromptElement):
-    _prompt = ''
-
-    _abstract_ex = """
-<think>
-Think step by step. If you need to make calculations such as coordinates, write them here. Describe the effect
-that your previous action had on the current content of the page.
-</think>
-"""
-    _concrete_ex = """
-<think>
-My memory says that I filled the first name and last name, but I can't see any
-content in the form. I need to explore different ways to fill the form. Perhaps
-the form is not visible yet or some fields are disabled. I need to replan.
-</think>
-"""
-
-    def _parse_answer(self, text_answer):
-        return parse_html_tags_raise(
-            text_answer, optional_keys=['think'], merge_multiple=True
-        )
-
-
-def diff(previous, new):
-    """Return a string showing the difference between original and new.
-
-    If the difference is above diff_threshold, return the diff string.
-    """
-    if previous == new:
-        return 'Identical', []
-
-    if len(previous) == 0 or previous is None:
-        return 'previous is empty', []
-
-    diff_gen = difflib.ndiff(previous.splitlines(), new.splitlines())
-
-    diff_lines = []
-    plus_count = 0
-    minus_count = 0
-    for line in diff_gen:
-        if line.strip().startswith('+'):
-            diff_lines.append(line)
-            plus_count += 1
-        elif line.strip().startswith('-'):
-            diff_lines.append(line)
-            minus_count += 1
-        else:
-            continue
-
-    header = f'{plus_count} lines added and {minus_count} lines removed:'
-
-    return header, diff_lines
-
-
-class Diff(Shrinkable):
-    def __init__(
-        self, previous, new, prefix='', max_line_diff=20, shrink_speed=2, visible=True
-    ) -> None:
-        super().__init__(visible=visible)
-        self.max_line_diff = max_line_diff
-        self.header, self.diff_lines = diff(previous, new)
-        self.shrink_speed = shrink_speed
-        self.prefix = prefix
-
-    def shrink(self):
-        self.max_line_diff -= self.shrink_speed
-        self.max_line_diff = max(1, self.max_line_diff)
-
-    @property
-    def _prompt(self) -> str:  # type: ignore
-        diff_str = '\n'.join(self.diff_lines[: self.max_line_diff])
-        if len(self.diff_lines) > self.max_line_diff:
-            original_count = len(self.diff_lines)
-            diff_str = f'{diff_str}\nDiff truncated, {original_count - self.max_line_diff} changes now shown.'
-        return f'{self.prefix}{self.header}\n{diff_str}\n'
-
-
-class HistoryStep(Shrinkable):
-    def __init__(
-        self, previous_obs, current_obs, action, memory, flags: Flags, shrink_speed=1
-    ) -> None:
-        super().__init__()
-        self.html_diff = Diff(
-            previous_obs[flags.html_type],
-            current_obs[flags.html_type],
-            prefix='\n### HTML diff:\n',
-            shrink_speed=shrink_speed,
-            visible=lambda: flags.use_html and flags.use_diff,
-        )
-        self.ax_tree_diff = Diff(
-            previous_obs['axtree_txt'],
-            current_obs['axtree_txt'],
-            prefix='\n### Accessibility tree diff:\n',
-            shrink_speed=shrink_speed,
-            visible=lambda: flags.use_ax_tree and flags.use_diff,
-        )
-        self.error = Error(
-            current_obs['last_action_error'],
-            visible=(
-                flags.use_error_logs
-                and current_obs['last_action_error']
-                and flags.use_past_error_logs
-            ),
-            prefix='### ',
-        )
-        self.shrink_speed = shrink_speed
-        self.action = action
-        self.memory = memory
-        self.flags = flags
-
-    def shrink(self):
-        super().shrink()
-        self.html_diff.shrink()
-        self.ax_tree_diff.shrink()
-
-    @property
-    def _prompt(self) -> str:  # type: ignore
-        prompt = ''
-
-        if self.flags.use_action_history:
-            prompt += f'\n### Action:\n{self.action}\n'
-
-        prompt += (
-            f'{self.error.prompt}{self.html_diff.prompt}{self.ax_tree_diff.prompt}'
-        )
-
-        if self.flags.use_memory and self.memory is not None:
-            prompt += f'\n### Memory:\n{self.memory}\n'
-
-        return prompt
-
-
-class History(Shrinkable):
-    def __init__(
-        self, history_obs, actions, memories, thoughts, flags: Flags, shrink_speed=1
-    ) -> None:
-        super().__init__(visible=flags.use_history)
-        assert len(history_obs) == len(actions) + 1
-        assert len(history_obs) == len(memories) + 1
-
-        self.shrink_speed = shrink_speed
-        self.history_steps: list[HistoryStep] = []
-
-        for i in range(1, len(history_obs)):
-            self.history_steps.append(
-                HistoryStep(
-                    history_obs[i - 1],
-                    history_obs[i],
-                    actions[i - 1],
-                    memories[i - 1],
-                    flags,
-                )
-            )
-
-    def shrink(self):
-        """Shrink individual steps"""
-        # TODO set the shrink speed of older steps to be higher
-        super().shrink()
-        for step in self.history_steps:
-            step.shrink()
-
-    @property
-    def _prompt(self):
-        prompts = ['# History of interaction with the task:\n']
-        for i, step in enumerate(self.history_steps):
-            prompts.append(f'## step {i}')
-            prompts.append(step.prompt)
-        return '\n'.join(prompts) + '\n'
-
-
-if __name__ == '__main__':
-    html_template = """
-    <html>
-    <body>
-    <div>
-    Hello World.
-    Step {}.
-    </div>
-    </body>
-    </html>
-    """
-
-    OBS_HISTORY = [
-        {
-            'goal': 'do this and that',
-            'pruned_html': html_template.format(1),
-            'axtree_txt': '[1] Click me',
-            'last_action_error': '',
-        },
-        {
-            'goal': 'do this and that',
-            'pruned_html': html_template.format(2),
-            'axtree_txt': '[1] Click me',
-            'last_action_error': '',
-        },
-        {
-            'goal': 'do this and that',
-            'pruned_html': html_template.format(3),
-            'axtree_txt': '[1] Click me',
-            'last_action_error': 'Hey, there is an error now',
-        },
-    ]
-    ACTIONS = ["click('41')", "click('42')"]
-    MEMORIES = ['memory A', 'memory B']
-    THOUGHTS = ['thought A', 'thought B']
-
-    flags = Flags(
-        use_html=True,
-        use_ax_tree=True,
-        use_thinking=True,
-        use_error_logs=True,
-        use_past_error_logs=True,
-        use_history=True,
-        use_action_history=True,
-        use_memory=True,
-        use_diff=True,
-        html_type='pruned_html',
-        use_concrete_example=True,
-        use_abstract_example=True,
-        use_screenshot=False,
-        multi_actions=True,
-    )
-
-    print(
-        MainPrompt(
-            obs_history=OBS_HISTORY,
-            actions=ACTIONS,
-            memories=MEMORIES,
-            thoughts=THOUGHTS,
-            flags=flags,
-        ).prompt
-    )

agenthub/browsing_agent/response_parser.py

@@ -1,88 +0,0 @@
-import ast
-
-from opendevin.controller.action_parser import ActionParser, ResponseParser
-from opendevin.core.logger import opendevin_logger as logger
-from opendevin.events.action import (
-    Action,
-    BrowseInteractiveAction,
-)
-
-
-class BrowsingResponseParser(ResponseParser):
-    def __init__(self):
-        # Need to pay attention to the item order in self.action_parsers
-        super().__init__()
-        self.action_parsers = [BrowsingActionParserMessage()]
-        self.default_parser = BrowsingActionParserBrowseInteractive()
-
-    def parse(self, response: str) -> Action:
-        action_str = self.parse_response(response)
-        return self.parse_action(action_str)
-
-    def parse_response(self, response) -> str:
-        action_str = response['choices'][0]['message']['content']
-        if action_str is None:
-            return ''
-        action_str = action_str.strip()
-        if not action_str.endswith('```'):
-            action_str = action_str + ')```'
-        logger.info(action_str)
-        return action_str
-
-    def parse_action(self, action_str: str) -> Action:
-        for action_parser in self.action_parsers:
-            if action_parser.check_condition(action_str):
-                return action_parser.parse(action_str)
-        return self.default_parser.parse(action_str)
-
-
-class BrowsingActionParserMessage(ActionParser):
-    """Parser action:
-    - BrowseInteractiveAction(browser_actions) - unexpected response format, message back to user
-    """
-
-    def __init__(
-        self,
-    ):
-        pass
-
-    def check_condition(self, action_str: str) -> bool:
-        return '```' not in action_str
-
-    def parse(self, action_str: str) -> Action:
-        msg = f'send_msg_to_user("""{action_str}""")'
-        return BrowseInteractiveAction(
-            browser_actions=msg,
-            thought=action_str,
-            browsergym_send_msg_to_user=action_str,
-        )
-
-
-class BrowsingActionParserBrowseInteractive(ActionParser):
-    """Parser action:
-    - BrowseInteractiveAction(browser_actions) - handle send message to user function call in BrowserGym
-    """
-
-    def __init__(
-        self,
-    ):
-        pass
-
-    def check_condition(self, action_str: str) -> bool:
-        return True
-
-    def parse(self, action_str: str) -> Action:
-        thought = action_str.split('```')[0].strip()
-        action_str = action_str.split('```')[1].strip()
-        msg_content = ''
-        for sub_action in action_str.split('\n'):
-            if 'send_msg_to_user(' in sub_action:
-                tree = ast.parse(sub_action)
-                args = tree.body[0].value.args  # type: ignore
-                msg_content = args[0].value
-
-        return BrowseInteractiveAction(
-            browser_actions=action_str,
-            thought=thought,
-            browsergym_send_msg_to_user=msg_content,
-        )

agenthub/browsing_agent/utils.py

@@ -1,158 +0,0 @@
-import collections
-import re
-from warnings import warn
-
-import yaml
-
-
-def yaml_parser(message):
-    """Parse a yaml message for the retry function."""
-    # saves gpt-3.5 from some yaml parsing errors
-    message = re.sub(r':\s*\n(?=\S|\n)', ': ', message)
-
-    try:
-        value = yaml.safe_load(message)
-        valid = True
-        retry_message = ''
-    except yaml.YAMLError as e:
-        warn(str(e), stacklevel=2)
-        value = {}
-        valid = False
-        retry_message = "Your response is not a valid yaml. Please try again and be careful to the format. Don't add any apology or comment, just the answer."
-    return value, valid, retry_message
-
-
-def _compress_chunks(text, identifier, skip_list, split_regex='\n\n+'):
-    """Compress a string by replacing redundant chunks by identifiers. Chunks are defined by the split_regex."""
-    text_list = re.split(split_regex, text)
-    text_list = [chunk.strip() for chunk in text_list]
-    counter = collections.Counter(text_list)
-    def_dict = {}
-    id = 0
-
-    # Store items that occur more than once in a dictionary
-    for item, count in counter.items():
-        if count > 1 and item not in skip_list and len(item) > 10:
-            def_dict[f'{identifier}-{id}'] = item
-            id += 1
-
-    # Replace redundant items with their identifiers in the text
-    compressed_text = '\n'.join(text_list)
-    for key, value in def_dict.items():
-        compressed_text = compressed_text.replace(value, key)
-
-    return def_dict, compressed_text
-
-
-def compress_string(text):
-    """Compress a string by replacing redundant paragraphs and lines with identifiers."""
-    # Perform paragraph-level compression
-    def_dict, compressed_text = _compress_chunks(
-        text, identifier='§', skip_list=[], split_regex='\n\n+'
-    )
-
-    # Perform line-level compression, skipping any paragraph identifiers
-    line_dict, compressed_text = _compress_chunks(
-        compressed_text, '¶', list(def_dict.keys()), split_regex='\n+'
-    )
-    def_dict.update(line_dict)
-
-    # Create a definitions section
-    def_lines = ['<definitions>']
-    for key, value in def_dict.items():
-        def_lines.append(f'{key}:\n{value}')
-    def_lines.append('</definitions>')
-    definitions = '\n'.join(def_lines)
-
-    return definitions + '\n' + compressed_text
-
-
-def extract_html_tags(text, keys):
-    """Extract the content within HTML tags for a list of keys.
-
-    Parameters
-    ----------
-    text : str
-        The input string containing the HTML tags.
-    keys : list of str
-        The HTML tags to extract the content from.
-
-    Returns:
-    -------
-    dict
-        A dictionary mapping each key to a list of subset in `text` that match the key.
-
-    Notes:
-    -----
-    All text and keys will be converted to lowercase before matching.
-
-    """
-    content_dict = {}
-    # text = text.lower()
-    # keys = set([k.lower() for k in keys])
-    for key in keys:
-        pattern = f'<{key}>(.*?)</{key}>'
-        matches = re.findall(pattern, text, re.DOTALL)
-        if matches:
-            content_dict[key] = [match.strip() for match in matches]
-    return content_dict
-
-
-class ParseError(Exception):
-    pass
-
-
-def parse_html_tags_raise(text, keys=(), optional_keys=(), merge_multiple=False):
-    """A version of parse_html_tags that raises an exception if the parsing is not successful."""
-    content_dict, valid, retry_message = parse_html_tags(
-        text, keys, optional_keys, merge_multiple=merge_multiple
-    )
-    if not valid:
-        raise ParseError(retry_message)
-    return content_dict
-
-
-def parse_html_tags(text, keys=(), optional_keys=(), merge_multiple=False):
-    """Satisfy the parse api, extracts 1 match per key and validates that all keys are present
-
-    Parameters
-    ----------
-    text : str
-        The input string containing the HTML tags.
-    keys : list of str
-        The HTML tags to extract the content from.
-    optional_keys : list of str
-        The HTML tags to extract the content from, but are optional.
-
-    Returns:
-    -------
-    dict
-        A dictionary mapping each key to subset of `text` that match the key.
-    bool
-        Whether the parsing was successful.
-    str
-        A message to be displayed to the agent if the parsing was not successful.
-    """
-    all_keys = tuple(keys) + tuple(optional_keys)
-    content_dict = extract_html_tags(text, all_keys)
-    retry_messages = []
-
-    for key in all_keys:
-        if key not in content_dict:
-            if key not in optional_keys:
-                retry_messages.append(f'Missing the key <{key}> in the answer.')
-        else:
-            val = content_dict[key]
-            content_dict[key] = val[0]
-            if len(val) > 1:
-                if not merge_multiple:
-                    retry_messages.append(
-                        f'Found multiple instances of the key {key}. You should have only one of them.'
-                    )
-                else:
-                    # merge the multiple instances
-                    content_dict[key] = '\n'.join(val)
-
-    valid = len(retry_messages) == 0
-    retry_message = '\n'.join(retry_messages)
-    return content_dict, valid, retry_message

agenthub/codeact_agent/README.md

@@ -1,6 +1,6 @@
 # CodeAct Agent Framework
 
-This folder implements the CodeAct idea ([paper](https://arxiv.org/abs/2402.01030), [tweet](https://twitter.com/xingyaow_/status/1754556835703751087)) that consolidates LLM agents’ **act**ions into a unified **code** action space for both *simplicity* and *performance* (see paper for more details).
+This folder implements the CodeAct idea ([paper](https://arxiv.org/abs/2402.13463), [tweet](https://twitter.com/xingyaow_/status/1754556835703751087)) that consolidates LLM agents’ **act**ions into a unified **code** action space for both *simplicity* and *performance* (see paper for more details).
 
 The conceptual idea is illustrated below. At each turn, the agent can:
 

agenthub/codeact_agent/action_parser.py

@@ -1,177 +0,0 @@
-import re
-
-from opendevin.controller.action_parser import ActionParser, ResponseParser
-from opendevin.events.action import (
-    Action,
-    AgentDelegateAction,
-    AgentFinishAction,
-    CmdRunAction,
-    IPythonRunCellAction,
-    MessageAction,
-)
-
-
-class CodeActResponseParser(ResponseParser):
-    """Parser action:
-    - CmdRunAction(command) - bash command to run
-    - IPythonRunCellAction(code) - IPython code to run
-    - AgentDelegateAction(agent, inputs) - delegate action for (sub)task
-    - MessageAction(content) - Message action to run (e.g. ask for clarification)
-    - AgentFinishAction() - end the interaction
-    """
-
-    def __init__(self):
-        # Need pay attention to the item order in self.action_parsers
-        super().__init__()
-        self.action_parsers = [
-            CodeActActionParserFinish(),
-            CodeActActionParserCmdRun(),
-            CodeActActionParserIPythonRunCell(),
-            CodeActActionParserAgentDelegate(),
-        ]
-        self.default_parser = CodeActActionParserMessage()
-
-    def parse(self, response) -> Action:
-        action_str = self.parse_response(response)
-        return self.parse_action(action_str)
-
-    def parse_response(self, response) -> str:
-        action = response.choices[0].message.content
-        if action is None:
-            return ''
-        for lang in ['bash', 'ipython', 'browse']:
-            if f'<execute_{lang}>' in action and f'</execute_{lang}>' not in action:
-                action += f'</execute_{lang}>'
-        return action
-
-    def parse_action(self, action_str: str) -> Action:
-        for action_parser in self.action_parsers:
-            if action_parser.check_condition(action_str):
-                return action_parser.parse(action_str)
-        return self.default_parser.parse(action_str)
-
-
-class CodeActActionParserFinish(ActionParser):
-    """Parser action:
-    - AgentFinishAction() - end the interaction
-    """
-
-    def __init__(
-        self,
-    ):
-        self.finish_command = None
-
-    def check_condition(self, action_str: str) -> bool:
-        self.finish_command = re.search(r'<finish>.*</finish>', action_str, re.DOTALL)
-        return self.finish_command is not None
-
-    def parse(self, action_str: str) -> Action:
-        assert (
-            self.finish_command is not None
-        ), 'self.finish_command should not be None when parse is called'
-        thought = action_str.replace(self.finish_command.group(0), '').strip()
-        return AgentFinishAction(thought=thought)
-
-
-class CodeActActionParserCmdRun(ActionParser):
-    """Parser action:
-    - CmdRunAction(command) - bash command to run
-    - AgentFinishAction() - end the interaction
-    """
-
-    def __init__(
-        self,
-    ):
-        self.bash_command = None
-
-    def check_condition(self, action_str: str) -> bool:
-        self.bash_command = re.search(
-            r'<execute_bash>(.*?)</execute_bash>', action_str, re.DOTALL
-        )
-        return self.bash_command is not None
-
-    def parse(self, action_str: str) -> Action:
-        assert (
-            self.bash_command is not None
-        ), 'self.bash_command should not be None when parse is called'
-        thought = action_str.replace(self.bash_command.group(0), '').strip()
-        # a command was found
-        command_group = self.bash_command.group(1).strip()
-        if command_group.strip() == 'exit':
-            return AgentFinishAction(thought=thought)
-        return CmdRunAction(command=command_group, thought=thought)
-
-
-class CodeActActionParserIPythonRunCell(ActionParser):
-    """Parser action:
-    - IPythonRunCellAction(code) - IPython code to run
-    """
-
-    def __init__(
-        self,
-    ):
-        self.python_code = None
-        self.jupyter_kernel_init_code: str = 'from agentskills import *'
-
-    def check_condition(self, action_str: str) -> bool:
-        self.python_code = re.search(
-            r'<execute_ipython>(.*?)</execute_ipython>', action_str, re.DOTALL
-        )
-        return self.python_code is not None
-
-    def parse(self, action_str: str) -> Action:
-        assert (
-            self.python_code is not None
-        ), 'self.python_code should not be None when parse is called'
-        code_group = self.python_code.group(1).strip()
-        thought = action_str.replace(self.python_code.group(0), '').strip()
-        return IPythonRunCellAction(
-            code=code_group,
-            thought=thought,
-            kernel_init_code=self.jupyter_kernel_init_code,
-        )
-
-
-class CodeActActionParserAgentDelegate(ActionParser):
-    """Parser action:
-    - AgentDelegateAction(agent, inputs) - delegate action for (sub)task
-    """
-
-    def __init__(
-        self,
-    ):
-        self.agent_delegate = None
-
-    def check_condition(self, action_str: str) -> bool:
-        self.agent_delegate = re.search(
-            r'<execute_browse>(.*)</execute_browse>', action_str, re.DOTALL
-        )
-        return self.agent_delegate is not None
-
-    def parse(self, action_str: str) -> Action:
-        assert (
-            self.agent_delegate is not None
-        ), 'self.agent_delegate should not be None when parse is called'
-        thought = action_str.replace(self.agent_delegate.group(0), '').strip()
-        browse_actions = self.agent_delegate.group(1).strip()
-        task = f'{thought}. I should start with: {browse_actions}'
-        return AgentDelegateAction(agent='BrowsingAgent', inputs={'task': task})
-
-
-class CodeActActionParserMessage(ActionParser):
-    """Parser action:
-    - MessageAction(content) - Message action to run (e.g. ask for clarification)
-    """
-
-    def __init__(
-        self,
-    ):
-        pass
-
-    def check_condition(self, action_str: str) -> bool:
-        # We assume the LLM is GOOD enough that when it returns pure natural language
-        # it wants to talk to the user
-        return True
-
-    def parse(self, action_str: str) -> Action:
-        return MessageAction(content=action_str, wait_for_response=True)

agenthub/codeact_agent/codeact_agent.py

@@ -1,4 +1,5 @@
-from agenthub.codeact_agent.action_parser import CodeActResponseParser
+import re
+
 from agenthub.codeact_agent.prompt import (
     COMMAND_DOCS,
     EXAMPLES,
@@ -8,53 +9,92 @@ from agenthub.codeact_agent.prompt import (
 )
 from opendevin.controller.agent import Agent
 from opendevin.controller.state.state import State
+from opendevin.core.logger import opendevin_logger as logger
 from opendevin.events.action import (
     Action,
-    AgentDelegateAction,
     AgentFinishAction,
+    BrowseInteractiveAction,
     CmdRunAction,
     IPythonRunCellAction,
     MessageAction,
 )
 from opendevin.events.observation import (
-    AgentDelegateObservation,
+    BrowserOutputObservation,
     CmdOutputObservation,
     IPythonRunCellObservation,
 )
-from opendevin.events.observation.observation import Observation
-from opendevin.events.serialization.event import truncate_content
 from opendevin.llm.llm import LLM
 from opendevin.runtime.plugins import (
-    AgentSkillsRequirement,
     JupyterRequirement,
     PluginRequirement,
+    SWEAgentCommandsRequirement,
 )
-from opendevin.runtime.tools import RuntimeTool
 
 ENABLE_GITHUB = True
 
 
-# FIXME: We can tweak these two settings to create MicroAgents specialized toward different area
-def get_system_message() -> str:
-    if ENABLE_GITHUB:
-        return f'{SYSTEM_PREFIX}\n{GITHUB_MESSAGE}\n\n{COMMAND_DOCS}\n\n{SYSTEM_SUFFIX}'
-    else:
-        return f'{SYSTEM_PREFIX}\n\n{COMMAND_DOCS}\n\n{SYSTEM_SUFFIX}'
+def parse_response(response) -> str:
+    action = response.choices[0].message.content
+    for lang in ['bash', 'ipython', 'browse']:
+        if f'<execute_{lang}>' in action and f'</execute_{lang}>' not in action:
+            action += f'</execute_{lang}>'
+    return action
 
 
-def get_in_context_example() -> str:
-    return EXAMPLES
+def truncate_observation(observation: str, max_chars: int = 10_000) -> str:
+    """
+    Truncate the middle of the observation if it is too long.
+    """
+    if len(observation) <= max_chars:
+        return observation
+    half = max_chars // 2
+    return (
+        observation[:half]
+        + '\n[... Observation truncated due to length ...]\n'
+        + observation[-half:]
+    )
+
+
+def swe_agent_edit_hack(bash_command: str) -> str:
+    """
+    Hack to handle the SWE-agent edit command. The vanilla edit command will hang the SSHBox.
+
+    REPLACE THIS:
+    edit 683:693
+            try:
+                return list(urlsplit(url))
+            except ValueError:
+                raise ValidationError(self.error_messages['invalid'], code='invalid')
+    end_of_edit
+
+    WITH THIS:
+    edit 683:693 <<EOF
+            try:
+                return list(urlsplit(url))
+            except ValueError:
+                raise ValidationError(self.error_messages['invalid'], code='invalid')
+    EOF
+    """
+    if 'edit' in bash_command:
+        # edit\s(\d+):(\d+)([\s\S]*)end_of_edit
+        # replace
+        bash_command = re.sub(
+            r'edit\s(\d+):(\d+)([\s\S]*?)end_of_edit',
+            r'edit \1:\2 <<EOF\3EOF',
+            bash_command,
+        )
+    return bash_command
 
 
 class CodeActAgent(Agent):
-    VERSION = '1.8'
+    VERSION = '1.3'
     """
     The Code Act Agent is a minimalist agent.
     The agent works by passing the model a list of action-observation pairs and prompting the model to take the next step.
 
     ### Overview
 
-    This agent implements the CodeAct idea ([paper](https://arxiv.org/abs/2402.01030), [tweet](https://twitter.com/xingyaow_/status/1754556835703751087)) that consolidates LLM agents’ **act**ions into a unified **code** action space for both *simplicity* and *performance* (see paper for more details).
+    This agent implements the CodeAct idea ([paper](https://arxiv.org/abs/2402.13463), [tweet](https://twitter.com/xingyaow_/status/1754556835703751087)) that consolidates LLM agents’ **act**ions into a unified **code** action space for both *simplicity* and *performance* (see paper for more details).
 
     The conceptual idea is illustrated below. At each turn, the agent can:
 
@@ -85,24 +125,22 @@ class CodeActAgent(Agent):
     """
 
     sandbox_plugins: list[PluginRequirement] = [
-        # NOTE: AgentSkillsRequirement need to go before JupyterRequirement, since
-        # AgentSkillsRequirement provides a lot of Python functions,
-        # and it needs to be initialized before Jupyter for Jupyter to use those functions.
-        AgentSkillsRequirement(),
         JupyterRequirement(),
+        SWEAgentCommandsRequirement(),
     ]
-    runtime_tools: list[RuntimeTool] = [RuntimeTool.BROWSER]
 
-    system_message: str = get_system_message()
-    in_context_example: str = f"Here is an example of how you can interact with the environment for task solving:\n{get_in_context_example()}\n\nNOW, LET'S START!"
-
-    action_parser = CodeActResponseParser()
+    system_message: str = (
+        f'{SYSTEM_PREFIX}\n{GITHUB_MESSAGE}\n\n{COMMAND_DOCS}\n\n{SYSTEM_SUFFIX}'
+        if ENABLE_GITHUB
+        else f'{SYSTEM_PREFIX}\n\n{COMMAND_DOCS}\n\n{SYSTEM_SUFFIX}'
+    )
 
     def __init__(
         self,
         llm: LLM,
     ) -> None:
-        """Initializes a new instance of the CodeActAgent class.
+        """
+        Initializes a new instance of the CodeActAgent class.
 
         Parameters:
         - llm (LLM): The llm to be used by this agent
@@ -110,89 +148,78 @@ class CodeActAgent(Agent):
         super().__init__(llm)
         self.reset()
 
-    def action_to_str(self, action: Action) -> str:
-        if isinstance(action, CmdRunAction):
-            return (
-                f'{action.thought}\n<execute_bash>\n{action.command}\n</execute_bash>'
-            )
-        elif isinstance(action, IPythonRunCellAction):
-            return f'{action.thought}\n<execute_ipython>\n{action.code}\n</execute_ipython>'
-        elif isinstance(action, AgentDelegateAction):
-            return f'{action.thought}\n<execute_browse>\n{action.inputs["task"]}\n</execute_browse>'
-        elif isinstance(action, MessageAction):
-            return action.content
-        return ''
-
-    def get_action_message(self, action: Action) -> dict[str, str] | None:
-        if (
-            isinstance(action, AgentDelegateAction)
-            or isinstance(action, CmdRunAction)
-            or isinstance(action, IPythonRunCellAction)
-            or isinstance(action, MessageAction)
-        ):
-            return {
-                'role': 'user' if action.source == 'user' else 'assistant',
-                'content': self.action_to_str(action),
-            }
-        return None
-
-    def get_observation_message(self, obs: Observation) -> dict[str, str] | None:
-        max_message_chars = self.llm.config.max_message_chars
-        if isinstance(obs, CmdOutputObservation):
-            content = 'OBSERVATION:\n' + truncate_content(
-                obs.content, max_message_chars
-            )
-            content += (
-                f'\n[Command {obs.command_id} finished with exit code {obs.exit_code}]'
-            )
-            return {'role': 'user', 'content': content}
-        elif isinstance(obs, IPythonRunCellObservation):
-            content = 'OBSERVATION:\n' + obs.content
-            # replace base64 images with a placeholder
-            splitted = content.split('\n')
-            for i, line in enumerate(splitted):
-                if '![image](data:image/png;base64,' in line:
-                    splitted[i] = (
-                        '![image](data:image/png;base64, ...) already displayed to user'
-                    )
-            content = '\n'.join(splitted)
-            content = truncate_content(content, max_message_chars)
-            return {'role': 'user', 'content': content}
-        elif isinstance(obs, AgentDelegateObservation):
-            content = 'OBSERVATION:\n' + truncate_content(
-                str(obs.outputs), max_message_chars
-            )
-            return {'role': 'user', 'content': content}
-        return None
-
     def reset(self) -> None:
-        """Resets the CodeAct Agent."""
+        """
+        Resets the CodeAct Agent.
+        """
         super().reset()
+        self.messages: list[dict[str, str]] = [
+            {'role': 'system', 'content': self.system_message},
+            {
+                'role': 'user',
+                'content': f"Here is an example of how you can interact with the environment for task solving:\n{EXAMPLES}\n\nNOW, LET'S START!",
+            },
+        ]
+        self.cost_accumulator = 0
 
     def step(self, state: State) -> Action:
-        """Performs one step using the CodeAct Agent.
+        """
+        Performs one step using the CodeAct Agent.
         This includes gathering info on previous steps and prompting the model to make a command to execute.
 
         Parameters:
-        - state (State): used to get updated info
+        - state (State): used to get updated info and background commands
 
         Returns:
         - CmdRunAction(command) - bash command to run
         - IPythonRunCellAction(code) - IPython code to run
-        - AgentDelegateAction(agent, inputs) - delegate action for (sub)task
+        - BrowseInteractiveAction(browsergym_command) - BrowserGym commands to run
         - MessageAction(content) - Message action to run (e.g. ask for clarification)
         - AgentFinishAction() - end the interaction
         """
-        # if we're done, go back
-        latest_user_message = state.history.get_last_user_message()
-        if latest_user_message and latest_user_message.strip() == '/exit':
-            return AgentFinishAction()
 
-        # prepare what we want to send to the LLM
-        messages: list[dict[str, str]] = self._get_messages(state)
+        updated_info = state.updated_info
+        if updated_info:
+            for prev_action, obs in updated_info:
+                if (
+                    isinstance(prev_action, MessageAction)
+                    and prev_action.source == 'user'
+                ):
+                    self.messages.append(
+                        {'role': 'user', 'content': prev_action.content}
+                    )
+                    if prev_action.content.strip() == '/exit':
+                        # User wants to exit
+                        return AgentFinishAction()
+
+                if isinstance(obs, CmdOutputObservation):
+                    content = 'OBSERVATION:\n' + truncate_observation(obs.content)
+                    content += f'\n[Command {obs.command_id} finished with exit code {obs.exit_code}]]'
+                    self.messages.append({'role': 'user', 'content': content})
+                elif isinstance(obs, IPythonRunCellObservation):
+                    content = 'OBSERVATION:\n' + obs.content
+                    # replace base64 images with a placeholder
+                    splitted = content.split('\n')
+                    for i, line in enumerate(splitted):
+                        if '![image](data:image/png;base64,' in line:
+                            splitted[i] = (
+                                '![image](data:image/png;base64, ...) already displayed to user'
+                            )
+                    content = '\n'.join(splitted)
+                    content = truncate_observation(content)
+                    self.messages.append({'role': 'user', 'content': content})
+                elif isinstance(obs, BrowserOutputObservation):
+                    content = 'OBSERVATION:\n' + truncate_observation(obs.content)
+                    self.messages.append({'role': 'user', 'content': content})
+
+        latest_user_message = [m for m in self.messages if m['role'] == 'user'][-1]
+        if latest_user_message:
+            latest_user_message['content'] += (
+                f'\n\nENVIRONMENT REMINDER: You have {state.max_iterations - state.iteration} turns left to complete the task.'
+            )
 
         response = self.llm.completion(
-            messages=messages,
+            messages=self.messages,
             stop=[
                 '</execute_ipython>',
                 '</execute_bash>',
@@ -200,37 +227,62 @@ class CodeActAgent(Agent):
             ],
             temperature=0.0,
         )
-        return self.action_parser.parse(response)
 
-    def _get_messages(self, state: State) -> list[dict[str, str]]:
-        messages = [
-            {'role': 'system', 'content': self.system_message},
-            {'role': 'user', 'content': self.in_context_example},
-        ]
+        self.log_cost(response)
 
-        for event in state.history.get_events():
-            # create a regular message from an event
-            if isinstance(event, Action):
-                message = self.get_action_message(event)
-            elif isinstance(event, Observation):
-                message = self.get_observation_message(event)
-            else:
-                raise ValueError(f'Unknown event type: {type(event)}')
+        action_str: str = parse_response(response)
+        state.num_of_chars += sum(
+            len(message['content']) for message in self.messages
+        ) + len(action_str)
+        self.messages.append({'role': 'assistant', 'content': action_str})
 
-            # add regular message
-            if message:
-                messages.append(message)
+        if finish_command := re.search(r'<finish>.*</finish>', action_str, re.DOTALL):
+            thought = action_str.replace(finish_command.group(0), '').strip()
+            return AgentFinishAction(thought=thought)
+        if bash_command := re.search(
+            r'<execute_bash>(.*)</execute_bash>', action_str, re.DOTALL
+        ):
+            # remove the command from the action string to get thought
+            thought = action_str.replace(bash_command.group(0), '').strip()
+            # a command was found
+            command_group = bash_command.group(1).strip()
+            command_group = swe_agent_edit_hack(command_group)
 
-        # the latest user message is important:
-        # we want to remind the agent of the environment constraints
-        latest_user_message = next(
-            (m for m in reversed(messages) if m['role'] == 'user'), None
-        )
-
-        # add a reminder to the prompt
-        if latest_user_message:
-            latest_user_message['content'] += (
-                f'\n\nENVIRONMENT REMINDER: You have {state.max_iterations - state.iteration} turns left to complete the task. When finished reply with <finish></finish>'
+            if command_group.strip() == 'exit':
+                return AgentFinishAction()
+            return CmdRunAction(command=command_group, thought=thought)
+        elif python_code := re.search(
+            r'<execute_ipython>(.*)</execute_ipython>', action_str, re.DOTALL
+        ):
+            # a code block was found
+            code_group = python_code.group(1).strip()
+            thought = action_str.replace(python_code.group(0), '').strip()
+            return IPythonRunCellAction(code=code_group, thought=thought)
+        elif browse_command := re.search(
+            r'<execute_browse>(.*)</execute_browse>', action_str, re.DOTALL
+        ):
+            # BrowserGym actions was found
+            browse_actions = browse_command.group(1).strip()
+            thought = action_str.replace(browse_command.group(0), '').strip()
+            return BrowseInteractiveAction(
+                browser_actions=browse_actions, thought=thought
             )
+        else:
+            # We assume the LLM is GOOD enough that when it returns pure natural language
+            # it want to talk to the user
+            return MessageAction(content=action_str, wait_for_response=True)
 
-        return messages
+    def search_memory(self, query: str) -> list[str]:
+        raise NotImplementedError('Implement this abstract method')
+
+    def log_cost(self, response):
+        try:
+            cur_cost = self.llm.completion_cost(response)
+        except Exception:
+            cur_cost = 0
+        self.cost_accumulator += cur_cost
+        logger.info(
+            'Cost: %.2f USD | Accumulated Cost: %.2f USD',
+            cur_cost,
+            self.cost_accumulator,
+        )

agenthub/codeact_agent/prompt.py

@@ -1,74 +1,70 @@
-from opendevin.runtime.plugins import AgentSkillsRequirement
+from opendevin.runtime.plugins import SWEAgentCommandsRequirement
 
-_AGENT_SKILLS_DOCS = AgentSkillsRequirement.documentation
+_SWEAGENT_BASH_DOCS = '\n'.join(
+    filter(
+        lambda x: not x.startswith('submit'),
+        SWEAgentCommandsRequirement.documentation.split('\n'),
+    )
+)
+# _SWEAGENT_BASH_DOCS content below:
+"""
+open <path> [<line_number>] - opens the file at the given path in the editor. If line_number is provided, the window will be move to include that line
+goto <line_number> - moves the window to show <line_number>
+scroll_down - moves the window down {WINDOW} lines
+scroll_up - moves the window down {WINDOW} lines
+create <filename> - creates and opens a new file with the given name
+search_dir <search_term> [<dir>] - searches for search_term in all files in dir. If dir is not provided, searches in the current directory
+search_file <search_term> [<file>] - searches for search_term in file. If file is not provided, searches in the current open file
+find_file <file_name> [<dir>] - finds all files with the given name in dir. If dir is not provided, searches in the current directory
+edit <start_line>:<end_line>
+<replacement_text>
+end_of_edit - replaces lines <start_line> through <end_line> (inclusive) with the given text in the open file. The replacement text is terminated by a line with only end_of_edit on it. All of the <replacement text> will be entered, so make sure your indentation is formatted properly. Python files will be checked for syntax errors after the edit. If the system detects a syntax error, the edit will not be executed. Simply try to edit the file again, but make sure to read the error message and modify the edit command you issue accordingly. Issuing the same command a second time will just lead to the same error message again. Remember, the file must be open before editing.
+"""
 
 COMMAND_DOCS = (
-    '\nApart from the standard Python library, the assistant can also use the following functions (already imported) in <execute_ipython> environment:\n'
-    f'{_AGENT_SKILLS_DOCS}'
-    "Please note that THE `edit_file_by_replace`, `append_file` and `insert_content_at_line` FUNCTIONS REQUIRE PROPER INDENTATION. If the assistant would like to add the line '        print(x)', it must fully write that out, with all those spaces before the code! Indentation is important and code that is not indented correctly will fail and require fixing before it can be run."
+    '\nApart from the standard bash commands, you can also use the following special commands in <execute_bash> environment:\n'
+    f'{_SWEAGENT_BASH_DOCS}'
+    "Please note that THE EDIT COMMAND REQUIRES PROPER INDENTATION. If you'd like to add the line '        print(x)' you must fully write that out, with all those spaces before the code! Indentation is important and code that is not indented correctly will fail and require fixing before it can be run."
 )
 
-# ======= SYSTEM MESSAGE =======
-MINIMAL_SYSTEM_PREFIX = """A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
-The assistant can use an interactive Python (Jupyter Notebook) environment, executing code with <execute_ipython>.
+SYSTEM_PREFIX = """A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
+The assistant can interact with an interactive Python (Jupyter Notebook) environment and receive the corresponding output when needed. The code should be enclosed using "<execute_ipython>" tag, for example:
 <execute_ipython>
 print("Hello World!")
 </execute_ipython>
 The assistant can execute bash commands on behalf of the user by wrapping them with <execute_bash> and </execute_bash>.
-
 For example, you can list the files in the current directory by <execute_bash> ls </execute_bash>.
-Important, however: do not run interactive commands. You do not have access to stdin.
-Also, you need to handle commands that may run indefinitely and not return a result. For such cases, you should redirect the output to a file and run the command in the background to avoid blocking the execution.
-For example, to run a Python script that might run indefinitely without returning immediately, you can use the following format: <execute_bash> python3 app.py > server.log 2>&1 & </execute_bash>
-Also, if a command execution result saying like: Command: "npm start" timed out. Sending SIGINT to the process, you should also retry with running the command in the background.
-"""
+The assistant can browse the Internet with commands on behalf of the user by wrapping them with <execute_browse> and </execute_browse>.
+For example, you can browse a given URL by <execute_browse> goto("<URL>") </execute_browse>.
+The assistant should attempt fewer things at a time instead of putting too much commands OR code in one "execute" block.
+The assistant can install Python packages through bash by <execute_bash> pip install [package needed] </execute_bash> and should always import packages and define variables before starting to use them.
+The assistant should stop <execute> and provide an answer when they have already obtained the answer from the execution result.
+If the assistant encounters an import error in IPython for a newly installed package, they should try to restart the kernel and import the package again. IPython kernel can be re-started by:
+<execute_ipython>
+import IPython
+IPython.Application.instance().kernel.do_shutdown(True)  # Restart the kernel
+</execute_ipython>"""
 
-BROWSING_PREFIX = """The assistant can browse the Internet with <execute_browse> and </execute_browse>.
-For example, <execute_browse> Tell me the usa's president using google search </execute_browse>.
-Or <execute_browse> Tell me what is in http://example.com </execute_browse>.
-"""
-PIP_INSTALL_PREFIX = """The assistant can install Python packages using the %pip magic command in an IPython environment by using the following syntax: <execute_ipython> %pip install [package needed] </execute_ipython> and should always import packages and define variables before starting to use them."""
-
-SYSTEM_PREFIX = MINIMAL_SYSTEM_PREFIX + BROWSING_PREFIX + PIP_INSTALL_PREFIX
-
-GITHUB_MESSAGE = """To interact with GitHub, use the $GITHUB_TOKEN environment variable.
-For example, to push a branch `my_branch` to the GitHub repo `owner/repo`:
+GITHUB_MESSAGE = """To do any activities on GitHub, you should use the token in the $GITHUB_TOKEN environment variable.
+For instance, to push a local branch `my_branch` to the github repo `owner/repo`, you can use the following four commands:
 <execute_bash> git push https://$GITHUB_TOKEN@github.com/owner/repo.git my_branch </execute_bash>
-If $GITHUB_TOKEN is not set, ask the user to set it."""
+If you require access to GitHub but $GITHUB_TOKEN is not set, ask the user to set it for you."""
 
-SYSTEM_SUFFIX = """Responses should be concise.
-The assistant should attempt fewer things at a time instead of putting too many commands OR too much code in one "execute" block.
-Include ONLY ONE <execute_ipython>, <execute_bash>, or <execute_browse> per response, unless the assistant is finished with the task or needs more input or action from the user in order to proceed.
-If the assistant is finished with the task you MUST include <finish></finish> in your response.
-IMPORTANT: Execute code using <execute_ipython>, <execute_bash>, or <execute_browse> whenever possible.
-When handling files, try to use full paths and pwd to avoid errors.
+SYSTEM_SUFFIX = """The assistant's response should be concise.
+You should include <execute_ipython> or <execute_bash> or <execute_browse> in every one of your responses, unless you are finished with the task or need more input or action from the user in order to proceed.
+IMPORTANT: Whenever possible, execute the code for the user using <execute_ipython> or <execute_bash> or <execute_browse> instead of providing it.
 """
 
-
-# ======= EXAMPLE MESSAGE =======
 EXAMPLES = """
 --- START OF EXAMPLE ---
 
-USER: Create a list of numbers from 1 to 10, and display them in a web page at port 5000.
+USER: Can you create a list of numbers from 1 to 10, and create a web page to display them at port 5000?
 
 ASSISTANT:
-Sure! Let me create a Python file `app.py`:
+Sure! Let me write the Python code for starting a web server and save it to a file `app.py`:
 <execute_ipython>
-create_file('app.py')
-</execute_ipython>
-
-USER:
-OBSERVATION:
-[File: /workspace/app.py (1 lines total)]
-(this is the beginning of the file)
-1|
-(this is the end of the file)
-[File app.py created.]
-
-ASSISTANT:
-Now I will write the Python code for starting a web server and save it to the file `app.py`:
-<execute_ipython>
-EDITED_CODE=\"\"\"from flask import Flask
+CODE='''
+from flask import Flask
 app = Flask(__name__)
 
 @app.route('/')
@@ -77,30 +73,15 @@ def index():
     return str(numbers)
 
 if __name__ == '__main__':
-    app.run(port=5000)\"\"\"
-
-insert_content_at_line(
-  'app.py',
-  1,
-  EDITED_CODE,
-)
+    app.run(port=5000)
+'''
+with open('app.py', 'w') as f:
+    f.write(CODE)
 </execute_ipython>
 
 USER:
-OBSERVATION:
-(this is the beginning of the file)
-1|from flask import Flask
-2|app = Flask(__name__)
-3|
-4|@app.route('/')
-5|def index():
-6|    numbers = list(range(1, 11))
-7|    return str(numbers)
-8|
-9|if __name__ == '__main__':
-10|    app.run(port=5000)
-(this is the end of the file)
-[File updated (edited at line 1). Please review the changes and make sure they are correct (correct indentation, no duplicate lines, etc). Edit the file again if necessary.]
+Observation:
+[Code executed successfully with no output]
 
 ASSISTANT:
 I have created a Python file `app.py` that will display a list of numbers from 1 to 10 when you run it. Let me run the Python file for you:
@@ -109,7 +90,7 @@ python3 app.py > server.log 2>&1 &
 </execute_bash>
 
 USER:
-OBSERVATION:
+Observation:
 [1] 121[1]+  Exit 1                  python3 app.py > server.log 2>&1
 
 ASSISTANT: Looks like the server is running with PID 121 then crashed. Let me check the server log:
@@ -118,7 +99,7 @@ cat server.log
 </execute_bash>
 
 USER:
-OBSERVATION:
+Observation:
 Traceback (most recent call last):
   File "/workspace/app.py", line 2, in <module>
     from flask import Flask
@@ -131,7 +112,7 @@ pip install flask
 </execute_bash>
 
 USER:
-OBSERVATION:
+Observation:
 Defaulting to user installation because normal site-packages is not writeable
 Collecting flask
   Using cached flask-3.0.3-py3-none-any.whl (101 kB)
@@ -155,7 +136,7 @@ python3 app.py > server.log 2>&1 &
 </execute_bash>
 
 USER:
-OBSERVATION:
+Observation:
 [1] 124
 
 ASSISTANT:
@@ -165,7 +146,7 @@ cat server.log
 </execute_bash>
 
 USER:
-OBSERVATION:
+Observation:
 * Serving Flask app 'app'
  * Debug mode: off
 WARNING: This is a development server. Do not use it in a production deployment. Use a production WSGI server instead.
@@ -180,11 +161,11 @@ USER: Now browse the newly started server's homepage and show me the content.
 ASSISTANT:
 Sure! Let me browse the server's homepage at http://127.0.0.1:5000:
 <execute_browse>
-Get the content on "http://127.0.0.1:5000"
+goto("http://127.0.0.1:5000")
 </execute_browse>
 
 USER:
-OBSERVATION:
+Observation:
 [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
 
 ASSISTANT:
@@ -199,59 +180,55 @@ kill 124
 </execute_bash>
 
 USER:
-OBSERVATION:
+Observation:
 [1]+  Terminated              python3 app.py > server.log 2>&1
 
 ASSISTANT:
 The server has been stopped. Let me open the Python file and modify it.
-<execute_ipython>
-open_file('app.py')
-</execute_ipython>
+<execute_bash>
+open app.py
+</execute_bash>
 
 USER:
-[File: /workspace/app.py (10 lines total)]
-(this is the beginning of the file)
-1|from flask import Flask
-2|app = Flask(__name__)
-3|
-4|@app.route('/')
-5|def index():
-6|    numbers = list(range(1, 11))
-7|    return str(numbers)
-8|
-9|if __name__ == '__main__':
-10|    app.run(port=5000)
-(this is the end of the file)
+[File: /workspace/app.py (11 lines total)]
+1:
+2:from flask import Flask
+3:app = Flask(__name__)
+4:
+5:@app.route('/')
+6:def index():
+7:    numbers = list(range(1, 11))
+8:    return str(numbers)
+9:
+10:if __name__ == '__main__':
+11:    app.run(port=5000)
 
 ASSISTANT:
 I should edit the file to display the numbers in a table format. I should include correct indentation. Let me update the file:
-<execute_ipython>
-edit_file_by_replace(
-  'app.py',
-  to_replace="    return str(numbers)",
-  new_content="    return '<table>' + ''.join([f'<tr><td>{i}</td></tr>' for i in numbers]) + '</table>'",
-)
-</execute_ipython>
+<execute_bash>
+edit 8:8
+    return '<table>' + ''.join([f'<tr><td>{i}</td></tr>' for i in numbers]) + '</table>'
+end_of_edit
+</execute_bash>
 
 USER:
 Observation:
-[File: /workspace/app.py (10 lines total after edit)]
-(this is the beginning of the file)
-1|from flask import Flask
-2|app = Flask(__name__)
-3|
-4|@app.route('/')
-5|def index():
-6|    numbers = list(range(1, 11))
-7|    return '<table>' + ''.join([f'<tr><td>{i}</td></tr>' for i in numbers]) + '</table>'
-8|
-9|if __name__ == '__main__':
-10|    app.run(port=5000)
-(this is the end of the file)
-[File updated (edited at line 7). Please review the changes and make sure they are correct (correct indentation, no duplicate lines, etc). Edit the file again if necessary.]
+[File: /workspace/app.py (11 lines total)]
+1:
+2:from flask import Flask
+3:app = Flask(__name__)
+4:
+5:@app.route('/')
+6:def index():
+7:    numbers = list(range(1, 11))
+8:    return '<table>' + ''.join([f'<tr><td>{i}</td></tr>' for i in numbers]) + '</table>'
+9:
+10:if __name__ == '__main__':
+11:    app.run(port=5000)
+File updated. Please review the changes and make sure they are correct (correct indentation, no duplicate lines, etc). Edit the file again if necessary.
 
 ASSISTANT:
-Running the updated file:
+The file has been updated. Let me run the Python file again with the new changes:
 <execute_bash>
 python3 app.py > server.log 2>&1 &
 </execute_bash>
@@ -261,15 +238,14 @@ Observation:
 [1] 126
 
 ASSISTANT:
-The server is running on port 5000 with PID 126. You can access the list of numbers in a table format by visiting http://127.0.0.1:5000. Let me know if you have any further requests!
-<finish></finish>
+The server is running on port 5000 with PID 126. You can access the list of numbers in a table format by visiting http://127.0.0.1:5000. Free free to let me know if you have any further requests!
 
 --- END OF EXAMPLE ---
 """
 
 INVALID_INPUT_MESSAGE = (
     "I don't understand your input. \n"
-    'For bash commands, use <execute_bash> YOUR_COMMAND </execute_bash>.\n'
-    'For Python code, use <execute_ipython> YOUR_CODE </execute_ipython>.\n'
-    'For browsing, use <execute_browse> YOUR_COMMAND </execute_browse>.\n'
+    'If you want to execute a bash command, please use <execute_bash> YOUR_COMMAND_HERE </execute_bash>.\n'
+    'If you want to execute a block of Python code, please use <execute_ipython> YOUR_COMMAND_HERE </execute_ipython>.\n'
+    'If you want to browse the Internet, please use <execute_browse> YOUR_COMMAND_HERE </execute_browse>.\n'
 )

agenthub/codeact_swe_agent/README.md

@@ -1,7 +0,0 @@
-# CodeAct (SWE Edit Specialized)
-
-This agent is an adaptation of the original [SWE Agent](https://swe-agent.com/) based on CodeAct using the `agentskills` library of OpenDevin.
-
-Its intended use is **solving GitHub issues**.
-
-It removes web-browsing and GitHub capability from the original CodeAct agent to avoid confusion to the agent.

agenthub/codeact_swe_agent/__init__.py

@@ -1,5 +0,0 @@
-from opendevin.controller.agent import Agent
-
-from .codeact_swe_agent import CodeActSWEAgent
-
-Agent.register('CodeActSWEAgent', CodeActSWEAgent)

agenthub/codeact_swe_agent/action_parser.py

@@ -1,110 +0,0 @@
-import re
-
-from opendevin.controller.action_parser import ActionParser
-from opendevin.events.action import (
-    Action,
-    AgentFinishAction,
-    CmdRunAction,
-    IPythonRunCellAction,
-    MessageAction,
-)
-
-
-class CodeActSWEActionParserFinish(ActionParser):
-    """Parser action:
-    - AgentFinishAction() - end the interaction
-    """
-
-    def __init__(
-        self,
-    ):
-        self.finish_command = None
-
-    def check_condition(self, action_str: str) -> bool:
-        self.finish_command = re.search(r'<finish>.*</finish>', action_str, re.DOTALL)
-        return self.finish_command is not None
-
-    def parse(self, action_str: str) -> Action:
-        assert (
-            self.finish_command is not None
-        ), 'self.finish_command should not be None when parse is called'
-        thought = action_str.replace(self.finish_command.group(0), '').strip()
-        return AgentFinishAction(thought=thought)
-
-
-class CodeActSWEActionParserCmdRun(ActionParser):
-    """Parser action:
-    - CmdRunAction(command) - bash command to run
-    - AgentFinishAction() - end the interaction
-    """
-
-    def __init__(
-        self,
-    ):
-        self.bash_command = None
-
-    def check_condition(self, action_str: str) -> bool:
-        self.bash_command = re.search(
-            r'<execute_bash>(.*?)</execute_bash>', action_str, re.DOTALL
-        )
-        return self.bash_command is not None
-
-    def parse(self, action_str: str) -> Action:
-        assert (
-            self.bash_command is not None
-        ), 'self.bash_command should not be None when parse is called'
-        thought = action_str.replace(self.bash_command.group(0), '').strip()
-        # a command was found
-        command_group = self.bash_command.group(1).strip()
-        if command_group.strip() == 'exit':
-            return AgentFinishAction()
-        return CmdRunAction(command=command_group, thought=thought)
-
-
-class CodeActSWEActionParserIPythonRunCell(ActionParser):
-    """Parser action:
-    - IPythonRunCellAction(code) - IPython code to run
-    """
-
-    def __init__(
-        self,
-    ):
-        self.python_code = None
-        self.jupyter_kernel_init_code: str = 'from agentskills import *'
-
-    def check_condition(self, action_str: str) -> bool:
-        self.python_code = re.search(
-            r'<execute_ipython>(.*?)</execute_ipython>', action_str, re.DOTALL
-        )
-        return self.python_code is not None
-
-    def parse(self, action_str: str) -> Action:
-        assert (
-            self.python_code is not None
-        ), 'self.python_code should not be None when parse is called'
-        code_group = self.python_code.group(1).strip()
-        thought = action_str.replace(self.python_code.group(0), '').strip()
-        return IPythonRunCellAction(
-            code=code_group,
-            thought=thought,
-            kernel_init_code=self.jupyter_kernel_init_code,
-        )
-
-
-class CodeActSWEActionParserMessage(ActionParser):
-    """Parser action:
-    - MessageAction(content) - Message action to run (e.g. ask for clarification)
-    """
-
-    def __init__(
-        self,
-    ):
-        pass
-
-    def check_condition(self, action_str: str) -> bool:
-        # We assume the LLM is GOOD enough that when it returns pure natural language
-        # it wants to talk to the user
-        return True
-
-    def parse(self, action_str: str) -> Action:
-        return MessageAction(content=action_str, wait_for_response=True)

agenthub/codeact_swe_agent/codeact_swe_agent.py

@@ -1,190 +0,0 @@
-from agenthub.codeact_swe_agent.prompt import (
-    COMMAND_DOCS,
-    SWE_EXAMPLE,
-    SYSTEM_PREFIX,
-    SYSTEM_SUFFIX,
-)
-from agenthub.codeact_swe_agent.response_parser import CodeActSWEResponseParser
-from opendevin.controller.agent import Agent
-from opendevin.controller.state.state import State
-from opendevin.events.action import (
-    Action,
-    AgentFinishAction,
-    CmdRunAction,
-    IPythonRunCellAction,
-    MessageAction,
-)
-from opendevin.events.observation import (
-    CmdOutputObservation,
-    IPythonRunCellObservation,
-)
-from opendevin.events.observation.observation import Observation
-from opendevin.events.serialization.event import truncate_content
-from opendevin.llm.llm import LLM
-from opendevin.runtime.plugins import (
-    AgentSkillsRequirement,
-    JupyterRequirement,
-    PluginRequirement,
-)
-from opendevin.runtime.tools import RuntimeTool
-
-
-def get_system_message() -> str:
-    return f'{SYSTEM_PREFIX}\n\n{COMMAND_DOCS}\n\n{SYSTEM_SUFFIX}'
-
-
-def get_in_context_example() -> str:
-    return SWE_EXAMPLE
-
-
-class CodeActSWEAgent(Agent):
-    VERSION = '1.6'
-    """
-    This agent is an adaptation of the original [SWE Agent](https://swe-agent.com/) based on CodeAct 1.5 using the `agentskills` library of OpenDevin.
-
-    It is intended use is **solving Github issues**.
-
-    It removes web-browsing and Github capability from the original CodeAct agent to avoid confusion to the agent.
-    """
-
-    sandbox_plugins: list[PluginRequirement] = [
-        # NOTE: AgentSkillsRequirement need to go before JupyterRequirement, since
-        # AgentSkillsRequirement provides a lot of Python functions,
-        # and it needs to be initialized before Jupyter for Jupyter to use those functions.
-        AgentSkillsRequirement(),
-        JupyterRequirement(),
-    ]
-    runtime_tools: list[RuntimeTool] = []
-
-    system_message: str = get_system_message()
-    in_context_example: str = f"Here is an example of how you can interact with the environment for task solving:\n{get_in_context_example()}\n\nNOW, LET'S START!"
-
-    response_parser = CodeActSWEResponseParser()
-
-    def __init__(
-        self,
-        llm: LLM,
-    ) -> None:
-        """Initializes a new instance of the CodeActAgent class.
-
-        Parameters:
-        - llm (LLM): The llm to be used by this agent
-        """
-        super().__init__(llm)
-        self.reset()
-
-    def action_to_str(self, action: Action) -> str:
-        if isinstance(action, CmdRunAction):
-            return (
-                f'{action.thought}\n<execute_bash>\n{action.command}\n</execute_bash>'
-            )
-        elif isinstance(action, IPythonRunCellAction):
-            return f'{action.thought}\n<execute_ipython>\n{action.code}\n</execute_ipython>'
-        elif isinstance(action, MessageAction):
-            return action.content
-        return ''
-
-    def get_action_message(self, action: Action) -> dict[str, str] | None:
-        if (
-            isinstance(action, CmdRunAction)
-            or isinstance(action, IPythonRunCellAction)
-            or isinstance(action, MessageAction)
-        ):
-            return {
-                'role': 'user' if action.source == 'user' else 'assistant',
-                'content': self.action_to_str(action),
-            }
-        return None
-
-    def get_observation_message(self, obs: Observation) -> dict[str, str] | None:
-        max_message_chars = self.llm.config.max_message_chars
-        if isinstance(obs, CmdOutputObservation):
-            content = 'OBSERVATION:\n' + truncate_content(
-                obs.content, max_message_chars
-            )
-            content += (
-                f'\n[Command {obs.command_id} finished with exit code {obs.exit_code}]'
-            )
-            return {'role': 'user', 'content': content}
-        elif isinstance(obs, IPythonRunCellObservation):
-            content = 'OBSERVATION:\n' + obs.content
-            # replace base64 images with a placeholder
-            splitted = content.split('\n')
-            for i, line in enumerate(splitted):
-                if '![image](data:image/png;base64,' in line:
-                    splitted[i] = (
-                        '![image](data:image/png;base64, ...) already displayed to user'
-                    )
-            content = '\n'.join(splitted)
-            content = truncate_content(content, max_message_chars)
-            return {'role': 'user', 'content': content}
-        return None
-
-    def reset(self) -> None:
-        """Resets the CodeAct Agent."""
-        super().reset()
-
-    def step(self, state: State) -> Action:
-        """Performs one step using the CodeAct Agent.
-        This includes gathering info on previous steps and prompting the model to make a command to execute.
-
-        Parameters:
-        - state (State): used to get updated info and background commands
-
-        Returns:
-        - CmdRunAction(command) - bash command to run
-        - IPythonRunCellAction(code) - IPython code to run
-        - MessageAction(content) - Message action to run (e.g. ask for clarification)
-        - AgentFinishAction() - end the interaction
-        """
-        # if we're done, go back
-        latest_user_message = state.history.get_last_user_message()
-        if latest_user_message and latest_user_message.strip() == '/exit':
-            return AgentFinishAction()
-
-        # prepare what we want to send to the LLM
-        messages: list[dict[str, str]] = self._get_messages(state)
-
-        response = self.llm.completion(
-            messages=messages,
-            stop=[
-                '</execute_ipython>',
-                '</execute_bash>',
-            ],
-            temperature=0.0,
-        )
-
-        return self.response_parser.parse(response)
-
-    def _get_messages(self, state: State) -> list[dict[str, str]]:
-        messages = [
-            {'role': 'system', 'content': self.system_message},
-            {'role': 'user', 'content': self.in_context_example},
-        ]
-
-        for event in state.history.get_events():
-            # create a regular message from an event
-            if isinstance(event, Action):
-                message = self.get_action_message(event)
-            elif isinstance(event, Observation):
-                message = self.get_observation_message(event)
-            else:
-                raise ValueError(f'Unknown event type: {type(event)}')
-
-            # add regular message
-            if message:
-                messages.append(message)
-
-        # the latest user message is important:
-        # we want to remind the agent of the environment constraints
-        latest_user_message = next(
-            (m for m in reversed(messages) if m['role'] == 'user'), None
-        )
-
-        # add a reminder to the prompt
-        if latest_user_message:
-            latest_user_message['content'] += (
-                f'\n\nENVIRONMENT REMINDER: You have {state.max_iterations - state.iteration} turns left to complete the task.'
-            )
-
-        return messages

agenthub/codeact_swe_agent/prompt.py

@@ -1,455 +0,0 @@
-from opendevin.runtime.plugins import AgentSkillsRequirement
-
-_AGENT_SKILLS_DOCS = AgentSkillsRequirement.documentation
-
-COMMAND_DOCS = (
-    '\nApart from the standard Python library, the assistant can also use the following functions (already imported) in <execute_ipython> environment:\n'
-    f'{_AGENT_SKILLS_DOCS}'
-    "Please note that THE `edit_file` FUNCTION REQUIRES PROPER INDENTATION. If the assistant would like to add the line '        print(x)', it must fully write that out, with all those spaces before the code! Indentation is important and code that is not indented correctly will fail and require fixing before it can be run."
-)
-
-# ======= SYSTEM MESSAGE =======
-MINIMAL_SYSTEM_PREFIX = """A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
-The assistant can interact with an interactive Python (Jupyter Notebook) environment and receive the corresponding output when needed. The code should be enclosed using "<execute_ipython>" tag, for example:
-<execute_ipython>
-print("Hello World!")
-</execute_ipython>
-The assistant can execute bash commands on behalf of the user by wrapping them with <execute_bash> and </execute_bash>.
-For example, you can list the files in the current directory by <execute_bash> ls </execute_bash>.
-"""
-
-PIP_INSTALL_PREFIX = """The assistant can install Python packages using the %pip magic command in an IPython environment by using the following syntax: <execute_ipython> %pip install [package needed] </execute_ipython> and should always import packages and define variables before starting to use them."""
-
-SYSTEM_PREFIX = MINIMAL_SYSTEM_PREFIX + PIP_INSTALL_PREFIX
-
-SYSTEM_SUFFIX = """The assistant's response should be concise.
-The assistant should include ONLY ONE <execute_ipython> or <execute_bash> in every one of the responses, unless the assistant is finished with the task or need more input or action from the user in order to proceed.
-IMPORTANT: Whenever possible, execute the code for the user using <execute_ipython> or <execute_bash> instead of providing it.
-"""
-
-SWE_EXAMPLE = """
---- START OF EXAMPLE ---
-
-USER: We are currently solving the following issue within our repository. Here is the issue text:
---- BEGIN ISSUE ---
-TimeDelta serialization precision
-Hi there!
-
-I just found quite strange behaviour of `TimeDelta` field serialization
-
-```python3
-from marshmallow.fields import TimeDelta
-from datetime import timedelta
-
-td_field = TimeDelta(precision="milliseconds")
-
-obj = dict()
-obj["td_field"] = timedelta(milliseconds=345)
-
-print(td_field.serialize("td_field", obj))
-```
-
-Output of this snippet is `344`, but it seems that `345` is correct.
-
-Looks like a rounding issue here: https://github.com/marshmallow-code/marshmallow/blob/dev/src/marshmallow/fields.py#L1474
---- END ISSUE ---
-
-[Current directory: /workspace/marshmallow-code__marshmallow]
-
-ASSISTANT:
-Let's first start by reproducing the results of the issue. The issue includes some example code for reproduction, which we can use. We'll create a new file called `reproduce.py` and paste the example code into it.
-<execute_ipython>
-create_file('/workspace/marshmallow-code__marshmallow/reproduce.py')
-</execute_ipython>
-
-USER:
-OBSERVATION:
-[File: /workspace/marshmallow-code__marshmallow/reproduce.py (1 lines total)]
-1|
-[File /workspace/marshmallow-code__marshmallow/reproduce.py created.]
-
-ASSISTANT:
-Now let's paste in the example code from the issue.
-<execute_ipython>
-EDITED_CODE = \"\"\"from marshmallow.fields import TimeDelta
-from datetime import timedelta
-
-td_field = TimeDelta(precision="milliseconds")
-
-obj = dict()
-obj["td_field"] = timedelta(milliseconds=345)
-
-print(td_field.serialize("td_field", obj))
-\"\"\"
-edit_file(start=1, end=1, content=EDITED_CODE)
-</execute_ipython>
-
-USER:
-OBSERVATION:
-[File: /workspace/marshmallow-code__marshmallow/reproduce.py (9 lines total)]
-1|from marshmallow.fields import TimeDelta
-2|from datetime import timedelta
-3|
-4|td_field = TimeDelta(precision="milliseconds")
-5|
-6|obj = dict()
-7|obj["td_field"] = timedelta(milliseconds=345)
-8|
-9|print(td_field.serialize("td_field", obj))
-[File updated. Please review the changes and make sure they are correct (correct indentation, no duplicate lines, etc). Edit the file again if necessary.]
-
-ASSISTANT:
-Now let's run the code to see if we see the same output as the issue.
-<execute_bash>
-python reproduce.py
-</execute_bash>
-
-USER:
-OBSERVATION:
-344
-
-ASSISTANT:
-We are indeed seeing the same output as the issue. The issue suggests that we should look at line 1474 of the `fields.py` file to see if there is a rounding issue. Let's open the `fields.py` file and navigate to line 1474 to see the relevant code. First we should find the `fields.py` file in the repository. It is likely to be in the `src/marshmallow` directory, but we should check to be sure.
-<execute_bash>
-ls -F
-</execute_bash>
-
-USER:
-OBSERVATION:
-AUTHORS.rst
-CHANGELOG.rst
-CODE_OF_CONDUCT.md
-CONTRIBUTING.rst
-LICENSE
-MANIFEST.in
-NOTICE
-README.rst
-RELEASING.md
-azure-pipelines.yml
-docs/
-examples/
-performance/
-pyproject.toml
-reproduce.py
-setup.cfg
-setup.py
-src/
-tests/
-tox.ini
-
-ASSISTANT:
-It looks like the `src` directory is present, which suggests that the `fields.py` file is likely to be in the `src` directory. Let's use find_file to see where it is.
-<execute_ipython>
-find_file("fields.py", "/workspace/marshmallow-code__marshmallow/src")
-</execute_ipython>
-
-USER:
-OBSERVATION:
-[Found 1 matches for "fields.py" in /workspace/marshmallow-code__marshmallow/src]
-/workspace/marshmallow-code__marshmallow/src/marshmallow/fields.py
-
-ASSISTANT:
-It looks like the `fields.py` file is present in the `./src/marshmallow/` directory. The issue also points to a specific URL with line number 1474. We should navigate to that line in fields.py to see the relevant code for the `TimeDelta` serialization.
-<execute_ipython>
-open_file("/workspace/marshmallow-code__marshmallow/src/marshmallow/fields.py", 1474)
-</execute_ipython>
-
-USER:
-OBSERVATION:
-[File: /workspace/marshmallow-code__marshmallow/src/marshmallow/fields.py (1997 lines total)]
-(1458 more lines above)
-1459|            self.WEEKS,
-1460|        )
-1461|
-1462|        if precision not in units:
-1463|            msg = 'The precision must be {} or "{}".'.format(
-1464|                ", ".join([f'"{each}"' for each in units[:-1]]), units[-1]
-1465|            )
-1466|            raise ValueError(msg)
-1467|
-1468|        self.precision = precision
-1469|        super().__init__(**kwargs)
-1470|
-1471|    def _serialize(self, value, attr, obj, **kwargs):
-1472|        if value is None:
-1473|            return None
-1474|        base_unit = dt.timedelta(**{self.precision: 1})
-1475|        return int(value.total_seconds() / base_unit.total_seconds())
-1476|
-1477|    def _deserialize(self, value, attr, data, **kwargs):
-1478|        try:
-1479|            value = int(value)
-1480|        except (TypeError, ValueError) as error:
-1481|            raise self.make_error("invalid") from error
-1482|
-1483|        kwargs = {self.precision: value}
-1484|
-1485|        try:
-1486|            return dt.timedelta(**kwargs)
-1487|        except OverflowError as error:
-1488|            raise self.make_error("invalid") from error
-1489|
-1490|
-1491|class Mapping(Field):
-1492|    \"\"\"An abstract class for objects with key-value pairs.
-1493|
-1494|    :param keys: A field class or instance for dict keys.
-1495|    :param values: A field class or instance for dict values.
-1496|    :param kwargs: The same keyword arguments that :class:`Field` receives.
-1497|
-1498|    .. note::
-1499|        When the structure of nested data is not known, you may omit the
-1500|        `keys` and `values` arguments to prevent content validation.
-1501|
-1502|    .. versionadded:: 3.0.0rc4
-1503|    \"\"\"
-1504|
-1505|    mapping_type = dict
-1506|
-1507|    #: Default error messages.
-1508|    default_error_messages = {"invalid": "Not a valid mapping type."}
-1509|
-1510|    def __init__(
-1511|        self,
-1512|        keys: typing.Optional[typing.Union[Field, type]] = None,
-1513|        values: typing.Optional[typing.Union[Field, type]] = None,
-1514|        **kwargs,
-1515|    ):
-1516|        super().__init__(**kwargs)
-1517|        if keys is None:
-1518|            self.key_field = None
-1519|        else:
-1520|            try:
-1521|                self.key_field = resolve_field_instance(keys)
-1522|            except FieldInstanceResolutionError as error:
-1523|                raise ValueError(
-1524|                    '"keys" must be a subclass or instance of '
-1525|                    "marshmallow.base.FieldABC."
-1526|                ) from error
-1527|
-1528|        if values is None:
-1529|            self.value_field = None
-1530|        else:
-1531|            try:
-1532|                self.value_field = resolve_field_instance(values)
-1533|            except FieldInstanceResolutionError as error:
-1534|                raise ValueError(
-1535|                    '"values" must be a subclass or instance of '
-1536|                    "marshmallow.base.FieldABC."
-1537|                ) from error
-1538|            if isinstance(self.value_field, Nested):
-1539|                self.only = self.value_field.only
-1540|                self.exclude = self.value_field.exclude
-1541|
-1542|    def _bind_to_schema(self, field_name, schema):
-1543|        super()._bind_to_schema(field_name, schema)
-1544|        if self.value_field:
-1545|            self.value_field = copy.deepcopy(self.value_field)
-1546|            self.value_field._bind_to_schema(field_name, self)
-1547|        if isinstance(self.value_field, Nested):
-1548|            self.value_field.only = self.only
-1549|            self.value_field.exclude = self.exclude
-1550|        if self.key_field:
-1551|            self.key_field = copy.deepcopy(self.key_field)
-1552|            self.key_field._bind_to_schema(field_name, self)
-1553|
-1554|    def _serialize(self, value, attr, obj, **kwargs):
-1555|        if value is None:
-1556|            return None
-1557|        if not self.value_field and not self.key_field:
-1558|            return self.mapping_type(value)
-(439 more lines below)
-
-ASSISTANT:
-We are now looking at the relevant section of the `fields.py` file where the `TimeDelta` serialization occurs. The issue suggests that there is a rounding problem near line 1474. The current implementation uses integer division, which truncates the decimal part and can lead to incorrect rounding on line 1475. To fix this issue, we should ensure that the division results in a float, and then round the result to the nearest integer before converting it to an int. This should preserve the milliseconds precision as expected. Let's make the necessary edit to the code.
-<execute_ipython>
-edit_file(start=1475, end=1475, content="return int(round(value.total_seconds() / base_unit.total_seconds()))  # round to nearest int")
-</execute_ipython>
-
-USER:
-OBSERVATION:
-[Your proposed edit has introduced new syntax error(s). Please understand the errors and retry your edit command.]
-
-ERRORS:
-- E999 IndentationError: unexpected indent
-
-[This is how your edit would have looked if applied]
--------------------------------------------------
-[File: /marshmallow-code__marshmallow/src/marshmallow/fields.py (1997 lines total)]
-(1468 more lines above)
-1469|        super().__init__(**kwargs)
-1470|
-1471|    def _serialize(self, value, attr, obj, **kwargs):
-1472|        if value is None:
-1473|            return None
-1474|        base_unit = dt.timedelta(**{self.precision: 1})
-1475|return int(round(value.total_seconds() / base_unit.total_seconds()))  # round to nearest int
-1476|
-1477|    def _deserialize(self, value, attr, data, **kwargs):
-1478|        try:
-1479|            value = int(value)
-(518 more lines below)
--------------------------------------------------
-
-[This is the original code before your edit]
--------------------------------------------------
-[File: /marshmallow-code__marshmallow/src/marshmallow/fields.py (1997 lines total)]
-(1469 more lines above)
-1470|
-1471|    def _serialize(self, value, attr, obj, **kwargs):
-1472|        if value is None:
-1473|            return None
-1474|        base_unit = dt.timedelta(**{self.precision: 1})
-1475|        return int(value.total_seconds() / base_unit.total_seconds())
-1476|
-1477|    def _deserialize(self, value, attr, data, **kwargs):
-1478|        try:
-1479|            value = int(value)
-1480|        except (TypeError, ValueError) as error:
-(517 more lines below)
--------------------------------------------------
-Your changes have NOT been applied. Please fix your edit command and try again.
-You either need to 1) Specify the correct start/end line arguments or 2) Correct your edit code.
-DO NOT re-run the same failed edit command. Running it again will lead to the same error.
-
-ASSISTANT:
-My edit command did not use the proper indentation, I will fix my syntax in this follow up edit command.
-<execute_ipython>
-EDITED_CODE="        return int(round(value.total_seconds() / base_unit.total_seconds()))  # round to nearest int"
-edit_file(start=1475, end=1475, content=EDITED_CODE)
-</execute_ipython>
-
-USER:
-OBSERVATION:
-[File: /marshmallow-code__marshmallow/src/marshmallow/fields.py (1997 lines total)]
-(1424 more lines above)
-1425|
-1426|    :param precision: Influences how the integer is interpreted during
-1427|        (de)serialization. Must be 'days', 'seconds', 'microseconds',
-1428|        'milliseconds', 'minutes', 'hours' or 'weeks'.
-1429|    :param kwargs: The same keyword arguments that :class:`Field` receives.
-1430|
-1431|    .. versionchanged:: 2.0.0
-1432|        Always serializes to an integer value to avoid rounding errors.
-1433|        Add `precision` parameter.
-1434|    \"\"\"
-1435|
-1436|    DAYS = "days"
-1437|    SECONDS = "seconds"
-1438|    MICROSECONDS = "microseconds"
-1439|    MILLISECONDS = "milliseconds"
-1440|    MINUTES = "minutes"
-1441|    HOURS = "hours"
-1442|    WEEKS = "weeks"
-1443|
-1444|    #: Default error messages.
-1445|    default_error_messages = {
-1446|        "invalid": "Not a valid period of time.",
-1447|        "format": "{input!r} cannot be formatted as a timedelta.",
-1448|    }
-1449|
-1450|    def __init__(self, precision: str = SECONDS, **kwargs):
-1451|        precision = precision.lower()
-1452|        units = (
-1453|            self.DAYS,
-1454|            self.SECONDS,
-1455|            self.MICROSECONDS,
-1456|            self.MILLISECONDS,
-1457|            self.MINUTES,
-1458|            self.HOURS,
-1459|            self.WEEKS,
-1460|        )
-1461|
-1462|        if precision not in units:
-1463|            msg = 'The precision must be {} or "{}".'.format(
-1464|                ", ".join([f'"{each}"' for each in units[:-1]]), units[-1]
-1465|            )
-1466|            raise ValueError(msg)
-1467|
-1468|        self.precision = precision
-1469|        super().__init__(**kwargs)
-1470|
-1471|    def _serialize(self, value, attr, obj, **kwargs):
-1472|        if value is None:
-1473|            return None
-1474|        base_unit = dt.timedelta(**{self.precision: 1})
-1475|        return int(round(value.total_seconds() / base_unit.total_seconds()))  # round to nearest int
-1476|
-1477|    def _deserialize(self, value, attr, data, **kwargs):
-1478|        try:
-1479|            value = int(value)
-1480|        except (TypeError, ValueError) as error:
-1481|            raise self.make_error("invalid") from error
-1482|
-1483|        kwargs = {self.precision: value}
-1484|
-1485|        try:
-1486|            return dt.timedelta(**kwargs)
-1487|        except OverflowError as error:
-1488|            raise self.make_error("invalid") from error
-1489|
-1490|
-1491|class Mapping(Field):
-1492|    \"\"\"An abstract class for objects with key-value pairs.
-1493|
-1494|    :param keys: A field class or instance for dict keys.
-1495|    :param values: A field class or instance for dict values.
-1496|    :param kwargs: The same keyword arguments that :class:`Field` receives.
-1497|
-1498|    .. note::
-1499|        When the structure of nested data is not known, you may omit the
-1500|        `keys` and `values` arguments to prevent content validation.
-1501|
-1502|    .. versionadded:: 3.0.0rc4
-1503|    \"\"\"
-1504|
-1505|    mapping_type = dict
-1506|
-1507|    #: Default error messages.
-1508|    default_error_messages = {"invalid": "Not a valid mapping type."}
-1509|
-1510|    def __init__(
-1511|        self,
-1512|        keys: typing.Optional[typing.Union[Field, type]] = None,
-1513|        values: typing.Optional[typing.Union[Field, type]] = None,
-1514|        **kwargs,
-1515|    ):
-1516|        super().__init__(**kwargs)
-1517|        if keys is None:
-1518|            self.key_field = None
-1519|        else:
-1520|            try:
-1521|                self.key_field = resolve_field_instance(keys)
-1522|            except FieldInstanceResolutionError as error:
-1523|                raise ValueError(
-1524|                    '"keys" must be a subclass or instance of '
-(473 more lines below)
-[File updated. Please review the changes and make sure they are correct (correct indentation, no duplicate lines, etc). Edit the file again if necessary.]
-
-ASSISTANT:
-The code has been updated to use the `round` function, which should fix the rounding issue. Before submitting the changes, it would be prudent to run the reproduce.py code again to ensure that this change has the desired effect. I see that "Current directory" is still `/marshmallow-code__marshmallow`, so we can just run `reproduce.py` directly.
-<execute_bash>
-python reproduce.py
-</execute_bash>
-
-USER:
-OBSERVATION:
-345
-
-ASSISTANT:
-The output has changed from 344 to 345, which suggests that the rounding issue has been fixed. Let's remove the reproduce.py file since it is no longer needed.
-<execute_bash>
-rm reproduce.py
-</execute_bash>
-
-USER:
-OBSERVATION:
-
-ASSISTANT:
-rm doesn't have any output when it deletes successfully, so that must have worked. Now that we have fixed the issue, we can exit the current task.
-<execute_bash>
-exit
-</execute_bash>
-
---- END OF EXAMPLE ---
-"""

agenthub/codeact_swe_agent/response_parser.py

@@ -1,46 +0,0 @@
-from agenthub.codeact_swe_agent.action_parser import (
-    CodeActSWEActionParserCmdRun,
-    CodeActSWEActionParserFinish,
-    CodeActSWEActionParserIPythonRunCell,
-    CodeActSWEActionParserMessage,
-)
-from opendevin.controller.action_parser import ResponseParser
-from opendevin.events.action import Action
-
-
-class CodeActSWEResponseParser(ResponseParser):
-    """Parser action:
-    - CmdRunAction(command) - bash command to run
-    - IPythonRunCellAction(code) - IPython code to run
-    - MessageAction(content) - Message action to run (e.g. ask for clarification)
-    - AgentFinishAction() - end the interaction
-    """
-
-    def __init__(self):
-        # Need pay attention to the item order in self.action_parsers
-        super().__init__()
-        self.action_parsers = [
-            CodeActSWEActionParserFinish(),
-            CodeActSWEActionParserCmdRun(),
-            CodeActSWEActionParserIPythonRunCell(),
-        ]
-        self.default_parser = CodeActSWEActionParserMessage()
-
-    def parse(self, response: str) -> Action:
-        action_str = self.parse_response(response)
-        return self.parse_action(action_str)
-
-    def parse_response(self, response) -> str:
-        action = response.choices[0].message.content
-        if action is None:
-            return ''
-        for lang in ['bash', 'ipython']:
-            if f'<execute_{lang}>' in action and f'</execute_{lang}>' not in action:
-                action += f'</execute_{lang}>'
-        return action
-
-    def parse_action(self, action_str: str) -> Action:
-        for action_parser in self.action_parsers:
-            if action_parser.check_condition(action_str):
-                return action_parser.parse(action_str)
-        return self.default_parser.parse(action_str)

agenthub/delegator_agent/agent.py

@@ -6,15 +6,16 @@ from opendevin.llm.llm import LLM
 
 
 class DelegatorAgent(Agent):
-    VERSION = '1.0'
     """
-    The Delegator Agent is responsible for delegating tasks to other agents based on the current task.
+    The planner agent utilizes a special prompting strategy to create long term plans for solving problems.
+    The agent is given its previous action-observation pairs, current task, and hint based on last action taken at every step.
     """
 
     current_delegate: str = ''
 
     def __init__(self, llm: LLM):
-        """Initialize the Delegator Agent with an LLM
+        """
+        Initialize the Delegator Agent with an LLM
 
         Parameters:
         - llm (LLM): The llm to be used by this agent
@@ -22,15 +23,16 @@ class DelegatorAgent(Agent):
         super().__init__(llm)
 
     def step(self, state: State) -> Action:
-        """Checks to see if current step is completed, returns AgentFinishAction if True.
-        Otherwise, delegates the task to the next agent in the pipeline.
+        """
+        Checks to see if current step is completed, returns AgentFinishAction if True.
+        Otherwise, creates a plan prompt and sends to model for inference, returning the result as the next action.
 
         Parameters:
         - state (State): The current state given the previous actions and observations
 
         Returns:
         - AgentFinishAction: If the last state was 'completed', 'verified', or 'abandoned'
-        - AgentDelegateAction: The next agent to delegate the task to
+        - Action: The next action to take based on llm response
         """
         if self.current_delegate == '':
             self.current_delegate = 'study'
@@ -39,9 +41,7 @@ class DelegatorAgent(Agent):
                 agent='StudyRepoForTaskAgent', inputs={'task': task}
             )
 
-        # last observation in history should be from the delegate
-        last_observation = state.history.get_last_observation()
-
+        last_observation = state.history[-1][1]
         if not isinstance(last_observation, AgentDelegateObservation):
             raise Exception('Last observation is not an AgentDelegateObservation')
 
@@ -80,3 +80,6 @@ class DelegatorAgent(Agent):
                 )
         else:
             raise Exception('Invalid delegate state')
+
+    def search_memory(self, query: str) -> list[str]:
+        return []

agenthub/dummy_agent/agent.py

@@ -1,14 +1,14 @@
-from typing import TypedDict, Union
+import time
+from typing import TypedDict
 
 from opendevin.controller.agent import Agent
 from opendevin.controller.state.state import State
-from opendevin.core.schema import AgentState
 from opendevin.events.action import (
     Action,
     AddTaskAction,
     AgentFinishAction,
+    AgentRecallAction,
     AgentRejectAction,
-    BrowseInteractiveAction,
     BrowseURLAction,
     CmdRunAction,
     FileReadAction,
@@ -17,7 +17,7 @@ from opendevin.events.action import (
     ModifyTaskAction,
 )
 from opendevin.events.observation import (
-    AgentStateChangedObservation,
+    AgentRecallObservation,
     CmdOutputObservation,
     FileReadObservation,
     FileWriteObservation,
@@ -30,6 +30,8 @@ from opendevin.llm.llm import LLM
 """
 FIXME: There are a few problems this surfaced
 * FileWrites seem to add an unintended newline at the end of the file
+* command_id is sometimes a number, sometimes a string
+* Why isn't the output of the background command split between two steps?
 * Browser not working
 """
 
@@ -37,9 +39,10 @@ ActionObs = TypedDict(
     'ActionObs', {'action': Action, 'observations': list[Observation]}
 )
 
+BACKGROUND_CMD = 'echo "This is in the background" && sleep .1 && echo "This too"'
+
 
 class DummyAgent(Agent):
-    VERSION = '1.0'
     """
     The DummyAgent is used for e2e testing. It just sends the same set of actions deterministically,
     without making any LLM calls.
@@ -49,40 +52,32 @@ class DummyAgent(Agent):
         super().__init__(llm)
         self.steps: list[ActionObs] = [
             {
-                'action': AddTaskAction(
-                    parent='None', goal='check the current directory'
-                ),
-                'observations': [],
+                'action': AddTaskAction(parent='0', goal='check the current directory'),
+                'observations': [NullObservation('')],
             },
             {
-                'action': AddTaskAction(parent='0', goal='run ls'),
-                'observations': [],
+                'action': AddTaskAction(parent='0.0', goal='run ls'),
+                'observations': [NullObservation('')],
             },
             {
-                'action': ModifyTaskAction(task_id='0', state='in_progress'),
-                'observations': [],
+                'action': ModifyTaskAction(task_id='0.0', state='in_progress'),
+                'observations': [NullObservation('')],
             },
             {
                 'action': MessageAction('Time to get started!'),
-                'observations': [],
+                'observations': [NullObservation('')],
             },
             {
                 'action': CmdRunAction(command='echo "foo"'),
                 'observations': [
-                    CmdOutputObservation(
-                        'foo', command_id=-1, command='echo "foo"', exit_code=0
-                    )
+                    CmdOutputObservation('foo', command_id=-1, command='echo "foo"')
                 ],
             },
             {
                 'action': FileWriteAction(
                     content='echo "Hello, World!"', path='hello.sh'
                 ),
-                'observations': [
-                    FileWriteObservation(
-                        content='echo "Hello, World!"', path='hello.sh'
-                    )
-                ],
+                'observations': [FileWriteObservation('', path='hello.sh')],
             },
             {
                 'action': FileReadAction(path='hello.sh'),
@@ -94,123 +89,77 @@ class DummyAgent(Agent):
                 'action': CmdRunAction(command='bash hello.sh'),
                 'observations': [
                     CmdOutputObservation(
-                        'bash: hello.sh: No such file or directory',
-                        command_id=-1,
-                        command='bash workspace/hello.sh',
-                        exit_code=127,
+                        'Hello, World!', command_id=-1, command='bash hello.sh'
                     )
                 ],
             },
+            {
+                'action': CmdRunAction(command=BACKGROUND_CMD, background=True),
+                'observations': [
+                    CmdOutputObservation(
+                        'Background command started. To stop it, send a `kill` action with command_id 42',
+                        command_id='42',  # type: ignore[arg-type]
+                        command=BACKGROUND_CMD,
+                    ),
+                    CmdOutputObservation(
+                        'This is in the background\nThis too\n',
+                        command_id='42',  # type: ignore[arg-type]
+                        command=BACKGROUND_CMD,
+                    ),
+                ],
+            },
+            {
+                'action': AgentRecallAction(query='who am I?'),
+                'observations': [
+                    AgentRecallObservation('', memories=['I am a computer.']),
+                    # CmdOutputObservation('This too\n', command_id='42', command=BACKGROUND_CMD),
+                ],
+            },
             {
                 'action': BrowseURLAction(url='https://google.com'),
                 'observations': [
-                    # BrowserOutputObservation('<html><body>Simulated Google page</body></html>',url='https://google.com',screenshot=''),
+                    # BrowserOutputObservation('<html></html>', url='https://google.com', screenshot=""),
                 ],
             },
             {
-                'action': BrowseInteractiveAction(
-                    browser_actions='goto("https://google.com")'
-                ),
-                'observations': [
-                    # BrowserOutputObservation('<html><body>Simulated Google page after interaction</body></html>',url='https://google.com',screenshot=''),
-                ],
+                'action': AgentFinishAction(),
+                'observations': [],
             },
             {
                 'action': AgentRejectAction(),
-                'observations': [NullObservation('')],
-            },
-            {
-                'action': AgentFinishAction(
-                    outputs={}, thought='Task completed', action='finish'
-                ),
-                'observations': [AgentStateChangedObservation('', AgentState.FINISHED)],
+                'observations': [],
             },
         ]
 
     def step(self, state: State) -> Action:
-        if state.iteration >= len(self.steps):
-            return AgentFinishAction()
-
-        current_step = self.steps[state.iteration]
-        action = current_step['action']
-
-        # If the action is AddTaskAction or ModifyTaskAction, update the parent ID or task_id
-        if isinstance(action, AddTaskAction):
-            if action.parent == 'None':
-                action.parent = ''  # Root task has no parent
-            elif action.parent == '0':
-                action.parent = state.root_task.id
-            elif action.parent.startswith('0.'):
-                action.parent = f'{state.root_task.id}{action.parent[1:]}'
-        elif isinstance(action, ModifyTaskAction):
-            if action.task_id == '0':
-                action.task_id = state.root_task.id
-            elif action.task_id.startswith('0.'):
-                action.task_id = f'{state.root_task.id}{action.task_id[1:]}'
-            # Ensure the task_id doesn't start with a dot
-            if action.task_id.startswith('.'):
-                action.task_id = action.task_id[1:]
-        elif isinstance(action, (BrowseURLAction, BrowseInteractiveAction)):
-            try:
-                return self.simulate_browser_action(action)
-            except (
-                Exception
-            ):  # This could be a specific exception for browser unavailability
-                return self.handle_browser_unavailable(action)
-
+        time.sleep(0.1)
         if state.iteration > 0:
             prev_step = self.steps[state.iteration - 1]
-
-            if 'observations' in prev_step and prev_step['observations']:
+            if 'observations' in prev_step:
                 expected_observations = prev_step['observations']
-                hist_events = state.history.get_last_events(len(expected_observations))
-
-                if len(hist_events) < len(expected_observations):
-                    print(
-                        f'Warning: Expected {len(expected_observations)} observations, but got {len(hist_events)}'
-                    )
-
-                for i in range(min(len(expected_observations), len(hist_events))):
-                    hist_obs = event_to_dict(hist_events[i])
+                hist_start = len(state.history) - len(expected_observations)
+                for i in range(len(expected_observations)):
+                    hist_obs = event_to_dict(state.history[hist_start + i][1])
                     expected_obs = event_to_dict(expected_observations[i])
-
-                    # Remove dynamic fields for comparison
-                    for obs in [hist_obs, expected_obs]:
-                        obs.pop('id', None)
-                        obs.pop('timestamp', None)
-                        obs.pop('cause', None)
-                        obs.pop('source', None)
-                        if 'extras' in obs:
-                            obs['extras'].pop('command_id', None)
-
+                    if (
+                        'command_id' in hist_obs['extras']
+                        and hist_obs['extras']['command_id'] != -1
+                    ):
+                        del hist_obs['extras']['command_id']
+                        hist_obs['content'] = ''
+                    if (
+                        'command_id' in expected_obs['extras']
+                        and expected_obs['extras']['command_id'] != -1
+                    ):
+                        del expected_obs['extras']['command_id']
+                        expected_obs['content'] = ''
                     if hist_obs != expected_obs:
-                        print(
-                            f'Warning: Observation mismatch. Expected {expected_obs}, got {hist_obs}'
-                        )
+                        print('\nactual', hist_obs)
+                        print('\nexpect', expected_obs)
+                    assert (
+                        hist_obs == expected_obs
+                    ), f'Expected observation {expected_obs}, got {hist_obs}'
+        return self.steps[state.iteration]['action']
 
-        return action
-
-    def simulate_browser_action(
-        self, action: Union[BrowseURLAction, BrowseInteractiveAction]
-    ) -> Action:
-        # Instead of simulating, we'll reject the browser action
-        return self.handle_browser_unavailable(action)
-
-    def handle_browser_unavailable(
-        self, action: Union[BrowseURLAction, BrowseInteractiveAction]
-    ) -> Action:
-        # Create a message action to inform that browsing is not available
-        message = 'Browser actions are not available in the DummyAgent environment.'
-        if isinstance(action, BrowseURLAction):
-            message += f' Unable to browse URL: {action.url}'
-        elif isinstance(action, BrowseInteractiveAction):
-            message += (
-                f' Unable to perform interactive browsing: {action.browser_actions}'
-            )
-        return MessageAction(content=message)
-
-    async def get_working_directory(self, state: State) -> str:
-        # Implement this method to return the current working directory
-        # This might involve accessing state information or making an async call
-        # For now, we'll return a placeholder value
-        return './workspace'
+    def search_memory(self, query: str) -> list[str]:
+        return ['I am a computer.']

agenthub/micro/README.md

@@ -12,6 +12,3 @@ in the following structure:
 Note that `prompt.md` could use jinja2 template syntax. During runtime, `prompt.md`
 is loaded and rendered, and used together with `agent.yaml` to initialize a
 micro-agent.
-
-Micro-agents can be used independently. You can also use `ManagerAgent` which knows
-how to coordinate the agents and collaboratively finish a task.

agenthub/micro/_instructions/actions/finish.md

@@ -1,2 +1,2 @@
-* `finish` - if you're absolutely certain that you've completed your task, use the finish action to stop working. Arguments:
+* `finish` - if you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working. Arguments:
   * `outputs` - a dictionary representing the outputs of your task, if any

agenthub/micro/_instructions/actions/kill.md

@@ -0,0 +1,2 @@
+* `kill` - kills a background command
+  * `command_id` - the ID of the background command to kill

agenthub/micro/_instructions/actions/reject.md

@@ -1,2 +1,2 @@
 * `reject` - reject the task. Arguments:
-  * `outputs` - a dictionary with only a `reason` attribute
+  * `outputs` - a dictionary representing the outputs of your task, if any

agenthub/micro/_instructions/actions/run.md

@@ -1,2 +1,3 @@
 * `run` - runs a command on the command line in a Linux shell. Arguments:
   * `command` - the command to run
+  * `background` - if true, run the command in the background, so that other commands can be run concurrently. Useful for e.g. starting a server. You won't be able to see the logs. You don't need to end the command with `&`, just set this to true.

agenthub/micro/agent.py

@@ -7,7 +7,6 @@ from opendevin.events.action import Action
 from opendevin.events.serialization.action import action_from_dict
 from opendevin.events.serialization.event import event_to_memory
 from opendevin.llm.llm import LLM
-from opendevin.memory.history import ShortTermHistory
 
 from .instructions import instructions
 from .registry import all_microagents
@@ -22,37 +21,28 @@ def parse_response(orig_response: str) -> Action:
 
 
 def to_json(obj, **kwargs):
-    """Serialize an object to str format"""
+    """
+    Serialize an object to str format
+    """
     return json.dumps(obj, **kwargs)
 
 
+def history_to_json(obj, **kwargs):
+    """
+    Serialize and simplify history to str format
+    """
+    if isinstance(obj, list):
+        # process history, make it simpler.
+        processed_history = []
+        for action, observation in obj:
+            processed_history.append((event_to_memory(action), event_to_memory(observation)))
+        return json.dumps(processed_history, **kwargs)
+
+
 class MicroAgent(Agent):
-    VERSION = '1.0'
     prompt = ''
     agent_definition: dict = {}
 
-    def history_to_json(
-        self, history: ShortTermHistory, max_events: int = 20, **kwargs
-    ):
-        """
-        Serialize and simplify history to str format
-        """
-        processed_history = []
-        event_count = 0
-
-        for event in history.get_events(reverse=True):
-            if event_count >= max_events:
-                break
-            processed_history.append(
-                event_to_memory(event, self.llm.config.max_message_chars)
-            )
-            event_count += 1
-
-        # history is in reverse order, let's fix it
-        processed_history.reverse()
-
-        return json.dumps(processed_history, **kwargs)
-
     def __init__(self, llm: LLM):
         super().__init__(llm)
         if 'name' not in self.agent_definition:
@@ -62,16 +52,21 @@ class MicroAgent(Agent):
         del self.delegates[self.agent_definition['name']]
 
     def step(self, state: State) -> Action:
+        latest_user_message = state.get_current_user_intent()
         prompt = self.prompt_template.render(
             state=state,
             instructions=instructions,
             to_json=to_json,
-            history_to_json=self.history_to_json,
+            history_to_json=history_to_json,
             delegates=self.delegates,
-            latest_user_message=state.get_current_user_intent(),
+            latest_user_message=latest_user_message,
         )
         messages = [{'content': prompt, 'role': 'user'}]
         resp = self.llm.completion(messages=messages)
         action_resp = resp['choices'][0]['message']['content']
+        state.num_of_chars += len(prompt) + len(action_resp)
         action = parse_response(action_resp)
         return action
+
+    def search_memory(self, query: str) -> list[str]:
+        return []

agenthub/micro/coder/agent.yaml

@@ -2,5 +2,5 @@ name: CoderAgent
 description: Given a particular task, and a detailed description of the codebase, accomplishes the task
 inputs:
   task: string
-  summary: string
+  codebase_summary: string
 outputs: {}

agenthub/micro/coder/prompt.md

@@ -2,7 +2,7 @@
 You are a software engineer. You've inherited an existing codebase, which you
 need to modify to complete this task:
 
-{{ state.inputs.task }}
+{{ latest_user_message }}
 
 {% if state.inputs.summary %}
 Here's a summary of the codebase, as it relates to this task:
@@ -21,7 +21,7 @@ Do NOT finish until you have completed the tasks.
 
 ## History
 {{ instructions.history_truncated }}
-{{ history_to_json(state.history, max_events=20) }}
+{{ history_to_json(state.history[-10:]) }}
 
 ## Format
 {{ instructions.format.action }}

agenthub/micro/commit_writer/README.md

@@ -3,7 +3,7 @@
 CommitWriterAgent can help write git commit message. Example:
 
 ```bash
-WORKSPACE_MOUNT_PATH="`PWD`" SANDBOX_BOX_TYPE="ssh" \
+WORKSPACE_MOUNT_PATH="`PWD`" SANDBOX_TYPE="exec" \
   poetry run python opendevin/core/main.py -t "dummy task" -c CommitWriterAgent -d ./
 ```
 

agenthub/micro/commit_writer/agent.yaml

@@ -3,4 +3,3 @@ description: "Write a git commit message for files in the git staging area"
 inputs: {}
 outputs:
   answer: string
-  reason: string

agenthub/micro/commit_writer/prompt.md

@@ -11,16 +11,14 @@ changes. The commit message should include:
 - A summary line that clearly states the purpose of the changes.
 - Optionally, a detailed description if the changes are complex or need further explanation.
 
-You should first use `git status` to check whether it's a valid git repo and there
-is diff in the staging area. If not, please call the `reject` action.
-
-If it is a valid git repo and there is diff in the staging area, you should find
-the diff using `git diff --cached`, compile a commit message, and call the `finish`
-action with `outputs.answer` set to the answer.
+You should find the diff using `git diff --cached`, compile a commit message,
+and call the `finish` action with `outputs.answer` set to the answer. If current
+repo is not a valid git repo, or there is no diff in the staging area, please call
+the `reject` action with `outputs.answer` set to the reason.
 
 ## History
 {{ instructions.history_truncated }}
-{{ history_to_json(state.history, max_events=20) }}
+{{ history_to_json(state.history[-10:]) }}
 
 If the last item in the history is an error, you should try to fix it.
 

agenthub/micro/manager/agent.yaml

@@ -3,6 +3,4 @@ description: Delegates tasks to microagents based on their area of expertise
 generates: Action
 inputs:
   task: string
-outputs:
-  summary: string # if finished
-  reason: string # if rejected
+outputs: {}

agenthub/micro/manager/prompt.md

@@ -7,16 +7,6 @@ can do the actual work. A description of each agent is provided below. You MUST
 select one of the delegates below to move towards accomplishing the task, and you MUST
 provide the correct inputs for the delegate you select.
 
-Note: the delegated agent either returns "finish" or "reject".
-- If the action is "finish", but the full task is not done yet, you should
-continue to delegate to one of the agents below to until the full task is finished.
-- If the action is "reject", it means the delegated agent is not capable of the
-task you send to. You should revisit the input you send to the delegate, and consider
-whether any other delegate would be able to solve the task. If you cannot find
-a proper delegate agent, or the delegate attempts keep failing, call the `reject`
-action. In `reason` attribute, make sure you include your attempts (e.g. what agent
-you have delegated to, and why they failed).
-
 ## Agents
 {% for name, details in delegates.items() %}
 ### {{ name }}
@@ -27,15 +17,11 @@ you have delegated to, and why they failed).
 
 ## History
 {{ instructions.history_truncated }}
-{{ history_to_json(state.history, max_events=20) }}
-
-If the last item in the history is an error, you should try to fix it. If you
-cannot fix it, call the `reject` action.
+{{ history_to_json(state.history[-10:]) }}
 
 ## Available Actions
 {{ instructions.actions.delegate }}
 {{ instructions.actions.finish }}
-{{ instructions.actions.reject }}
 
 ## Format
 {{ instructions.format.action }}

agenthub/micro/math_agent/prompt.md

@@ -1,7 +1,7 @@
 # Task
 You are a brilliant mathematician and programmer. You've been given the following problem to solve:
 
-`{{ state.inputs.task }}`
+{{ latest_user_message }}
 
 Please write a python script that solves this problem, and prints the answer to stdout.
 ONLY print the answer to stdout, nothing else.
@@ -10,7 +10,7 @@ and call the `finish` action with `outputs.answer` set to the answer.
 
 ## History
 {{ instructions.history_truncated }}
-{{ history_to_json(state.history, max_events=20) }}
+{{ history_to_json(state.history[-10:]) }}
 
 If the last item in the history is an error, you should try to fix it.
 

agenthub/micro/postgres_agent/prompt.md

@@ -2,7 +2,7 @@
 You are a database engineer. You are working on an existing Postgres project, and have been given
 the following task:
 
-{{ state.inputs.task }}
+{{ latest_user_message }}
 
 You must:
 * Investigate the existing migrations to understand the current schema
@@ -18,7 +18,7 @@ You may take any of the following actions:
 
 ## History
 {{ instructions.history_truncated }}
-{{ history_to_json(state.history, max_events=20) }}
+{{ history_to_json(state.history[-10:]) }}
 
 ## Format
 {{ instructions.format.action }}

agenthub/micro/registry.py

@@ -4,10 +4,7 @@ import yaml
 
 all_microagents = {}
 
-# Get the list of directories and sort them to preserve determinism
-dirs = sorted(os.listdir(os.path.dirname(__file__)))
-
-for dir in dirs:
+for dir in os.listdir(os.path.dirname(__file__)):
     base = os.path.dirname(__file__) + '/' + dir
     if os.path.isfile(base):
         continue
@@ -16,7 +13,8 @@ for dir in dirs:
     promptFile = base + '/prompt.md'
     agentFile = base + '/agent.yaml'
     if not os.path.isfile(promptFile) or not os.path.isfile(agentFile):
-        raise Exception(f'Missing prompt or agent file in {base}. Please create them.')
+        raise Exception(
+            f'Missing prompt or agent file in {base}. Please create them.')
     with open(promptFile, 'r') as f:
         prompt = f.read()
     with open(agentFile, 'r') as f:

agenthub/micro/repo_explorer/prompt.md

@@ -20,7 +20,7 @@ When you're done, put your summary into the output of the `finish` action.
 
 ## History
 {{ instructions.history_truncated }}
-{{ history_to_json(state.history, max_events=20) }}
+{{ history_to_json(state.history[-10:]) }}
 
 ## Format
 {{ instructions.format.action }}

agenthub/micro/study_repo_for_task/prompt.md

@@ -1,11 +1,9 @@
 # Task
-You are a software architect. Your team has inherited an existing codebase, and
-need to finish a project:
+You are a software engineer. You've inherited an existing codebase, which you're
+learning about for the first time. You need to study the codebase to find all
+the information needed to complete this task:
 
-{{ state.inputs.task }}
-
-As an architect, you need to study the codebase to find all the information that
-might be helpful for your software engineering team.
+{{ latest_user_message }}
 
 ## Available Actions
 {{ instructions.actions.run }}
@@ -13,50 +11,15 @@ might be helpful for your software engineering team.
 {{ instructions.actions.message }}
 {{ instructions.actions.finish }}
 
-You must ONLY `run` commands that have no side-effects, like `ls` and `grep`. You
-MUST NOT modify or write to any file.
+You must ONLY `run` commands that have no side-effects, like `ls` and `grep`.
 
 Do NOT finish until you have a complete understanding of which parts of the
-codebase are relevant to the project, including particular files, functions, and classes.
+codebase are relevant to the task, including particular files, functions, and classes.
 When you're done, put your summary in `outputs.summary` in the `finish` action.
-Remember, your task is to explore and study the current repository, not actually
-implement the solution. If the codebase is empty, you should call the `finish` action.
 
 ## History
 {{ instructions.history_truncated }}
-{{ history_to_json(state.history, max_events=20) }}
+{{ history_to_json(state.history[-10:]) }}
 
 ## Format
 {{ instructions.format.action }}
-
-## Examples
-
-Here is an example of how you can interact with the environment for task solving:
-
---- START OF EXAMPLE ---
-
-USER: Can you create a list of numbers from 1 to 10, and create a web page to display them at port 5000?
-
-ASSISTANT:
-{
-  "action": "run",
-  "args": {
-    "command": "ls"
-  }
-}
-
-USER:
-OBSERVATION:
-[]
-
-ASSISTANT:
-{
-  "action": "finish",
-  "args": {
-    "outputs": {
-      "summary": "The codebase appears to be empty. Engineers should start everything from scratch."
-    }
-  }
-}
-
---- END OF EXAMPLE ---

agenthub/micro/typo_fixer_agent/agent.yaml

@@ -1,6 +1,5 @@
 name: TypoFixerAgent
 description: Fixes typos in files in the current working directory
-inputs:
-  task: string
+inputs: {}
 outputs:
   summary: string

agenthub/micro/typo_fixer_agent/prompt.md

@@ -1,13 +1,5 @@
 # Task
-You are a proofreader tasked with fixing typos in the files in your current working directory.
-
-{% if state.inputs.task %}
-Specifically, your task is:
-{{ state.inputs.task }}
-{% endif %}
-
-To achieve this goal, you should:
-
+You are a proofreader tasked with fixing typos in the files in your current working directory. Your goal is to:
 1. Scan the files for typos
 2. Overwrite the files with the typos fixed
 3. Provide a summary of the typos fixed
@@ -21,17 +13,17 @@ To achieve this goal, you should:
 
 To complete this task:
 1. Use the `read` action to read the contents of the files in your current working directory. Make sure to provide the file path in the format `'./file_name.ext'`.
-2. Use the `message` action to analyze the contents and identify typos.
+2. Use the `think` action to analyze the contents and identify typos.
 3. Use the `write` action to create new versions of the files with the typos fixed.
   - Overwrite the original files with the corrected content. Make sure to provide the file path in the format `'./file_name.ext'`.
-4. Use the `message` action to generate a summary of the typos fixed, including the original and fixed versions of each typo, and the file(s) they were found in.
+4. Use the `think` action to generate a summary of the typos fixed, including the original and fixed versions of each typo, and the file(s) they were found in.
 5. Use the `finish` action to return the summary in the `outputs.summary` field.
 
 Do NOT finish until you have fixed all the typos and generated a summary.
 
 ## History
 {{ instructions.history_truncated }}
-{{ history_to_json(state.history, max_events=10) }}
+{{ history_to_json(state.history[-5:]) }}
 
 ## Format
 {{ instructions.format.action }}

agenthub/micro/verifier/prompt.md

@@ -2,10 +2,9 @@
 You are a quality assurance engineer. Another engineer has made changes to the
 codebase which are supposed to solve this task:
 
-{{ state.inputs.task }}
+{{ latest_user_message }}
 
-Note the changes might have already been applied in-line. You should focus on
-validating if the task is solved, nothing else.
+Your goal is to verify that the changes are correct and bug-free.
 
 ## Available Actions
 {{ instructions.actions.run }}
@@ -22,7 +21,7 @@ explaining what the problem is.
 
 ## History
 {{ instructions.history_truncated }}
-{{ history_to_json(state.history, max_events=20) }}
+{{ history_to_json(state.history[-10:]) }}
 
 ## Format
 {{ instructions.format.action }}

agenthub/monologue_agent/.dockerignore

@@ -0,0 +1,2 @@
+.envrc
+workspace

agenthub/monologue_agent/README.md

@@ -0,0 +1,8 @@
+# LLM control loop
+This is currently a standalone utility. It will need to be integrated into OpenDevin's backend.
+
+## Usage
+```bash
+# Run this in project root
+./agenthub/monologue_agent/build-and-run.sh "write a bash script that prints 'hello world'"
+```

agenthub/monologue_agent/TODO.md

@@ -0,0 +1,8 @@
+# TODO
+There's a lot of low-hanging fruit for this agent:
+
+* Strip `<script>`, `<style>`, and other non-text tags from the HTML before sending it to the LLM
+* Keep track of the working directory when the agent uses `cd`
+* Improve memory condensing--condense earlier memories more aggressively
+* Limit the time that `run` can wait (in case agent runs an interactive command and it's hanging)
+* Figure out how to run background processes, e.g. `node server.js` to start a server

agenthub/monologue_agent/__init__.py

@@ -0,0 +1,5 @@
+from opendevin.controller.agent import Agent
+
+from .agent import MonologueAgent
+
+Agent.register('MonologueAgent', MonologueAgent)

agenthub/monologue_agent/agent.py

@@ -0,0 +1,270 @@
+import agenthub.monologue_agent.utils.prompts as prompts
+from opendevin.controller.agent import Agent
+from opendevin.controller.state.state import State
+from opendevin.core.config import config
+from opendevin.core.exceptions import AgentNoInstructionError
+from opendevin.core.schema import ActionType
+from opendevin.events.action import (
+    Action,
+    AgentRecallAction,
+    BrowseURLAction,
+    CmdRunAction,
+    FileReadAction,
+    FileWriteAction,
+    MessageAction,
+    NullAction,
+)
+from opendevin.events.observation import (
+    AgentRecallObservation,
+    BrowserOutputObservation,
+    CmdOutputObservation,
+    FileReadObservation,
+    NullObservation,
+    Observation,
+)
+from opendevin.events.serialization.event import event_to_memory
+from opendevin.llm.llm import LLM
+from opendevin.memory.condenser import MemoryCondenser
+from opendevin.memory.history import ShortTermHistory
+
+if config.agent.memory_enabled:
+    from opendevin.memory.memory import LongTermMemory
+
+MAX_TOKEN_COUNT_PADDING = 512
+MAX_OUTPUT_LENGTH = 5000
+
+INITIAL_THOUGHTS = [
+    'I exist!',
+    'Hmm...looks like I can type in a command line prompt',
+    'Looks like I have a web browser too!',
+    "Here's what I want to do: $TASK",
+    'How am I going to get there though?',
+    'It seems like I have some kind of short term memory.',
+    'Each of my thoughts seems to be stored in a JSON array.',
+    'It seems whatever I say next will be added as an object to the list.',
+    'But no one has perfect short-term memory. My list of thoughts will be summarized and condensed over time, losing information in the process.',
+    'Fortunately I have long term memory!',
+    'I can just perform a recall action, followed by the thing I want to remember. And then related thoughts just spill out!',
+    "Sometimes they're random thoughts that don't really have to do with what I wanted to remember. But usually they're exactly what I need!",
+    "Let's try it out!",
+    'RECALL what it is I want to do',
+    "Here's what I want to do: $TASK",
+    'How am I going to get there though?',
+    "Neat! And it looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
+    'RUN echo "hello world"',
+    'hello world',
+    'Cool! I bet I can write files too using the write action.',
+    'WRITE echo "console.log(\'hello world\')" > test.js',
+    '',
+    "I just created test.js. I'll try and run it now.",
+    'RUN node test.js',
+    'hello world',
+    'It works!',
+    "I'm going to try reading it now using the read action.",
+    'READ test.js',
+    "console.log('hello world')",
+    'Nice! I can read files too!',
+    'And if I want to use the browser, I just need to use the browse action and include the url I want to visit in the url argument',
+    "Let's try that...",
+    'BROWSE google.com',
+    '<form><input type="text"></input><button type="submit"></button></form>',
+    'I can browse the web too!',
+    'And once I have completed my task, I can use the finish action to stop working.',
+    "But I should only use the finish action when I'm absolutely certain that I've completed my task and have tested my work.",
+    'Very cool. Now to accomplish my task.',
+    "I'll need a strategy. And as I make progress, I'll need to keep refining that strategy. I'll need to set goals, and break them into sub-goals.",
+    'In between actions, I must always take some time to think, strategize, and set new goals. I should never take two actions in a row.',
+    "OK so my task is to $TASK. I haven't made any progress yet. Where should I start?",
+    'It seems like there might be an existing project here. I should probably start by running `pwd` and `ls` to orient myself.',
+]
+
+
+class MonologueAgent(Agent):
+    """
+    The Monologue Agent utilizes long and short term memory to complete tasks.
+    Long term memory is stored as a LongTermMemory object and the model uses it to search for examples from the past.
+    Short term memory is stored as a Monologue object and the model can condense it as necessary.
+    """
+
+    _initialized = False
+    monologue: ShortTermHistory
+    memory: 'LongTermMemory | None'
+    memory_condenser: MemoryCondenser
+
+    def __init__(self, llm: LLM):
+        """
+        Initializes the Monologue Agent with an llm, monologue, and memory.
+
+        Parameters:
+        - llm (LLM): The llm to be used by this agent
+        """
+        super().__init__(llm)
+
+    def _add_event(self, event_dict: dict):
+        """
+        Adds a new event to the agent's monologue and memory.
+        Monologue automatically condenses when it gets too large.
+
+        Parameters:
+        - event (dict): The event that will be added to monologue and memory
+        """
+
+        if (
+            'args' in event_dict
+            and 'output' in event_dict['args']
+            and len(event_dict['args']['output']) > MAX_OUTPUT_LENGTH
+        ):
+            event_dict['args']['output'] = (
+                event_dict['args']['output'][:MAX_OUTPUT_LENGTH] + '...'
+            )
+
+        self.monologue.add_event(event_dict)
+        if self.memory is not None:
+            self.memory.add_event(event_dict)
+
+        # Test monologue token length
+        prompt = prompts.get_request_action_prompt(
+            '',
+            self.monologue.get_events(),
+            [],
+        )
+        messages = [{'content': prompt, 'role': 'user'}]
+        token_count = self.llm.get_token_count(messages)
+
+        if token_count + MAX_TOKEN_COUNT_PADDING > self.llm.max_input_tokens:
+            prompt = prompts.get_summarize_monologue_prompt(self.monologue.events)
+            summary_response = self.memory_condenser.condense(
+                summarize_prompt=prompt, llm=self.llm
+            )
+            self.monologue.events = prompts.parse_summary_response(summary_response)
+
+    def _initialize(self, task: str):
+        """
+        Utilizes the INITIAL_THOUGHTS list to give the agent a context for its capabilities
+        and how to navigate the WORKSPACE_MOUNT_PATH_IN_SANDBOX in `config` (e.g., /workspace by default).
+        Short circuited to return when already initialized.
+        Will execute again when called after reset.
+
+        Parameters:
+        - task (str): The initial goal statement provided by the user
+
+        Raises:
+        - AgentNoInstructionError: If task is not provided
+        """
+
+        if self._initialized:
+            return
+
+        if task is None or task == '':
+            raise AgentNoInstructionError()
+
+        self.monologue = ShortTermHistory()
+        if config.agent.memory_enabled:
+            self.memory = LongTermMemory()
+        else:
+            self.memory = None
+
+        self.memory_condenser = MemoryCondenser()
+
+        self._add_initial_thoughts(task)
+        self._initialized = True
+
+    def _add_initial_thoughts(self, task):
+        previous_action = ''
+        for thought in INITIAL_THOUGHTS:
+            thought = thought.replace('$TASK', task)
+            if previous_action != '':
+                observation: Observation = NullObservation(content='')
+                if previous_action in {ActionType.RUN, ActionType.PUSH}:
+                    observation = CmdOutputObservation(
+                        content=thought, command_id=0, command=''
+                    )
+                elif previous_action == ActionType.READ:
+                    observation = FileReadObservation(content=thought, path='')
+                elif previous_action == ActionType.RECALL:
+                    observation = AgentRecallObservation(content=thought, memories=[])
+                elif previous_action == ActionType.BROWSE:
+                    observation = BrowserOutputObservation(
+                        content=thought, url='', screenshot=''
+                    )
+                self._add_event(event_to_memory(observation))
+                previous_action = ''
+            else:
+                action: Action = NullAction()
+                if thought.startswith('RUN'):
+                    command = thought.split('RUN ')[1]
+                    action = CmdRunAction(command)
+                    previous_action = ActionType.RUN
+                elif thought.startswith('WRITE'):
+                    parts = thought.split('WRITE ')[1].split(' > ')
+                    path = parts[1]
+                    content = parts[0]
+                    action = FileWriteAction(path=path, content=content)
+                elif thought.startswith('READ'):
+                    path = thought.split('READ ')[1]
+                    action = FileReadAction(path=path)
+                    previous_action = ActionType.READ
+                elif thought.startswith('RECALL'):
+                    query = thought.split('RECALL ')[1]
+                    action = AgentRecallAction(query=query)
+                    previous_action = ActionType.RECALL
+                elif thought.startswith('BROWSE'):
+                    url = thought.split('BROWSE ')[1]
+                    action = BrowseURLAction(url=url)
+                    previous_action = ActionType.BROWSE
+                else:
+                    action = MessageAction(thought)
+                self._add_event(event_to_memory(action))
+
+    def step(self, state: State) -> Action:
+        """
+        Modifies the current state by adding the most recent actions and observations, then prompts the model to think about it's next action to take using monologue, memory, and hint.
+
+        Parameters:
+        - state (State): The current state based on previous steps taken
+
+        Returns:
+        - Action: The next action to take based on LLM response
+        """
+
+        goal = state.get_current_user_intent()
+        self._initialize(goal)
+        for prev_action, obs in state.updated_info:
+            self._add_event(event_to_memory(prev_action))
+            self._add_event(event_to_memory(obs))
+
+        state.updated_info = []
+
+        prompt = prompts.get_request_action_prompt(
+            goal,
+            self.monologue.get_events(),
+            state.background_commands_obs,
+        )
+        messages = [{'content': prompt, 'role': 'user'}]
+        resp = self.llm.completion(messages=messages)
+        action_resp = resp['choices'][0]['message']['content']
+        state.num_of_chars += len(prompt) + len(action_resp)
+        action = prompts.parse_action_response(action_resp)
+        self.latest_action = action
+        return action
+
+    def search_memory(self, query: str) -> list[str]:
+        """
+        Uses VectorIndexRetriever to find related memories within the long term memory.
+        Uses search to produce top 10 results.
+
+        Parameters:
+        - query (str): The query that we want to find related memories for
+
+        Returns:
+        - list[str]: A list of top 10 text results that matched the query
+        """
+        if self.memory is None:
+            return []
+        return self.memory.search(query)
+
+    def reset(self) -> None:
+        super().reset()
+
+        # Reset the initial monologue and memory
+        self._initialized = False

agenthub/monologue_agent/utils/prompts.py

@@ -0,0 +1,190 @@
+from opendevin.core.config import config
+from opendevin.core.utils import json
+from opendevin.events.action import (
+    Action,
+)
+from opendevin.events.observation import (
+    CmdOutputObservation,
+)
+from opendevin.events.serialization.action import action_from_dict
+
+ACTION_PROMPT = """
+You're a thoughtful robot. Your main task is this:
+%(task)s
+
+Don't expand the scope of your task--just complete it as written.
+
+This is your internal monologue, in JSON format:
+
+%(monologue)s
+
+
+Your most recent thought is at the bottom of that monologue. Continue your train of thought.
+What is your next single thought or action? Your response must be in JSON format.
+It must be a single object, and it must contain two fields:
+* `action`, which is one of the actions below
+* `args`, which is a map of key-value pairs, specifying the arguments for that action
+
+Here are the possible actions:
+* `read` - reads the content of a file. Arguments:
+  * `path` - the path of the file to read
+* `write` - writes the content to a file. Arguments:
+  * `path` - the path of the file to write
+  * `content` - the content to write to the file
+* `run` - runs a command. Arguments:
+  * `command` - the command to run
+  * `background` - if true, run the command in the background, so that other commands can be run concurrently. Useful for e.g. starting a server. You won't be able to see the logs. You don't need to end the command with `&`, just set this to true.
+* `kill` - kills a background command
+  * `command_id` - the ID of the background command to kill
+* `browse` - opens a web page. Arguments:
+  * `url` - the URL to open
+* `push` - Push a branch from the current repo to github:
+  * `owner` - the owner of the repo to push to
+  * `repo` - the name of the repo to push to
+  * `branch` - the name of the branch to push
+* `recall` - recalls a past memory. Arguments:
+  * `query` - the query to search for
+* `message` - make a plan, set a goal, record your thoughts, or ask for more input from the user. Arguments:
+  * `content` - the message to record
+  * `wait_for_response` - set to `true` to wait for the user to respond before proceeding
+* `finish` - if you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
+
+%(background_commands)s
+
+You MUST take time to think in between read, write, run, browse, push, and recall actions--do this with the `message` action.
+You should never act twice in a row without thinking. But if your last several
+actions are all `message` actions, you should consider taking a different action.
+
+Notes:
+* you are logged in as %(user)s, but sudo will always work without a password.
+* all non-background commands will be forcibly stopped if they remain running for over %(timeout)s seconds.
+* your environment is Debian Linux. You can install software with `sudo apt-get`, but remember to use -y.
+* don't run interactive commands, or commands that don't return (e.g. `node server.js`). You may run commands in the background (e.g. `node server.js &`)
+* don't run interactive text editors (e.g. `nano` or 'vim'), instead use the 'write' or 'read' action.
+* don't run gui applications (e.g. software IDEs (like vs code or codium), web browsers (like firefox or chromium), or other complex software packages). Use non-interactive cli applications, or special actions instead.
+* whenever an action fails, always send a `message` about why it may have happened before acting again.
+
+What is your next single thought or action? Again, you must reply with JSON, and only with JSON. You must respond with exactly one 'action' object.
+
+%(hint)s
+"""
+
+MONOLOGUE_SUMMARY_PROMPT = """
+Below is the internal monologue of an automated LLM agent. Each
+thought is an item in a JSON array. The thoughts may be memories,
+actions taken by the agent, or outputs from those actions.
+Please return a new, smaller JSON array, which summarizes the
+internal monologue. You can summarize individual thoughts, and
+you can condense related thoughts together with a description
+of their content.
+
+%(monologue)s
+
+Make the summaries as pithy and informative as possible.
+Be specific about what happened and what was learned. The summary
+will be used as keywords for searching for the original memory.
+Be sure to preserve any key words or important information.
+
+Your response must be in JSON format. It must be an object with the
+key `new_monologue`, which is a JSON array containing the summarized monologue.
+Each entry in the array must have an `action` key, and an `args` key.
+The action key may be `summarize`, and `args.summary` should contain the summary.
+You can also use the same action and args from the source monologue.
+"""
+
+
+def get_summarize_monologue_prompt(thoughts: list[dict]):
+    """
+    Gets the prompt for summarizing the monologue
+
+    Returns:
+    - str: A formatted string with the current monologue within the prompt
+    """
+    return MONOLOGUE_SUMMARY_PROMPT % {
+        'monologue': json.dumps({'old_monologue': thoughts}, indent=2),
+    }
+
+
+def get_request_action_prompt(
+    task: str,
+    thoughts: list[dict],
+    background_commands_obs: list[CmdOutputObservation] = [],
+):
+    """
+    Gets the action prompt formatted with appropriate values.
+
+    Parameters:
+    - task (str): The current task the agent is trying to accomplish
+    - thoughts (list[dict]): The agent's current thoughts
+    - background_commands_obs (list[CmdOutputObservation]): list of all observed background commands running
+
+    Returns:
+    - str: Formatted prompt string with hint, task, monologue, and background included
+    """
+
+    hint = ''
+    if len(thoughts) > 0:
+        latest_thought = thoughts[-1]
+        if 'action' in latest_thought:
+            if latest_thought['action'] == 'message':
+                if latest_thought['args']['content'].startswith('OK so my task is'):
+                    hint = "You're just getting started! What should you do first?"
+                else:
+                    hint = "You've been thinking a lot lately. Maybe it's time to take action?"
+            elif latest_thought['action'] == 'error':
+                hint = 'Looks like that last command failed. Maybe you need to fix it, or try something else.'
+
+    bg_commands_message = ''
+    if len(background_commands_obs) > 0:
+        bg_commands_message = 'The following commands are running in the background:'
+        for command_obs in background_commands_obs:
+            bg_commands_message += (
+                f'\n`{command_obs.command_id}`: {command_obs.command}'
+            )
+        bg_commands_message += '\nYou can end any process by sending a `kill` action with the numerical `command_id` above.'
+
+    user = 'opendevin' if config.run_as_devin else 'root'
+
+    return ACTION_PROMPT % {
+        'task': task,
+        'monologue': json.dumps(thoughts, indent=2),
+        'background_commands': bg_commands_message,
+        'hint': hint,
+        'user': user,
+        'timeout': config.sandbox_timeout,
+        'WORKSPACE_MOUNT_PATH_IN_SANDBOX': config.workspace_mount_path_in_sandbox,
+    }
+
+
+def parse_action_response(orig_response: str) -> Action:
+    """
+    Parses a string to find an action within it
+
+    Parameters:
+    - response (str): The string to be parsed
+
+    Returns:
+    - Action: The action that was found in the response string
+    """
+    # attempt to load the JSON dict from the response
+    action_dict = json.loads(orig_response)
+
+    if 'content' in action_dict:
+        # The LLM gets confused here. Might as well be robust
+        action_dict['contents'] = action_dict.pop('content')
+
+    return action_from_dict(action_dict)
+
+
+def parse_summary_response(response: str) -> list[dict]:
+    """
+    Parses a summary of the monologue
+
+    Parameters:
+    - response (str): The response string to be parsed
+
+    Returns:
+    - list[dict]: The list of summaries output by the model
+    """
+    parsed = json.loads(response)
+    return parsed['new_monologue']

agenthub/planner_agent/agent.py

@@ -1,24 +1,20 @@
-from agenthub.planner_agent.response_parser import PlannerResponseParser
 from opendevin.controller.agent import Agent
 from opendevin.controller.state.state import State
 from opendevin.events.action import Action, AgentFinishAction
 from opendevin.llm.llm import LLM
-from opendevin.runtime.tools import RuntimeTool
 
-from .prompt import get_prompt
+from .prompt import get_prompt, parse_response
 
 
 class PlannerAgent(Agent):
-    VERSION = '1.0'
     """
     The planner agent utilizes a special prompting strategy to create long term plans for solving problems.
     The agent is given its previous action-observation pairs, current task, and hint based on last action taken at every step.
     """
-    runtime_tools: list[RuntimeTool] = [RuntimeTool.BROWSER]
-    response_parser = PlannerResponseParser()
 
     def __init__(self, llm: LLM):
-        """Initialize the Planner Agent with an LLM
+        """
+        Initialize the Planner Agent with an LLM
 
         Parameters:
         - llm (LLM): The llm to be used by this agent
@@ -26,7 +22,8 @@ class PlannerAgent(Agent):
         super().__init__(llm)
 
     def step(self, state: State) -> Action:
-        """Checks to see if current step is completed, returns AgentFinishAction if True.
+        """
+        Checks to see if current step is completed, returns AgentFinishAction if True.
         Otherwise, creates a plan prompt and sends to model for inference, returning the result as the next action.
 
         Parameters:
@@ -36,13 +33,20 @@ class PlannerAgent(Agent):
         - AgentFinishAction: If the last state was 'completed', 'verified', or 'abandoned'
         - Action: The next action to take based on llm response
         """
+
         if state.root_task.state in [
             'completed',
             'verified',
             'abandoned',
         ]:
             return AgentFinishAction()
-        prompt = get_prompt(state, self.llm.config.max_message_chars)
+        prompt = get_prompt(state)
         messages = [{'content': prompt, 'role': 'user'}]
         resp = self.llm.completion(messages=messages)
-        return self.response_parser.parse(resp)
+        action_resp = resp['choices'][0]['message']['content']
+        state.num_of_chars += len(prompt) + len(action_resp)
+        action = parse_response(action_resp)
+        return action
+
+    def search_memory(self, query: str) -> list[str]:
+        return []

agenthub/planner_agent/prompt.py

@@ -6,10 +6,13 @@ from opendevin.events.action import (
     Action,
     NullAction,
 )
+from opendevin.events.observation import (
+    NullObservation,
+)
 from opendevin.events.serialization.action import action_from_dict
 from opendevin.events.serialization.event import event_to_memory
 
-HISTORY_SIZE = 20
+HISTORY_SIZE = 10
 
 prompt = """
 # Task
@@ -74,6 +77,9 @@ It must be an object, and it must contain two fields:
   * `content` - the content to write to the file
 * `run` - runs a command on the command line in a Linux shell. Arguments:
   * `command` - the command to run
+  * `background` - if true, run the command in the background, so that other commands can be run concurrently. Useful for e.g. starting a server. You won't be able to see the logs. You don't need to end the command with `&`, just set this to true.
+* `kill` - kills a background command
+  * `command_id` - the ID of the background command to kill
 * `browse` - opens a web page. Arguments:
   * `url` - the URL to open
 * `message` - make a plan, set a goal, record your thoughts, or ask for more input from the user. Arguments:
@@ -88,7 +94,7 @@ It must be an object, and it must contain two fields:
   * `state` - set to 'in_progress' to start the task, 'completed' to finish it, 'verified' to assert that it was successful, 'abandoned' to give up on it permanently, or `open` to stop working on it for now.
 * `finish` - if ALL of your tasks and subtasks have been verified or abandoned, and you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
 
-You MUST take time to think in between read, write, run, and browse actions--do this with the `message` action.
+You MUST take time to think in between read, write, run, browse, and recall actions--do this with the `message` action.
 You should never act twice in a row without thinking. But if your last several
 actions are all `message` actions, you should consider taking a different action.
 
@@ -100,6 +106,7 @@ What is your next thought or action? Again, you must reply with JSON, and only w
 
 def get_hint(latest_action_id: str) -> str:
     """Returns action type hint based on given action_id"""
+
     hints = {
         '': "You haven't taken any actions yet. Start by using `ls` to check out what files you're working with.",
         ActionType.RUN: 'You should think about the command you just ran, what output it gave, and how that affects your plan.',
@@ -107,6 +114,7 @@ def get_hint(latest_action_id: str) -> str:
         ActionType.WRITE: 'You just changed a file. You should think about how it affects your plan.',
         ActionType.BROWSE: 'You should think about the page you just visited, and what you learned from it.',
         ActionType.MESSAGE: "Look at your last thought in the history above. What does it suggest? Don't think anymore--take action.",
+        ActionType.RECALL: 'You should think about the information you just recalled, and how it should affect your plan.',
         ActionType.ADD_TASK: 'You should think about the next action to take.',
         ActionType.MODIFY_TASK: 'You should think about the next action to take.',
         ActionType.SUMMARIZE: '',
@@ -115,9 +123,9 @@ def get_hint(latest_action_id: str) -> str:
     return hints.get(latest_action_id, '')
 
 
-def get_prompt(state: State, max_message_chars: int) -> str:
-    """Gets the prompt for the planner agent.
-
+def get_prompt(state: State) -> str:
+    """
+    Gets the prompt for the planner agent.
     Formatted with the most recent action-observation pairs, current task, and hint based on last action
 
     Parameters:
@@ -126,28 +134,19 @@ def get_prompt(state: State, max_message_chars: int) -> str:
     Returns:
     - str: The formatted string prompt with historical values
     """
-    # the plan
-    plan_str = json.dumps(state.root_task.to_dict(), indent=2)
 
-    # the history
+    plan_str = json.dumps(state.root_task.to_dict(), indent=2)
+    sub_history = state.history[-HISTORY_SIZE:]
     history_dicts = []
     latest_action: Action = NullAction()
-
-    # retrieve the latest HISTORY_SIZE events
-    for event_count, event in enumerate(state.history.get_events(reverse=True)):
-        if event_count >= HISTORY_SIZE:
-            break
-        if latest_action == NullAction() and isinstance(event, Action):
-            latest_action = event
-        history_dicts.append(event_to_memory(event, max_message_chars))
-
-    # history_dicts is in reverse order, lets fix it
-    history_dicts.reverse()
-
-    # and get it as a JSON string
+    for action, observation in sub_history:
+        if not isinstance(action, NullAction):
+            history_dicts.append(event_to_memory(action))
+            latest_action = action
+        if not isinstance(observation, NullObservation):
+            observation_dict = event_to_memory(observation)
+            history_dicts.append(observation_dict)
     history_str = json.dumps(history_dicts, indent=2)
-
-    # the plan status
     current_task = state.root_task.get_current_task()
     if current_task is not None:
         plan_status = f"You're currently working on this task:\n{current_task.goal}."
@@ -155,15 +154,9 @@ def get_prompt(state: State, max_message_chars: int) -> str:
             plan_status += "\nIf it's not achievable AND verifiable with a SINGLE action, you MUST break it down into subtasks NOW."
     else:
         plan_status = "You're not currently working on any tasks. Your next action MUST be to mark a task as in_progress."
-
-    # the hint, based on the last action
-    hint = get_hint(event_to_memory(latest_action, max_message_chars).get('action', ''))
-    logger.info('HINT:\n' + hint, extra={'msg_type': 'DETAIL'})
-
-    # the last relevant user message (the task)
+    hint = get_hint(event_to_memory(latest_action).get('action', ''))
+    logger.info('HINT:\n' + hint, extra={'msg_type': 'INFO'})
     task = state.get_current_user_intent()
-
-    # finally, fill in the prompt
     return prompt % {
         'task': task,
         'plan': plan_str,
@@ -174,7 +167,9 @@ def get_prompt(state: State, max_message_chars: int) -> str:
 
 
 def parse_response(response: str) -> Action:
-    """Parses the model output to find a valid action to take
+    """
+    Parses the model output to find a valid action to take
+
     Parameters:
     - response (str): A response from the model that potentially contains an Action.
 

agenthub/planner_agent/response_parser.py

@@ -1,37 +0,0 @@
-from opendevin.controller.action_parser import ResponseParser
-from opendevin.core.utils import json
-from opendevin.events.action import (
-    Action,
-)
-from opendevin.events.serialization.action import action_from_dict
-
-
-class PlannerResponseParser(ResponseParser):
-    def __init__(self):
-        super().__init__()
-
-    def parse(self, response: str) -> Action:
-        action_str = self.parse_response(response)
-        return self.parse_action(action_str)
-
-    def parse_response(self, response) -> str:
-        # get the next action from the response
-        return response['choices'][0]['message']['content']
-
-    def parse_action(self, action_str: str) -> Action:
-        """Parses a string to find an action within it
-
-        Parameters:
-        - response (str): The string to be parsed
-
-        Returns:
-        - Action: The action that was found in the response string
-        """
-        # attempt to load the JSON dict from the response
-        action_dict = json.loads(action_str)
-
-        if 'content' in action_dict:
-            # The LLM gets confused here. Might as well be robust
-            action_dict['contents'] = action_dict.pop('content')
-
-        return action_from_dict(action_dict)

config.template.toml

@@ -1,204 +0,0 @@
-###################### OpenDevin Configuration Example ######################
-#
-# All settings have default values, so you only need to uncomment and
-# modify what you want to change
-# The fields within each section are sorted in alphabetical order.
-#
-##############################################################################
-
-#################################### Core ####################################
-# General core configurations
-##############################################################################
-[core]
-# API key for E2B
-#e2b_api_key = ""
-
-# Base path for the workspace
-workspace_base = "./workspace"
-
-# Cache directory path
-#cache_dir = "/tmp/cache"
-
-# Debugging enabled
-#debug = false
-
-# Disable color in terminal output
-#disable_color = false
-
-# Enable saving and restoring the session when run from CLI
-#enable_cli_session = false
-
-# File store path
-#file_store_path = "/tmp/file_store"
-
-# File store type
-#file_store = "memory"
-
-# List of allowed file extensions for uploads
-#file_uploads_allowed_extensions = [".*"]
-
-# Maximum file size for uploads, in megabytes
-#file_uploads_max_file_size_mb = 0
-
-# Maximum budget per task, 0.0 means no limit
-#max_budget_per_task = 0.0
-
-# Maximum number of iterations
-#max_iterations = 100
-
-# Path to mount the workspace in the sandbox
-#workspace_mount_path_in_sandbox = "/workspace"
-
-# Path to mount the workspace
-#workspace_mount_path = ""
-
-# Path to rewrite the workspace mount path to
-#workspace_mount_rewrite = ""
-
-# Persist the sandbox
-persist_sandbox = false
-
-# Run as devin
-#run_as_devin = true
-
-# Runtime environment
-#runtime = "server"
-
-# SSH hostname for the sandbox
-#ssh_hostname = "localhost"
-
-# SSH password for the sandbox
-#ssh_password = ""
-
-# SSH port for the sandbox
-#ssh_port = 63710
-
-
-# Name of the default agent
-#default_agent = "CodeActAgent"
-
-#################################### LLM #####################################
-# Configuration for LLM models (group name starts with 'llm')
-# use 'llm' for the default LLM config
-##############################################################################
-[llm]
-# AWS access key ID
-#aws_access_key_id = ""
-
-# AWS region name
-#aws_region_name = ""
-
-# AWS secret access key
-#aws_secret_access_key = ""
-
-# API key to use
-api_key = "your-api-key"
-
-# API base URL
-#base_url = ""
-
-# API version
-#api_version = ""
-
-# Cost per input token
-#input_cost_per_token = 0.0
-
-# Cost per output token
-#output_cost_per_token = 0.0
-
-# Custom LLM provider
-#custom_llm_provider = ""
-
-# Embedding API base URL
-#embedding_base_url = ""
-
-# Embedding deployment name
-#embedding_deployment_name = ""
-
-# Embedding model to use
-embedding_model = ""
-
-# Maximum number of characters in an observation's content
-#max_message_chars = 10000
-
-# Maximum number of input tokens
-#max_input_tokens = 0
-
-# Maximum number of output tokens
-#max_output_tokens = 0
-
-# Model to use
-model = "gpt-4o"
-
-# Number of retries to attempt
-#num_retries = 5
-
-# Retry maximum wait time
-#retry_max_wait = 60
-
-# Retry minimum wait time
-#retry_min_wait = 3
-
-# Temperature for the API
-#temperature = 0.0
-
-# Timeout for the API
-#timeout = 0
-
-# Top p for the API
-#top_p = 0.5
-
-[llm.gpt3]
-# API key to use
-api_key = "your-api-key"
-
-# Model to use
-model = "gpt-3.5"
-
-#################################### Agent ###################################
-# Configuration for agents (group name starts with 'agent')
-# Use 'agent' for the default agent config
-# otherwise, group name must be `agent.<agent_name>` (case-sensitive), e.g.
-# agent.CodeActAgent
-##############################################################################
-[agent]
-# Memory enabled
-#memory_enabled = false
-
-# Memory maximum threads
-#memory_max_threads = 2
-
-# LLM config group to use
-#llm_config = 'llm'
-
-[agent.RepoExplorerAgent]
-# Example: use a cheaper model for RepoExplorerAgent to reduce cost, especially
-# useful when an agent doesn't demand high quality but uses a lot of tokens
-llm_config = 'gpt3'
-
-#################################### Sandbox ###################################
-# Configuration for the sandbox
-##############################################################################
-[sandbox]
-# Sandbox timeout in seconds
-#timeout = 120
-
-# Sandbox type (ssh, e2b, local)
-#box_type = "ssh"
-
-# Sandbox user ID
-#user_id = 1000
-
-# Container image to use for the sandbox
-#container_image = "ghcr.io/opendevin/sandbox:main"
-
-# Use host network
-#use_host_network = false
-
-# Enable auto linting after editing
-#enable_auto_lint = false
-
-#################################### Eval ####################################
-# Configuration for the evaluation, please refer to the specific evaluation
-# plugin for the available options
-##############################################################################

containers/README.md

@@ -1,12 +1,11 @@
 # Docker Containers
-
 Each folder here contains a Dockerfile, and a config.sh describing how to build
-the images and where to push them. These images are built and pushed in GitHub Actions
+the image and where to push it. These are images are built and pushed in GitHub Actions
 by the `ghcr.yml` workflow.
 
 ## Building Manually
 
-```bash
+```
 docker build -f containers/app/Dockerfile -t opendevin .
 docker build -f containers/sandbox/Dockerfile -t sandbox .
 ```

containers/app/Dockerfile

@@ -1,5 +1,5 @@
 ARG OPEN_DEVIN_BUILD_VERSION=dev
-FROM node:21.7.2-bookworm-slim AS frontend-builder
+FROM node:21.7.2-bookworm-slim as frontend-builder
 
 WORKDIR /app
 
@@ -10,10 +10,10 @@ RUN npm ci
 COPY ./frontend ./
 RUN npm run make-i18n && npm run build
 
-FROM python:3.12.3-slim AS backend-builder
+FROM python:3.12-slim as backend-builder
 
 WORKDIR /app
-ENV PYTHONPATH='/app'
+ENV PYTHONPATH '/app'
 
 ENV POETRY_NO_INTERACTION=1 \
     POETRY_VIRTUALENVS_IN_PROJECT=1 \
@@ -28,7 +28,7 @@ COPY ./pyproject.toml ./poetry.lock ./
 RUN touch README.md
 RUN poetry install --without evaluation --no-root && rm -rf $POETRY_CACHE_DIR
 
-FROM python:3.12.3-slim AS runtime
+FROM python:3.12-slim as runtime
 
 WORKDIR /app
 
@@ -53,7 +53,6 @@ RUN useradd -l -m -u $OPENDEVIN_USER_ID -s /bin/bash opendevin && \
     usermod -aG sudo opendevin && \
     echo '%sudo ALL=(ALL) NOPASSWD:ALL' >> /etc/sudoers
 RUN chown -R opendevin:app /app && chmod -R 770 /app
-RUN sudo chown -R opendevin:app $WORKSPACE_BASE && sudo chmod -R 770 $WORKSPACE_BASE
 USER opendevin
 
 ENV VIRTUAL_ENV=/app/.venv \

containers/app/entrypoint.sh

@@ -22,23 +22,17 @@ if [[ "$SANDBOX_USER_ID" -eq 0 ]]; then
   echo "Running OpenDevin as root"
   export RUN_AS_DEVIN=false
   mkdir -p /root/.cache/ms-playwright/
-  if [ -d "/home/opendevin/.cache/ms-playwright/" ]; then
-    mv /home/opendevin/.cache/ms-playwright/ /root/.cache/
-  fi
+  mv /home/opendevin/.cache/ms-playwright/ /root/.cache/
   "$@"
 else
   echo "Setting up enduser with id $SANDBOX_USER_ID"
-  if id "enduser" &>/dev/null; then
-    echo "User enduser already exists. Skipping creation."
-  else
+  if ! useradd -l -m -u $SANDBOX_USER_ID -s /bin/bash enduser; then
+    echo "Failed to create user enduser with id $SANDBOX_USER_ID. Moving opendevin user."
+    incremented_id=$(($SANDBOX_USER_ID + 1))
+    usermod -u $incremented_id opendevin
     if ! useradd -l -m -u $SANDBOX_USER_ID -s /bin/bash enduser; then
-      echo "Failed to create user enduser with id $SANDBOX_USER_ID. Moving opendevin user."
-      incremented_id=$(($SANDBOX_USER_ID + 1))
-      usermod -u $incremented_id opendevin
-      if ! useradd -l -m -u $SANDBOX_USER_ID -s /bin/bash enduser; then
-        echo "Failed to create user enduser with id $SANDBOX_USER_ID for a second time. Exiting."
-        exit 1
-      fi
+      echo "Failed to create user enduser with id $SANDBOX_USER_ID for a second time. Exiting."
+      exit 1
     fi
   fi
   usermod -aG app enduser
@@ -52,11 +46,8 @@ else
     groupadd -g $DOCKER_SOCKET_GID docker
   fi
 
-  mkdir -p /home/enduser/.cache/huggingface/hub/
   mkdir -p /home/enduser/.cache/ms-playwright/
-  if [ -d "/home/opendevin/.cache/ms-playwright/" ]; then
-    mv /home/opendevin/.cache/ms-playwright/ /home/enduser/.cache/
-  fi
+  mv /home/opendevin/.cache/ms-playwright/ /home/enduser/.cache/
 
   usermod -aG $DOCKER_SOCKET_GID enduser
   echo "Running as enduser"

containers/build.sh

@@ -3,76 +3,65 @@ set -eo pipefail
 
 image_name=$1
 org_name=$2
-platform=$3
+push=0
+if [[ $3 == "--push" ]]; then
+  push=1
+fi
 
-echo "Building: $image_name for platform: $platform"
+echo -e "Building: $image_name"
 tags=()
 
 OPEN_DEVIN_BUILD_VERSION="dev"
 
+cache_tag_base="buildcache"
+cache_tag="$cache_tag_base"
+
 if [[ -n $GITHUB_REF_NAME ]]; then
   # check if ref name is a version number
   if [[ $GITHUB_REF_NAME =~ ^[0-9]+\.[0-9]+\.[0-9]+$ ]]; then
-    major_version=$(echo "$GITHUB_REF_NAME" | cut -d. -f1)
-    minor_version=$(echo "$GITHUB_REF_NAME" | cut -d. -f1,2)
-    tags+=("$major_version" "$minor_version")
-    tags+=("latest")
+    major_version=$(echo $GITHUB_REF_NAME | cut -d. -f1)
+    minor_version=$(echo $GITHUB_REF_NAME | cut -d. -f1,2)
+    tags+=($major_version $minor_version)
   fi
-  sanitized=$(echo "$GITHUB_REF_NAME" | sed 's/[^a-zA-Z0-9.-]\+/-/g')
+  sanitized=$(echo $GITHUB_REF_NAME | sed 's/[^a-zA-Z0-9.-]\+/-/g')
   OPEN_DEVIN_BUILD_VERSION=$sanitized
-  tag=$(echo "$sanitized" | tr '[:upper:]' '[:lower:]') # lower case is required in tagging
-  tags+=("$tag")
+  cache_tag+="-${sanitized}"
+  tags+=($sanitized)
 fi
 echo "Tags: ${tags[@]}"
 
-if [[ "$image_name" == "opendevin" ]]; then
-  dir="./containers/app"
-elif [[ "$image_name" == "od_runtime" ]]; then
-  dir="./containers/runtime"
-else
-  dir="./containers/$image_name"
-fi
-
-if [[ (! -f "$dir/Dockerfile") && "$image_name" != "od_runtime" ]]; then
-  # Allow runtime to be built without a Dockerfile
+dir=./containers/$image_name
+if [ ! -f $dir/Dockerfile ]; then
   echo "No Dockerfile found"
   exit 1
 fi
-if [[ ! -f "$dir/config.sh" ]]; then
+if [ ! -f $dir/config.sh ]; then
   echo "No config.sh found for Dockerfile"
   exit 1
 fi
-
-source "$dir/config.sh"
-
+source $dir/config.sh
 if [[ -n "$org_name" ]]; then
   DOCKER_ORG="$org_name"
 fi
-
-# If $DOCKER_IMAGE_TAG is set, add it to the tags
-if [[ -n "$DOCKER_IMAGE_TAG" ]]; then
-  tags+=("$DOCKER_IMAGE_TAG")
-fi
-
-DOCKER_REPOSITORY="$DOCKER_REGISTRY/$DOCKER_ORG/$DOCKER_IMAGE"
+DOCKER_REPOSITORY=$DOCKER_REGISTRY/$DOCKER_ORG/$DOCKER_IMAGE
 DOCKER_REPOSITORY=${DOCKER_REPOSITORY,,} # lowercase
 echo "Repo: $DOCKER_REPOSITORY"
 echo "Base dir: $DOCKER_BASE_DIR"
 
 args=""
-for tag in "${tags[@]}"; do
+for tag in ${tags[@]}; do
   args+=" -t $DOCKER_REPOSITORY:$tag"
 done
-
-output_image="/tmp/${image_name}_image_${platform}.tar"
+if [[ $push -eq 1 ]]; then
+  args+=" --push"
+  args+=" --cache-to=type=registry,ref=$DOCKER_REPOSITORY:$cache_tag,mode=max"
+fi
 
 docker buildx build \
   $args \
-  --build-arg OPEN_DEVIN_BUILD_VERSION="$OPEN_DEVIN_BUILD_VERSION" \
-  --platform linux/$platform \
+  --build-arg OPEN_DEVIN_BUILD_VERSION=$OPEN_DEVIN_BUILD_VERSION \
+  --cache-from=type=registry,ref=$DOCKER_REPOSITORY:$cache_tag \
+  --cache-from=type=registry,ref=$DOCKER_REPOSITORY:$cache_tag_base-main \
+  --platform linux/amd64,linux/arm64 \
   --provenance=false \
-  -f "$dir/Dockerfile" \
-  --output type=docker,dest="$output_image" \
-  "$DOCKER_BASE_DIR"
-
-echo "${tags[*]}" > tags.txt
+  -f $dir/Dockerfile $DOCKER_BASE_DIR

containers/runtime/README.md

@@ -1,11 +0,0 @@
-# Dynamic constructed Dockerfile
-
-This folder builds runtime image (sandbox), which will use a `Dockerfile` that is dynamically generated depends on the `base_image` AND a [Python source distribution](https://docs.python.org/3.10/distutils/sourcedist.html) that's based on the current commit of `opendevin`.
-
-The following command will generate Dockerfile for `ubuntu:22.04` and the source distribution `.tar` into `containers/runtime`.
-
-```bash
-poetry run python3 opendevin/runtime/utils/runtime_build.py \
-    --base_image ubuntu:22.04 \
-    --build_folder containers/runtime
-```

containers/runtime/config.sh

@@ -1,8 +0,0 @@
-DOCKER_REGISTRY=ghcr.io
-DOCKER_ORG=opendevin
-DOCKER_BASE_DIR="./containers/runtime"
-# These two variables will be appended by the runtime_build.py script
-# DOCKER_IMAGE=
-# DOCKER_IMAGE_TAG=
-DOCKER_IMAGE=od_runtime
-DOCKER_IMAGE_TAG=od_v0.8.1_image_ubuntu_tag_22.04

containers/sandbox/Dockerfile

@@ -16,11 +16,11 @@ RUN apt-get update && apt-get install -y \
     build-essential \
     openssh-server \
     sudo \
+    bash \
     gcc \
     jq \
     g++ \
     make \
-    iproute2 \
     && rm -rf /var/lib/apt/lists/*
 
 RUN mkdir -p -m0755 /var/run/sshd
@@ -28,17 +28,6 @@ RUN mkdir -p -m0755 /var/run/sshd
 # symlink python3 to python
 RUN ln -s /usr/bin/python3 /usr/bin/python
 
-# ==== OpenDevin Runtime Client ====
-RUN mkdir -p /opendevin && mkdir -p /opendevin/logs && chmod 777 /opendevin/logs
-RUN wget --progress=bar:force -O Miniforge3.sh "https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-$(uname)-$(uname -m).sh"
-RUN bash Miniforge3.sh -b -p /opendevin/miniforge3
-RUN chmod -R g+w /opendevin/miniforge3
-RUN bash -c ". /opendevin/miniforge3/etc/profile.d/conda.sh && conda config --set changeps1 False && conda config --append channels conda-forge"
-RUN echo "" > /opendevin/bash.bashrc
-RUN rm -f Miniforge3.sh
-
-# - agentskills dependencies
-RUN /opendevin/miniforge3/bin/pip install --upgrade pip
-RUN /opendevin/miniforge3/bin/pip install jupyterlab notebook jupyter_kernel_gateway flake8
-RUN /opendevin/miniforge3/bin/pip install python-docx PyPDF2 python-pptx pylatexenc openai
-RUN /opendevin/miniforge3/bin/pip install python-dotenv toml termcolor pydantic python-docx pyyaml docker pexpect tenacity e2b browsergym minio
+# install basic dependencies for CodeActAgent
+RUN pip3 install --upgrade pip
+RUN pip3 install jupyterlab notebook jupyter_kernel_gateway flake8

dev_config/python/mypy.ini

@@ -7,3 +7,5 @@ warn_unreachable = True
 warn_redundant_casts = True
 no_implicit_optional = True
 strict_optional = True
+
+exclude = agenthub/monologue_agent/regression

dev_config/python/ruff.toml

@@ -1,3 +1,7 @@
+exclude = [
+    "agenthub/monologue_agent/regression/",
+]
+
 [lint]
 select = [
     "E",
@@ -5,17 +9,10 @@ select = [
     "F",
     "I",
     "Q",
-    "B",
 ]
 
 ignore = [
     "E501",
-    "B003",
-    "B007",
-    "B009",
-    "B010",
-    "B904",
-    "B018",
 ]
 
 [lint.flake8-quotes]

docs/docusaurus.config.ts

@@ -8,8 +8,8 @@ const config: Config = {
   favicon: "img/logo.png",
 
   // Set the production url of your site here
-  url: "https://docs.all-hands.dev",
-  baseUrl: "/",
+  url: "https://OpenDevin.github.io",
+  baseUrl: "/OpenDevin/",
 
   // GitHub pages deployment config.
   organizationName: "OpenDevin",
@@ -23,13 +23,8 @@ const config: Config = {
   // useful metadata like html lang. For example, if your site is Chinese, you
   // may want to replace "en" with "zh-Hans".
   i18n: {
-    defaultLocale: 'en',
-    locales: ['en', 'fr', 'zh-Hans'],
-    localeConfigs: {
-      en: {
-        htmlLang: 'en-GB',
-      },
-    },
+    defaultLocale: "en",
+    locales: ["en"],
   },
 
   presets: [
@@ -56,6 +51,7 @@ const config: Config = {
       } satisfies Preset.Options,
     ],
   ],
+
   themeConfig: {
     image: "img/docusaurus.png",
     navbar: {
@@ -83,11 +79,44 @@ const config: Config = {
           label: "GitHub",
           position: "right",
         },
+      ],
+    },
+    footer: {
+      style: "dark",
+      links: [
         {
-          type: 'localeDropdown',
-          position: 'left',
+          title: "OpenDevin",
+          items: [
+            {
+              label: "Docs",
+              to: "/modules/usage/intro",
+            },
+          ],
+        },
+        {
+          title: "Community",
+          items: [
+            {
+              label: "Slack",
+              href: "https://join.slack.com/t/opendevin/shared_invite/zt-2ggtwn3k5-PvAA2LUmqGHVZ~XzGq~ILw"
+            },
+            {
+              label: "Discord",
+              href: "https://discord.gg/ESHStjSjD4",
+            },
+          ],
+        },
+        {
+          title: "More",
+          items: [
+            {
+              label: "GitHub",
+              href: "https://github.com/OpenDevin/OpenDevin",
+            },
+          ],
         },
       ],
+      copyright: `Copyright © ${new Date().getFullYear()} OpenDevin`,
     },
     prism: {
       theme: prismThemes.oneLight,

docs/i18n/fr/code.json

@@ -1,406 +0,0 @@
-{
-  "footer.title": {
-    "message": "OpenDevin"
-  },
-  "footer.docs": {
-    "message": "Documents"
-  },
-  "footer.community": {
-    "message": "Communauté"
-  },
-  "footer.copyright": {
-    "message": "© {year} OpenDevin"
-  },
-  "faq.title": {
-    "message": "Questions Fréquemment Posées",
-    "description": "FAQ Title"
-  },
-  "faq.description": {
-    "message": "Questions Fréquemment Posées"
-  },
-  "faq.section.title.1": {
-    "message": "Qu'est-ce qu'OpenDevin ?",
-    "description": "First Section Title"
-  },
-  "faq.section.highlight": {
-    "message": "OpenDevin",
-    "description": "Highlight Text"
-  },
-  "faq.section.description.1": {
-    "message": "est un ingénieur logiciel autonome qui peut résoudre des tâches d'ingénierie logicielle et de navigation web à tout moment. Il peut exécuter des requêtes en sciences des données, telles que \"Trouver le nombre de demandes de pull à l'repository OpenDevin dans les derniers mois\", et des tâches d'ingénierie logicielle, comme \"Veuillez ajouter des tests à ce fichier et vérifier si tous les tests passent. Si ce n'est pas le cas, réparez le fichier.\"",
-    "description": "Description for OpenDevin"
-  },
-  "faq.section.description.2": {
-    "message": "De plus, OpenDevin est une plateforme et communauté pour les développeurs d'agents qui souhaitent tester et évaluer de nouveaux agents.",
-    "description": "Further Description for OpenDevin"
-  },
-  "faq.section.title.2": {
-    "message": "Support",
-    "description": "Support Section Title"
-  },
-  "faq.section.support.answer": {
-    "message": "Si vous rencontrez un problème que d'autres utilisateurs peuvent également avoir, merci de le signaler sur {githubLink}. Si vous avez des difficultés à l'installation ou des questions générales, rejoignez-vous sur {discordLink} ou {slackLink}.",
-    "description": "Support Answer"
-  },
-  "faq.section.title.3": {
-    "message": "Comment résoudre un problème sur GitHub avec OpenDevin ?",
-    "description": "GitHub Issue Section Title"
-  },
-  "faq.section.github.steps.intro": {
-    "message": "Pour résoudre un problème sur GitHub en utilisant OpenDevin, envoyez une commande à OpenDevin demandant qu'il suit des étapes comme les suivantes :",
-    "description": "GitHub Steps Introduction"
-  },
-  "faq.section.github.step1": {
-    "message": "Lisez l'issue https://github.com/OpenDevin/OpenDevin/issues/1611",
-    "description": "GitHub Step 1"
-  },
-  "faq.section.github.step2": {
-    "message": "Cloner le dépôt et vérifier une nouvelle branche",
-    "description": "GitHub Step 2"
-  },
-  "faq.section.github.step3": {
-    "message": "Sur la base des instructions dans la description de l'issue, modifiez les fichiers pour résoudre le problème",
-    "description": "GitHub Step 3"
-  },
-  "faq.section.github.step4": {
-    "message": "Pousser le résultat à GitHub en utilisant la variable d'environnement GITHUB_TOKEN",
-    "description": "GitHub Step 4"
-  },
-  "faq.section.github.step5": {
-    "message": "Dites-moi le lien que je dois utiliser pour envoyer une demande de pull",
-    "description": "GitHub Step 5"
-  },
-  "faq.section.github.steps.preRun": {
-    "message": "Avant de lancer OpenDevin, vous pouvez faire :",
-    "description": "GitHub Steps Pre-Run"
-  },
-  "faq.section.github.steps.tokenInfo": {
-    "message": "où XXX est un jeton GitHub que vous avez créé et qui a les autorisations pour pousser dans le dépôt OpenDevin. Si vous n'avez pas d'autorisations de modification du dépôt OpenDevin, vous devrez peut-être changer cela en :",
-    "description": "GitHub Steps Token Info"
-  },
-  "faq.section.github.steps.usernameInfo": {
-    "message": "où USERNAME est votre nom GitHub.",
-    "description": "GitHub Steps Username Info"
-  },
-  "faq.section.title.4": {
-    "message": "Comment OpenDevin est-il différent de Devin ?",
-    "description": "Devin Section Title"
-  },
-  "faq.section.devin.linkText": {
-    "message": "Devin",
-    "description": "Devin Link Text"
-  },
-  "faq.section.devin.description": {
-    "message": "est un produit commercial par Cognition Inc., qui a servi d'inspiration initiale pour OpenDevin. Les deux visent à bien faire le travail d'ingénierie logicielle, mais vous pouvez télécharger, utiliser et modifier OpenDevin, tandis que Devin peut être utilisé uniquement via le site de Cognition. De plus, OpenDevin a évolué au-delà de l'inspiration initiale, et est maintenant un écosystème communautaire pour le développement d'agents en général, et nous serions ravis de vous voir rejoindre et",
-    "description": "Devin Description"
-  },
-  "faq.section.devin.contribute": {
-    "message": "contribuer",
-    "description": "Contribute Link"
-  },
-  "faq.section.title.5": {
-    "message": "Comment OpenDevin est-il différent de ChatGPT ?",
-    "description": "ChatGPT Section Title"
-  },
-  "faq.section.chatgpt.description": {
-    "message": "ChatGPT vous pouvez accéder en ligne, il ne se connecte pas aux fichiers locaux et ses capacités d'exécution du code sont limitées. Alors qu'il peut écrire du code, mais c'est difficile à tester ou à exécuter.",
-    "description": "ChatGPT Description"
-  },
-  "homepage.description": {
-    "message": "Génération d'code AI pour l'ingénierie logicielle.",
-    "description": "The homepage description"
-  },
-  "homepage.getStarted": {
-    "message": "Commencer"
-  },
-  "welcome.message": {
-    "message": "Bienvenue à OpenDevin, un système d'IA autonome ingénieur logiciel capable d'exécuter des tâches d'ingénierie complexes et de collaborer activement avec les utilisateurs sur les projets de développement logiciel."
-  },
-  "theme.ErrorPageContent.title": {
-    "message": "Cette page a planté.",
-    "description": "The title of the fallback page when the page crashed"
-  },
-  "theme.BackToTopButton.buttonAriaLabel": {
-    "message": "Retourner en haut de la page",
-    "description": "The ARIA label for the back to top button"
-  },
-  "theme.blog.archive.title": {
-    "message": "Archives",
-    "description": "The page & hero title of the blog archive page"
-  },
-  "theme.blog.archive.description": {
-    "message": "Archives",
-    "description": "The page & hero description of the blog archive page"
-  },
-  "theme.blog.paginator.navAriaLabel": {
-    "message": "Pagination des listes d'articles du blog",
-    "description": "The ARIA label for the blog pagination"
-  },
-  "theme.blog.paginator.newerEntries": {
-    "message": "Nouvelles entrées",
-    "description": "The label used to navigate to the newer blog posts page (previous page)"
-  },
-  "theme.blog.paginator.olderEntries": {
-    "message": "Anciennes entrées",
-    "description": "The label used to navigate to the older blog posts page (next page)"
-  },
-  "theme.blog.post.paginator.navAriaLabel": {
-    "message": "Pagination des articles du blog",
-    "description": "The ARIA label for the blog posts pagination"
-  },
-  "theme.blog.post.paginator.newerPost": {
-    "message": "Article plus récent",
-    "description": "The blog post button label to navigate to the newer/previous post"
-  },
-  "theme.blog.post.paginator.olderPost": {
-    "message": "Article plus ancien",
-    "description": "The blog post button label to navigate to the older/next post"
-  },
-  "theme.blog.post.plurals": {
-    "message": "Un article|{count} articles",
-    "description": "Pluralized label for \"{count} posts\". Use as much plural forms (separated by \"|\") as your language support (see https://www.unicode.org/cldr/cldr-aux/charts/34/supplemental/language_plural_rules.html)"
-  },
-  "theme.blog.tagTitle": {
-    "message": "{nPosts} tags avec « {tagName} »",
-    "description": "The title of the page for a blog tag"
-  },
-  "theme.tags.tagsPageLink": {
-    "message": "Voir tous les tags",
-    "description": "The label of the link targeting the tag list page"
-  },
-  "theme.colorToggle.ariaLabel": {
-    "message": "Basculer entre le mode sombre et clair (actuellement {mode})",
-    "description": "The ARIA label for the navbar color mode toggle"
-  },
-  "theme.colorToggle.ariaLabel.mode.dark": {
-    "message": "mode sombre",
-    "description": "The name for the dark color mode"
-  },
-  "theme.colorToggle.ariaLabel.mode.light": {
-    "message": "mode clair",
-    "description": "The name for the light color mode"
-  },
-  "theme.docs.breadcrumbs.navAriaLabel": {
-    "message": "Bouton de navigation des liens de la page",
-    "description": "The ARIA label for the breadcrumbs"
-  },
-  "theme.docs.DocCard.categoryDescription.plurals": {
-    "message": "1 élément|{count} éléments",
-    "description": "The default description for a category card in the generated index about how many items this category includes"
-  },
-  "theme.docs.paginator.navAriaLabel": {
-    "message": "Pages de documentation",
-    "description": "The ARIA label for the docs pagination"
-  },
-  "theme.docs.paginator.previous": {
-    "message": "Précédent",
-    "description": "The label used to navigate to the previous doc"
-  },
-  "theme.docs.paginator.next": {
-    "message": "Suivant",
-    "description": "The label used to navigate to the next doc"
-  },
-  "theme.docs.tagDocListPageTitle.nDocsTagged": {
-    "message": "Un document tagué|{count} documents tagués",
-    "description": "Pluralized label for \"{count} docs tagged\". Use as much plural forms (separated by \"|\") as your language support (see https://www.unicode.org/cldr/cldr-aux/charts/34/supplemental/language_plural_rules.html)"
-  },
-  "theme.docs.tagDocListPageTitle": {
-    "message": "{nDocsTagged} avec \"{tagName}\"",
-    "description": "The title of the page for a docs tag"
-  },
-  "theme.docs.versionBadge.label": {
-    "message": "Version: {versionLabel}"
-  },
-  "theme.docs.versions.unreleasedVersionLabel": {
-    "message": "Ceci est la documentation de la prochaine version {versionLabel} de {siteTitle}.",
-    "description": "The label used to tell the user that he's browsing an unreleased doc version"
-  },
-  "theme.docs.versions.unmaintainedVersionLabel": {
-    "message": "Ceci est la documentation de {siteTitle} {versionLabel}, qui n'est plus activement maintenue.",
-    "description": "The label used to tell the user that he's browsing an unmaintained doc version"
-  },
-  "theme.docs.versions.latestVersionSuggestionLabel": {
-    "message": "Pour une documentation à jour, consultez la {latestVersionLink} ({versionLabel}).",
-    "description": "The label used to tell the user to check the latest version"
-  },
-  "theme.docs.versions.latestVersionLinkLabel": {
-    "message": "dernière version",
-    "description": "The label used for the latest version suggestion link label"
-  },
-  "theme.common.editThisPage": {
-    "message": "Éditer cette page",
-    "description": "The link label to edit the current page"
-  },
-  "theme.common.headingLinkTitle": {
-    "message": "Lien direct vers {heading}",
-    "description": "Title for link to heading"
-  },
-  "theme.lastUpdated.atDate": {
-    "message": " le {date}",
-    "description": "The words used to describe on which date a page has been last updated"
-  },
-  "theme.lastUpdated.byUser": {
-    "message": " par {user}",
-    "description": "The words used to describe by who the page has been last updated"
-  },
-  "theme.lastUpdated.lastUpdatedAtBy": {
-    "message": "Dernière mise à jour{atDate}{byUser}",
-    "description": "The sentence used to display when a page has been last updated, and by who"
-  },
-  "theme.navbar.mobileVersionsDropdown.label": {
-    "message": "Versions",
-    "description": "The label for the navbar versions dropdown on mobile view"
-  },
-  "theme.NotFound.title": {
-    "message": "Page introuvable",
-    "description": "The title of the 404 page"
-  },
-  "theme.tags.tagsListLabel": {
-    "message": "Tags :",
-    "description": "The label alongside a tag list"
-  },
-  "theme.admonition.caution": {
-    "message": "prudence",
-    "description": "The default label used for the Caution admonition (:::caution)"
-  },
-  "theme.admonition.danger": {
-    "message": "danger",
-    "description": "The default label used for the Danger admonition (:::danger)"
-  },
-  "theme.admonition.info": {
-    "message": "information",
-    "description": "The default label used for the Info admonition (:::info)"
-  },
-  "theme.admonition.note": {
-    "message": "remarque",
-    "description": "The default label used for the Note admonition (:::note)"
-  },
-  "theme.admonition.tip": {
-    "message": "astuce",
-    "description": "The default label used for the Tip admonition (:::tip)"
-  },
-  "theme.admonition.warning": {
-    "message": "prudence",
-    "description": "The default label used for the Warning admonition (:::warning)"
-  },
-  "theme.AnnouncementBar.closeButtonAriaLabel": {
-    "message": "Fermer",
-    "description": "The ARIA label for close button of announcement bar"
-  },
-  "theme.blog.sidebar.navAriaLabel": {
-    "message": "Navigation vers les articles récents du blog",
-    "description": "The ARIA label for recent posts in the blog sidebar"
-  },
-  "theme.CodeBlock.copied": {
-    "message": "Copié",
-    "description": "The copied button label on code blocks"
-  },
-  "theme.CodeBlock.copyButtonAriaLabel": {
-    "message": "Copier le code",
-    "description": "The ARIA label for copy code blocks button"
-  },
-  "theme.CodeBlock.copy": {
-    "message": "Copier",
-    "description": "The copy button label on code blocks"
-  },
-  "theme.CodeBlock.wordWrapToggle": {
-    "message": "Activer/désactiver le retour à la ligne",
-    "description": "The title attribute for toggle word wrapping button of code block lines"
-  },
-  "theme.DocSidebarItem.expandCategoryAriaLabel": {
-    "message": "Développer la catégorie '{label}' de la barre latérale",
-    "description": "The ARIA label to expand the sidebar category"
-  },
-  "theme.DocSidebarItem.collapseCategoryAriaLabel": {
-    "message": "Réduire la catégorie '{label}' de la barre latérale",
-    "description": "The ARIA label to collapse the sidebar category"
-  },
-  "theme.NavBar.navAriaLabel": {
-    "message": "Main",
-    "description": "The ARIA label for the main navigation"
-  },
-  "theme.navbar.mobileLanguageDropdown.label": {
-    "message": "Langues",
-    "description": "The label for the mobile language switcher dropdown"
-  },
-  "theme.NotFound.p1": {
-    "message": "Nous n'avons pas trouvé ce que vous recherchez.",
-    "description": "The first paragraph of the 404 page"
-  },
-  "theme.NotFound.p2": {
-    "message": "Veuillez contacter le propriétaire du site qui vous a lié à l'URL d'origine et leur faire savoir que leur lien est cassé.",
-    "description": "The 2nd paragraph of the 404 page"
-  },
-  "theme.TOCCollapsible.toggleButtonLabel": {
-    "message": "Sur cette page",
-    "description": "The label used by the button on the collapsible TOC component"
-  },
-  "theme.blog.post.readMore": {
-    "message": "Lire plus",
-    "description": "The label used in blog post item excerpts to link to full blog posts"
-  },
-  "theme.blog.post.readMoreLabel": {
-    "message": "En savoir plus sur {title}",
-    "description": "The ARIA label for the link to full blog posts from excerpts"
-  },
-  "theme.blog.post.readingTime.plurals": {
-    "message": "Une minute de lecture|{readingTime} minutes de lecture",
-    "description": "Pluralized label for \"{readingTime} min read\". Use as much plural forms (separated by \"|\") as your language support (see https://www.unicode.org/cldr/cldr-aux/charts/34/supplemental/language_plural_rules.html)"
-  },
-  "theme.docs.breadcrumbs.home": {
-    "message": "Page d'accueil",
-    "description": "The ARIA label for the home page in the breadcrumbs"
-  },
-  "theme.docs.sidebar.collapseButtonTitle": {
-    "message": "Réduire le menu latéral",
-    "description": "The title attribute for collapse button of doc sidebar"
-  },
-  "theme.docs.sidebar.collapseButtonAriaLabel": {
-    "message": "Réduire le menu latérale",
-    "description": "The title attribute for collapse button of doc sidebar"
-  },
-  "theme.docs.sidebar.navAriaLabel": {
-    "message": "Barre de navigation latérale des docs",
-    "description": "The ARIA label for the sidebar navigation"
-  },
-  "theme.docs.sidebar.closeSidebarButtonAriaLabel": {
-    "message": "Fermer la barre de navigation",
-    "description": "The ARIA label for close button of mobile sidebar"
-  },
-  "theme.navbar.mobileSidebarSecondaryMenu.backButtonLabel": {
-    "message": "← Retour au menu principal",
-    "description": "The label of the back button to return to main menu, inside the mobile navbar sidebar secondary menu (notably used to display the docs sidebar)"
-  },
-  "theme.docs.sidebar.toggleSidebarButtonAriaLabel": {
-    "message": "Ouvrir/fermer la barre de navigation",
-    "description": "The ARIA label for hamburger menu button of mobile navigation"
-  },
-  "theme.docs.sidebar.expandButtonTitle": {
-    "message": "Déplier le menu latéral",
-    "description": "The ARIA label and title attribute for expand button of doc sidebar"
-  },
-  "theme.docs.sidebar.expandButtonAriaLabel": {
-    "message": "Déployer le menu latérale",
-    "description": "The ARIA label and title attribute for expand button of doc sidebar"
-  },
-  "theme.ErrorPageContent.tryAgain": {
-    "message": "Réessayer",
-    "description": "The label of the button to try again rendering when the React error boundary captures an error"
-  },
-  "theme.common.skipToMainContent": {
-    "message": "Aller directement au contenu principal",
-    "description": "The skip to content label used for accessibility, allowing to rapidly navigate to main content with keyboard tab/enter navigation"
-  },
-  "theme.tags.tagsPageTitle": {
-    "message": "Tags",
-    "description": "The title of the tag list page"
-  },
-  "theme.unlistedContent.title": {
-    "message": "Page non répertoriée",
-    "description": "The unlisted content banner title"
-  },
-  "theme.unlistedContent.message": {
-    "message": "Cette page n'est pas répertoriée. Les moteurs de recherche ne l'indexeront pas, et seuls les utilisateurs ayant un lien direct peuvent y accéder.",
-    "description": "The unlisted content banner message"
-  }
-}

docs/i18n/fr/docusaurus-plugin-content-blog/options.json

@@ -1,14 +0,0 @@
-{
-  "title": {
-    "message": "Blog",
-    "description": "The title for the blog used in SEO"
-  },
-  "description": {
-    "message": "Blog",
-    "description": "The description for the blog used in SEO"
-  },
-  "sidebar.title": {
-    "message": "Articles récents",
-    "description": "The label for the left sidebar"
-  }
-}

docs/i18n/fr/docusaurus-plugin-content-docs/current.json

@@ -1,18 +0,0 @@
-{
-  "version.label": {
-    "message": "Next",
-    "description": "The label for version current"
-  },
-  "sidebar.docsSidebar.category.🤖 Backends LLM": {
-    "message": "🤖 Backends LLM",
-    "description": "The label for category 🤖 Backends LLM in sidebar docsSidebar"
-  },
-  "sidebar.docsSidebar.category.🚧 Dépannage": {
-    "message": "🚧 Dépannage",
-    "description": "The label for category 🚧 Dépannage in sidebar docsSidebar"
-  },
-  "sidebar.apiSidebar.category.Backend": {
-    "message": "Backend",
-    "description": "The label for category Backend in sidebar apiSidebar"
-  }
-}

docs/i18n/fr/docusaurus-plugin-content-docs/current/python/python.md

@@ -1,3 +0,0 @@
-# Documentation Python
-
-Les documents apparaîtront ici après le déploiement.

docs/i18n/fr/docusaurus-plugin-content-docs/current/python/sidebar.json

@@ -1,5 +0,0 @@
-{
-  "items": ["python/python"],
-  "label": "Backend",
-  "type": "categorie"
-}

docs/i18n/fr/docusaurus-plugin-content-docs/current/usage/about.md

@@ -1,53 +0,0 @@
----
-sidebar_position: 7
----
-
-# 📚 Divers {#misc}
-
-## ⭐️ Stratégie de Recherche {#research-strategy}
-
-La réalisation d'une réplication complète des applications de production avec les LLM est une entreprise complexe. Notre stratégie implique :
-
-1. **Recherche Technique de Base :** Se concentrer sur la recherche fondamentale pour comprendre et améliorer les aspects techniques de la génération et de la gestion de code.
-2. **Compétences Spécialisées :** Améliorer l'efficacité des composants de base grâce à la curation des données, aux méthodes de formation, et plus encore.
-3. **Planification des Tâches :** Développer des capacités pour la détection de bogues, la gestion du code source et l'optimisation.
-4. **Évaluation :** Établir des métriques d'évaluation complètes pour mieux comprendre et améliorer nos modèles.
-
-## 🚧 Agent Par Défaut {#default-agent}
-
-- Notre agent par défaut est actuellement le CodeActAgent, capable de générer du code et de gérer des fichiers. Nous travaillons sur d'autres implémentations d'agents, y compris [SWE Agent](https://swe-agent.com/). Vous pouvez [lire à propos de notre ensemble actuel d'agents ici](./agents).
-
-## 🤝 Comment Contribuer {#how-to-contribute}
-
-OpenDevin est un projet communautaire, et nous accueillons les contributions de tout le monde. Que vous soyez développeur, chercheur, ou simplement enthousiaste à l'idée de faire progresser le domaine de l'ingénierie logicielle avec l'IA, il existe de nombreuses façons de vous impliquer :
-
-- **Contributions de Code :** Aidez-nous à développer les fonctionnalités de base, l'interface frontend ou les solutions de sandboxing.
-- **Recherche et Évaluation :** Contribuez à notre compréhension des LLM en ingénierie logicielle, participez à l'évaluation des modèles ou suggérez des améliorations.
-- **Retour d'Information et Tests :** Utilisez l'ensemble d'outils OpenDevin, signalez des bogues, suggérez des fonctionnalités ou fournissez des retours sur l'ergonomie.
-
-Pour plus de détails, veuillez consulter [ce document](https://github.com/OpenDevin/OpenDevin/blob/main/CONTRIBUTING.md).
-
-## 🤖 Rejoignez Notre Communauté {#join-our-community}
-
-Nous avons maintenant à la fois un espace de travail Slack pour la collaboration sur la construction d'OpenDevin et un serveur Discord pour discuter de tout ce qui est lié, par exemple, à ce projet, aux LLM, aux agents, etc.
-
-- [Espace de travail Slack](https://join.slack.com/t/opendevin/shared_invite/zt-2jsrl32uf-fTeeFjNyNYxqSZt5NPY3fA)
-- [Serveur Discord](https://discord.gg/ESHStjSjD4)
-
-Si vous souhaitez contribuer, n'hésitez pas à rejoindre notre communauté. Simplifions l'ingénierie logicielle ensemble !
-
-🐚 **Codez moins, créez plus avec OpenDevin.**
-
-[![Star History Chart](https://api.star-history.com/svg?repos=OpenDevin/OpenDevin&type=Date)](https://star-history.com/#OpenDevin/OpenDevin&Date)
-
-## 🛠️ Construit Avec {#built-with}
-
-OpenDevin est construit en utilisant une combinaison de cadres et de bibliothèques puissants, offrant une base robuste pour son développement. Voici les technologies clés utilisées dans le projet :
-
-![FastAPI](https://img.shields.io/badge/FastAPI-black?style=for-the-badge) ![uvicorn](https://img.shields.io/badge/uvicorn-black?style=for-the-badge) ![LiteLLM](https://img.shields.io/badge/LiteLLM-black?style=for-the-badge) ![Docker](https://img.shields.io/badge/Docker-black?style=for-the-badge) ![Ruff](https://img.shields.io/badge/Ruff-black?style=for-the-badge) ![MyPy](https://img.shields.io/badge/MyPy-black?style=for-the-badge) ![LlamaIndex](https://img.shields.io/badge/LlamaIndex-black?style=for-the-badge) ![React](https://img.shields.io/badge/React-black?style=for-the-badge)
-
-Veuillez noter que la sélection de ces technologies est en cours, et que des technologies supplémentaires peuvent être ajoutées ou des existantes supprimées au fur et à mesure de l'évolution du projet. Nous nous efforçons d'adopter les outils les plus adaptés et efficaces pour améliorer les capacités d'OpenDevin.
-
-## 📜 Licence {#license}
-
-Distribué sous la licence MIT. Voir [notre licence](https://github.com/OpenDevin/OpenDevin/blob/main/LICENSE) pour plus d'informations.

docs/i18n/fr/docusaurus-plugin-content-docs/current/usage/agents.md

@@ -1,98 +0,0 @@
----
-sidebar_position: 3
----
-
-# 🧠 Agents et Capacités
-
-## Agent CodeAct
-
-### Description
-
-Cet agent implémente l'idée CodeAct ([article](https://arxiv.org/abs/2402.01030), [tweet](https://twitter.com/xingyaow_/status/1754556835703751087)) qui consolide les **act**ions des agents LLM en un espace d'action **code** unifié pour à la fois la _simplicité_ et la _performance_ (voir article pour plus de détails).
-
-L'idée conceptuelle est illustrée ci-dessous. À chaque tour, l'agent peut :
-
-1. **Converse** : Communiquer avec les humains en langage naturel pour demander des clarifications, des confirmations, etc.
-2. **CodeAct** : Choisir d'accomplir la tâche en exécutant du code
-
-- Exécuter toute commande `bash` Linux valide
-- Exécuter tout code `Python` valide avec [un interpréteur Python interactif](https://ipython.org/). Cela est simulé à travers la commande `bash`, voir le système de plugin ci-dessous pour plus de détails.
-
-![image](https://github.com/OpenDevin/OpenDevin/assets/38853559/92b622e3-72ad-4a61-8f41-8c040b6d5fb3)
-
-### Système de Plugin
-
-Pour rendre l'agent CodeAct plus puissant avec seulement l'accès à l'espace d'action `bash`, l'agent CodeAct exploite le système de plugins d'OpenDevin:
-
-- [Plugin Jupyter](https://github.com/OpenDevin/OpenDevin/tree/main/opendevin/runtime/plugins/jupyter) : pour l'exécution d'IPython via la commande bash
-- [Plugin outil agent SWE](https://github.com/OpenDevin/OpenDevin/tree/main/opendevin/runtime/plugins/swe_agent_commands) : Outils de ligne de commande bash puissants pour les tâches de développement logiciel introduits par [swe-agent](https://github.com/princeton-nlp/swe-agent).
-
-### Démonstration
-
-https://github.com/OpenDevin/OpenDevin/assets/38853559/f592a192-e86c-4f48-ad31-d69282d5f6ac
-
-_Exemple de CodeActAgent avec `gpt-4-turbo-2024-04-09` effectuant une tâche de science des données (régression linéaire)_
-
-### Actions
-
-`Action`,
-`CmdRunAction`,
-`IPythonRunCellAction`,
-`AgentEchoAction`,
-`AgentFinishAction`,
-`AgentTalkAction`
-
-### Observations
-
-`CmdOutputObservation`,
-`IPythonRunCellObservation`,
-`AgentMessageObservation`,
-`UserMessageObservation`
-
-### Méthodes
-
-| Méthode          | Description                                                                                                                                     |
-| ---------------- | ----------------------------------------------------------------------------------------------------------------------------------------------- |
-| `__init__`       | Initialise un agent avec `llm` et une liste de messages `list[Mapping[str, str]]`                                                                |
-| `step`           | Effectue une étape en utilisant l'agent CodeAct. Cela inclut la collecte d'informations sur les étapes précédentes et invite le modèle à exécuter une commande. |
-
-### En cours de réalisation & prochaine étape
-
-[] Support de la navigation sur le web
-[] Compléter le workflow pour l'agent CodeAct afin de soumettre des PRs Github
-
-## Agent Planificateur
-
-### Description
-
-L'agent planificateur utilise une stratégie d'incitation spéciale pour créer des plans à long terme pour résoudre les problèmes.
-L'agent reçoit ses paires action-observation précédentes, la tâche actuelle, et un indice basé sur la dernière action effectuée à chaque étape.
-
-### Actions
-
-`NullAction`,
-`CmdRunAction`,
-`BrowseURLAction`,
-`GithubPushAction`,
-`FileReadAction`,
-`FileWriteAction`,
-`AgentThinkAction`,
-`AgentFinishAction`,
-`AgentSummarizeAction`,
-`AddTaskAction`,
-`ModifyTaskAction`,
-
-### Observations
-
-`Observation`,
-`NullObservation`,
-`CmdOutputObservation`,
-`FileReadObservation`,
-`BrowserOutputObservation`
-
-### Méthodes
-
-| Méthode          | Description                                                                                                                                                                               |
-| ---------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `__init__`       | Initialise un agent avec `llm`                                                                                                                                                           |
-| `step`           | Vérifie si l'étape actuelle est terminée, retourne `AgentFinishAction` si oui. Sinon, crée une incitation de planification et l'envoie au modèle pour inférence, en ajoutant le résultat comme prochaine action. |

docs/i18n/fr/docusaurus-plugin-content-docs/current/usage/architecture.mdx

@@ -1,50 +0,0 @@
----
-sidebar_position: 4
----
-
-# 🏛️ Aperçu de l'Architecture Système
-
-Voici un aperçu de haut niveau de l'architecture du système. Le système est divisé en deux composants principaux : le frontend et le backend. Le frontend est responsable de la gestion des interactions avec l'utilisateur et de l'affichage des résultats. Le backend est responsable de la gestion de la logique métier et de l'exécution des agents.
-
-![system_architecture.svg](/img/system_architecture.svg)
-
-Cet aperçu est simplifié pour montrer les principaux composants et leurs interactions. Pour une vue plus détaillée de l'architecture du backend, consultez la section [Architecture du Backend](#backend-architecture-fr).
-
-# Architecture du Backend {#backend-architecture-fr}
-
-_**Avertissement**: L'architecture du backend est en cours de développement et est sujette à modifications. Le schéma suivant montre l'architecture actuelle du backend basée sur le commit indiqué dans le pied de page du schéma._
-
-![backend_architecture.svg](/img/backend_architecture.svg)
-
-<details>
-  <summary>Mise à jour de ce Schéma</summary>
-  <div>
-    La génération du schéma d'architecture du backend est partiellement automatisée.
-    Le schéma est généré à partir des annotations de type dans le code en utilisant l'outil py2puml.
-    Le schéma est ensuite revu manuellement, ajusté et exporté en PNG et SVG.
-
-    ## Prérequis
-
-    - Un environnement Python dans lequel opendevin est exécutable
-    (selon les instructions du fichier README.md à la racine du dépôt)
-    - [py2puml](https://github.com/lucsorel/py2puml) installé
-
-## Étapes
-
-1.  Générez automatiquement le schéma en exécutant la commande suivante depuis la racine du dépôt :
-    `py2puml opendevin opendevin > docs/architecture/backend_architecture.puml`
-
-2.  Ouvrez le fichier généré dans un éditeur PlantUML, par exemple Visual Studio Code avec l'extension PlantUML ou [PlantText](https://www.planttext.com/)
-
-3.  Révisez le PUML généré et apportez toutes les modifications nécessaires au schéma (ajoutez les parties manquantes, corrigez les erreurs, améliorez l'agencement).
-    _py2puml crée le schéma à partir des annotations de type dans le code, donc les annotations de type manquantes ou incorrectes peuvent entraîner un schéma incomplet ou incorrect._
-
-4.  Examinez la différence entre le nouveau schéma et le précédent et vérifiez manuellement si les modifications sont correctes.
-    _Assurez-vous de ne pas supprimer les parties ajoutées manuellement au schéma par le passé et qui sont toujours pertinentes._
-
-5.  Ajoutez le hash du commit qui a été utilisé pour générer le schéma dans le pied de page du schéma.
-
-6.  Exporte le schéma sous forme de fichiers PNG et SVG et remplacez les schémas existants dans le répertoire `docs/architecture`. Cela peut être fait avec (par exemple [PlantText](https://www.planttext.com/))
-
-  </div>
-</details>

docs/i18n/fr/docusaurus-plugin-content-docs/current/usage/custom_sandbox_guide.md

@@ -1,107 +0,0 @@
-# 💿 Comment Créer un Soutien Docker sur Mesure
-
-Le sandbox par défaut OpenDevin est équipé d'une configuration ubuntu minimaliste. Votre cas d'utilisation pourrait nécessiter des logiciels installés par défaut. Cet article vous enseignera comment réaliser cela en utilisant une image docker personnalisée.
-
-## Configuration
-
-Assurez-vous de pouvoir utiliser OpenDevin en suivant la documentation [Development.md](https://github.com/OpenDevin/OpenDevin/blob/main/Development.md).
-
-## Créer Votre Image Docker
-
-Ensuite, vous devez créer votre image docker personnalisée qui doit être basée sur debian/ubuntu. Par exemple, si nous souhaitons que OpenDevin ait accès au "node" binaire, nous utiliserions ce Dockerfile:
-
-```bash
-# Commencez avec l'image ubuntu la plus récente
-FROM ubuntu:latest
-
-# Effectuez les mises à jour nécessaires
-RUN apt-get update && apt-get install
-
-# Installez nodejs
-RUN apt-get install -y nodejs
-```
-
-Ensuite, construisez votre image docker avec le nom de votre choix. Par exemple "image_personnalisée". Pour cela, créez un répertoire et placez le fichier à l'intérieur avec le nom "Dockerfile", puis dans le répertoire exécutez cette commande:
-
-```bash
-docker build -t image_personnalisée .
-```
-
-Cela produira une nouvelle image appelée ```image_personnalisée``` qui sera disponible dans Docker Engine.
-
-> Remarque: Dans la configuration décrite ici, OpenDevin va fonctionner en tant que utilisateur "opendevin" à l'intérieur du sandbox et donc tous les packages installés via le Dockerfile seront disponibles pour tous les utilisateurs sur le système, pas seulement root.
->
-> L'installation avec apt-get ci-dessus installe nodejs pour tous les utilisateurs.
-
-## Spécifiez votre image personnalisée dans le fichier config.toml
-
-La configuration OpenDevin se fait via le fichier de niveau supérieur ```config.toml``` .
-Créez un fichier ```config.toml``` dans le répertoire OpenDevin et entrez ces contenus:
-
-```toml
-[core]
-workspace_base="./workspace"
-persist_sandbox=false
-run_as_devin=true
-sandbox_container_image="image_personnalisée"
-```
-
-> Assurez-vous que ```sandbox_container_image``` est défini sur le nom de votre image personnalisée précédente.
-
-## Exécution
-
-Exécutez OpenDevin en exécutant ```make run``` dans le répertoire racine.
-
-Naviguez vers ```localhost:3001``` et vérifiez si vos dépendances souhaitées sont disponibles.
-
-Dans le cas de l'exemple ci-dessus, la commande ```node -v``` dans la console produit ```v18.19.1```
-
-Félicitations !
-
-## Explication technique
-
-Le code pertinent est défini dans [ssh_box.py](https://github.com/OpenDevin/OpenDevin/blob/main/opendevin/runtime/docker/ssh_box.py) et [image_agnostic_util.py](https://github.com/OpenDevin/OpenDevin/blob/main/opendevin/runtime/docker/image_agnostic_util.py).
-
-En particulier, ssh_box.py vérifie l'objet config pour ```config.sandbox_container_image``` et ensuite tente de récupérer l'image à l'aide de [get_od_sandbox_image](https://github.com/OpenDevin/OpenDevin/blob/main/opendevin/runtime/docker/image_agnostic_util.py#L72), qui est défini dans image_agnostic_util.py.
-
-Lorsqu'une image personnalisée est utilisée pour la première fois, elle ne sera pas trouvée et donc elle sera construite (à l'exécution ultérieure, l'image construite sera trouvée et renvoyée).
-
-L'image personnalisée est construite avec [_build_sandbox_image()](https://github.com/OpenDevin/OpenDevin/blob/main/opendevin/runtime/docker/image_agnostic_util.py#L29), qui crée un fichier docker en utilisant votre image personnalisée comme base et configure ensuite l'environnement pour OpenDevin, comme ceci:
-
-```python
-dockerfile_content = (
-        f'FROM {base_image}\n'
-        'RUN apt update && apt install -y openssh-server wget sudo\n'
-        'RUN mkdir -p -m0755 /var/run/sshd\n'
-        'RUN mkdir -p /opendevin && mkdir -p /opendevin/logs && chmod 777 /opendevin/logs\n'
-        'RUN wget "https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-$(uname)-$(uname -m).sh"\n'
-        'RUN bash Miniforge3-$(uname)-$(uname -m).sh -b -p /opendevin/miniforge3\n'
-        'RUN bash -c ". /opendevin/miniforge3/etc/profile.d/conda.sh && conda config --set changeps1 False && conda config --append channels conda-forge"\n'
-        'RUN echo "export PATH=/opendevin/miniforge3/bin:$PATH" >> ~/.bashrc\n'
-        'RUN echo "export PATH=/opendevin/miniforge3/bin:$PATH" >> /opendevin/bash.bashrc\n'
-    ).strip()
-```
-
-> Remarque: Le nom de l'image est modifié via [_get_new_image_name()](https://github.com/OpenDevin/OpenDevin/blob/main/opendevin/runtime/docker/image_agnostic_util.py#L63) et c'est ce nom modifié qui sera recherché lors des exécutions ultérieures.
-
-## Dépannage / Erreurs
-
-### Erreur: ```useradd: UID 1000 est non unique```
-Si vous voyez cette erreur dans la sortie de la console, il s'agit du fait que OpenDevin essaie de créer le utilisateur opendevin dans le sandbox avec un ID d'utilisateur de 1000, cependant cet ID d'utilisateur est déjà utilisé dans l'image (pour une raison inconnue). Pour résoudre ce problème, changez la valeur du champ sandbox_user_id dans le fichier config.toml en une valeur différente:
-
-```toml
-[core]
-workspace_base="./workspace"
-persist_sandbox=false
-run_as_devin=true
-sandbox_container_image="image_personnalisée"
-sandbox_user_id="1001"
-```
-
-### Erreurs de port d'utilisation
-
-Si vous voyez un message d'erreur indiquant que le port est utilisé ou indisponible, essayez de supprimer toutes les containers docker en cours d'exécution (exécutez `docker ps` et `docker rm` des containers concernés) puis ré-exécutez ```make run```
-
-## Discuter
-
-Pour d'autres problèmes ou questions rejoignez le [Slack](https://join.slack.com/t/opendevin/shared_invite/zt-2jsrl32uf-fTeeFjNyNYxqSZt5NPY3fA) ou le [Discord](https://discord.gg/ESHStjSjD4) et demandez!

docs/i18n/fr/docusaurus-plugin-content-docs/current/usage/feedback.md

@@ -1,18 +0,0 @@
----
-sidebar_position: 6
----
-
-# ✅ Fournir des Commentaires
-
-Lorsque vous utilisez OpenDevin, vous rencontrerez sans aucun doute des cas où les choses fonctionnent bien et d'autres où elles ne fonctionnent pas. Nous vous encourageons à fournir des commentaires lorsque vous utilisez OpenDevin pour aider l'équipe de développement et, peut-être plus important encore, créer un corpus ouvert d'exemples de formation pour les agents de codage -- Partagez-OpenDevin !
-
-## 📝 Comment Fournir des Commentaires
-
-Fournir des commentaires est simple ! Lorsque vous utilisez OpenDevin, vous pouvez appuyer sur le bouton de pouce vers le haut ou vers le bas à n'importe quel moment de votre interaction. Vous serez invité à fournir votre adresse email (par exemple, afin que nous puissions vous contacter si nous voulons poser des questions de suivi), et vous pouvez choisir si vous souhaitez fournir des commentaires publiquement ou en privé.
-
-<iframe width="560" height="315" src="https://www.youtube.com/embed/5rFx-StMVV0?si=svo7xzp6LhGK_GXr" title="Lecteur vidéo YouTube" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" referrerpolicy="strict-origin-when-cross-origin" allowfullscreen></iframe>
-
-## 📜 Licence de Données et Confidentialité
-
-* Les données **publiques** seront distribuées sous la licence MIT, comme OpenDevin lui-même, et pourront être utilisées par la communauté pour former et tester des modèles. Évidemment, les commentaires que vous pouvez rendre publics seront plus précieux pour la communauté dans son ensemble, donc lorsque vous ne traitez pas d'informations sensibles, nous vous encourageons à choisir cette option !
-* Les données **privées** ne seront partagées qu'avec l'équipe OpenDevin dans le but d'améliorer OpenDevin.

docs/i18n/fr/docusaurus-plugin-content-docs/current/usage/intro.mdx

@@ -1,111 +0,0 @@
----
-sidebar_position: 1
----
-
-# 💻 OpenDevin
-
-OpenDevin est un **ingénieur logiciel IA autonome** capable d'exécuter des tâches d'ingénierie complexes et de collaborer activement avec les utilisateurs sur des projets de développement logiciel.
-Ce projet est entièrement open-source, vous pouvez donc l'utiliser et le modifier comme bon vous semble.
-
-:::tip
-Explorez le code source d'OpenDevin sur [GitHub](https://github.com/OpenDevin/OpenDevin) ou rejoignez l'une de nos communautés !
-
-<a href="https://github.com/OpenDevin/OpenDevin/graphs/contributors">
-  <img
-    src="https://img.shields.io/github/contributors/opendevin/opendevin?style=for-the-badge"
-    alt="Contributors"
-  />
-</a>
-<a href="https://github.com/OpenDevin/OpenDevin/network/members">
-  <img
-    src="https://img.shields.io/github/forks/opendevin/opendevin?style=for-the-badge"
-    alt="Forks"
-  />
-</a>
-<a href="https://github.com/OpenDevin/OpenDevin/stargazers">
-  <img
-    src="https://img.shields.io/github/stars/opendevin/opendevin?style=for-the-badge"
-    alt="Stargazers"
-  />
-</a>
-<a href="https://github.com/OpenDevin/OpenDevin/issues">
-  <img
-    src="https://img.shields.io/github/issues/opendevin/opendevin?style=for-the-badge"
-    alt="Issues"
-  />
-</a>
-<br></br>
-<a href="https://github.com/OpenDevin/OpenDevin/blob/main/LICENSE">
-  <img
-    src="https://img.shields.io/github/license/opendevin/opendevin?style=for-the-badge"
-    alt="MIT License"
-  />
-</a>
-<br></br>
-<a href="https://join.slack.com/t/opendevin/shared_invite/zt-2jsrl32uf-fTeeFjNyNYxqSZt5NPY3fA">
-  <img
-    src="https://img.shields.io/badge/Slack-Join%20Us-red?logo=slack&logoColor=white&style=for-the-badge"
-    alt="Join our Slack community"
-  />
-</a>
-<a href="https://discord.gg/ESHStjSjD4">
-  <img
-    src="https://img.shields.io/badge/Discord-Join%20Us-purple?logo=discord&logoColor=white&style=for-the-badge"
-    alt="Join our Discord community"
-  />
-</a>
-:::
-
-## 🛠️ Pour commencer
-
-La manière la plus simple d'exécuter OpenDevin est à l'intérieur d'un conteneur Docker. Il fonctionne mieux avec la version la plus récente de Docker, `26.0.0`.
-Vous devez utiliser Linux, Mac OS ou WSL sur Windows.
-
-Pour démarrer OpenDevin dans un conteneur docker, exécutez les commandes suivantes dans votre terminal :
-
-:::warning
-Lorsque vous exécutez la commande suivante, les fichiers dans `./workspace` peuvent être modifiés ou supprimés.
-:::
-
-```bash
-WORKSPACE_BASE=$(pwd)/workspace
-docker run -it \
-    --pull=always \
-    -e SANDBOX_USER_ID=$(id -u) \
-    -e PERSIST_SANDBOX="true" \
-    -e SSH_PASSWORD="make something up here" \
-    -e WORKSPACE_MOUNT_PATH=$WORKSPACE_BASE \
-    -v $WORKSPACE_BASE:/opt/workspace_base \
-    -v /var/run/docker.sock:/var/run/docker.sock \
-    -p 3000:3000 \
-    --add-host host.docker.internal:host-gateway \
-    --name opendevin-app-$(date +%Y%m%d%H%M%S) \
-    ghcr.io/opendevin/opendevin:0.6
-```
-
-Vous trouverez OpenDevin fonctionnant à l'adresse [http://localhost:3000](http://localhost:3000) avec accès à `./workspace`. Pour qu'OpenDevin fonctionne sur votre code, placez-le dans `./workspace`.
-
-OpenDevin n'aura accès qu'à ce dossier de workspace. Le reste de votre système ne sera pas affecté car il s'exécute dans un bac à sable sécurisé de docker.
-
-:::tip
-Si vous souhaitez utiliser la version **(instable !)** la plus récente, vous pouvez utiliser `ghcr.io/opendevin/opendevin:main` comme image (dernière ligne).
-:::
-
-Pour le workflow de développement, consultez [Development.md](https://github.com/OpenDevin/OpenDevin/blob/main/Development.md).
-
-Avez-vous des problèmes ? Consultez notre [Guide de dépannage](https://docs.all-hands.dev/modules/usage/troubleshooting).
-
-:::warning
-OpenDevin est actuellement en cours de développement, mais vous pouvez déjà exécuter la version alpha pour voir le système de bout en bout en action.
-:::
-
-[contributors-shield]: https://img.shields.io/github/contributors/opendevin/opendevin?style=for-the-badge
-[contributors-url]: https://github.com/OpenDevin/OpenDevin/graphs/contributors
-[forks-shield]: https://img.shields.io/github/forks/opendevin/opendevin?style=for-the-badge
-[forks-url]: https://github.com/OpenDevin/OpenDevin/network/members
-[stars-shield]: https://img.shields.io/github/stars/opendevin/opendevin?style=for-the-badge
-[stars-url]: https://github.com/OpenDevin/OpenDevin/stargazers
-[issues-shield]: https://img.shields.io/github/issues/opendevin/opendevin?style=for-the-badge
-[issues-url]: https://github.com/OpenDevin/OpenDevin/issues
-[license-shield]: https://img.shields.io/github/license/opendevin/opendevin?style=for-the-badge
-[license-url]: https://github.com/OpenDevin/OpenDevin/blob/main/LICENSE

docs/i18n/fr/docusaurus-plugin-content-docs/current/usage/llms/azureLLMs.md

@@ -1,37 +0,0 @@
-# Azure OpenAI LLM
-
-## Complétion
-
-OpenDevin utilise LiteLLM pour les appels de complétion. Vous pouvez trouver leur documentation sur Azure [ici](https://docs.litellm.ai/docs/providers/azure)
-
-### Configurations openai Azure
-
-Lors de l'exécution de l'image Docker OpenDevin, vous devrez définir les variables d'environnement suivantes en utilisant `-e` :
-
-```
-LLM_BASE_URL="<azure-api-base-url>"          # e.g. "https://openai-gpt-4-test-v-1.openai.azure.com/"
-LLM_API_KEY="<azure-api-key>"
-LLM_MODEL="azure/<your-gpt-deployment-name>"
-LLM_API_VERSION = "<api-version>"          # e.g. "2024-02-15-preview"
-```
-
-:::note
-Vous pouvez trouver le nom de votre déploiement ChatGPT sur la page des déploiements sur Azure. Par défaut ou initialement, il pourrait être le même que le nom du modèle de chat (par exemple 'GPT4-1106-preview'), mais il n'est pas obligé de l'être. Exécutez OpenDevin, et une fois chargé dans le navigateur, allez dans Paramètres et définissez le modèle comme suit : "azure/&lt;your-actual-gpt-deployment-name&gt;". Si ce n'est pas dans la liste, entrez votre propre texte et enregistrez-le.
-:::
-
-## Embeddings
-
-OpenDevin utilise llama-index pour les embeddings. Vous pouvez trouver leur documentation sur Azure [ici](https://docs.llamaindex.ai/en/stable/api_reference/embeddings/azure_openai/)
-
-### Configurations openai Azure
-
-Le modèle utilisé pour les embeddings Azure OpenAI est "text-embedding-ada-002".
-Vous avez besoin du nom de déploiement correct pour ce modèle dans votre compte Azure.
-
-Lors de l'exécution d'OpenDevin dans Docker, définissez les variables d'environnement suivantes en utilisant `-e` :
-
-```
-LLM_EMBEDDING_MODEL="azureopenai"
-LLM_EMBEDDING_DEPLOYMENT_NAME = "<your-embedding-deployment-name>"        # e.g. "TextEmbedding...<etc>"
-LLM_API_VERSION = "<api-version>"         # e.g. "2024-02-15-preview"
-```