Resolve merge conflict in agent_session.py

Co-authored-by: openhands <openhands@all-hands.dev>

CODE_OF_CONDUCT.md

@@ -18,24 +18,24 @@ diverse, inclusive, and healthy community.
 Examples of behavior that contributes to a positive environment for our
 community include:
 
-* Demonstrating empathy and kindness toward other people
-* Being respectful of differing opinions, viewpoints, and experiences
-* Giving and gracefully accepting constructive feedback
+* Demonstrating empathy and kindness toward other people.
+* Being respectful of differing opinions, viewpoints, and experiences.
+* Giving and gracefully accepting constructive feedback.
 * Accepting responsibility and apologizing to those affected by our mistakes,
-  and learning from the experience
+  and learning from the experience.
 * Focusing on what is best not just for us as individuals, but for the overall
-  community
+  community.
 
 Examples of unacceptable behavior include:
 
 * The use of sexualized language or imagery, and sexual attention or advances of
-  any kind
-* Trolling, insulting or derogatory comments, and personal or political attacks
-* Public or private harassment
+  any kind.
+* Trolling, insulting or derogatory comments, and personal or political attacks.
+* Public or private harassment.
 * Publishing others' private information, such as a physical or email address,
-  without their explicit permission
+  without their explicit permission.
 * Other conduct which could reasonably be considered inappropriate in a
-  professional setting
+  professional setting.
 
 ## Enforcement Responsibilities
 
@@ -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@all-hands.dev.
 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

@@ -11,11 +11,11 @@ To understand the codebase, please refer to the README in each module:
    - [agenthub](./openhands/agenthub/README.md)
    - [server](./openhands/server/README.md)
 
-## Setting up your development environment
+## Setting up Your Development Environment
 
 We have a separate doc [Development.md](https://github.com/All-Hands-AI/OpenHands/blob/main/Development.md) that tells you how to set up a development workflow.
 
-## How can I contribute?
+## How Can I Contribute?
 
 There are many ways that you can contribute:
 
@@ -23,7 +23,7 @@ There are many ways that you can contribute:
 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](#sending-pull-requests-to-openhands) (see details below). In particular, we have some [good first issues](https://github.com/All-Hands-AI/OpenHands/labels/good%20first%20issue) that may be ones to start on.
 
-## What can I build?
+## What Can I Build?
 Here are a few ways you can help improve the codebase.
 
 #### UI/UX
@@ -35,7 +35,7 @@ of the application, please open an issue first, or better, join the #frontend ch
 to gather consensus from our design team first.
 
 #### Improving the agent
-Our main agent is the CodeAct agent. You can [see its prompts here](https://github.com/All-Hands-AI/OpenHands/tree/main/openhands/agenthub/codeact_agent)
+Our main agent is the CodeAct agent. You can [see its prompts here](https://github.com/All-Hands-AI/OpenHands/tree/main/openhands/agenthub/codeact_agent).
 
 Changes to these prompts, and to the underlying behavior in Python, can have a huge impact on user experience.
 You can try modifying the prompts to see how they change the behavior of the agent as you use the app
@@ -63,7 +63,7 @@ At the moment, we have two kinds of tests: [`unit`](./tests/unit) and [`integrat
 ## Sending Pull Requests to OpenHands
 
 You'll need to fork our repository to send us a Pull Request. You can learn more
-about how to fork a GitHub repo and open a PR with your changes in [this article](https://medium.com/swlh/forks-and-pull-requests-how-to-contribute-to-github-repos-8843fac34ce8)
+about how to fork a GitHub repo and open a PR with your changes in [this article](https://medium.com/swlh/forks-and-pull-requests-how-to-contribute-to-github-repos-8843fac34ce8).
 
 ### 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:
@@ -103,7 +103,7 @@ Further, if you see an issue you like, please leave a "thumbs-up" or a comment,
 
 ### Making Pull Requests
 
-We're generally happy to consider all [PRs](https://github.com/All-Hands-AI/OpenHands/pulls), with the evaluation process varying based on the type of change:
+We're generally happy to consider all pull requests with the evaluation process varying based on the type of change:
 
 #### For Small Improvements
 

Development.md

@@ -3,7 +3,7 @@ This guide is for people working on OpenHands and editing the source code.
 If you wish to contribute your changes, check out the [CONTRIBUTING.md](https://github.com/All-Hands-AI/OpenHands/blob/main/CONTRIBUTING.md) on how to clone and setup the project initially before moving on.
 Otherwise, you can clone the OpenHands project directly.
 
-## Start the server for development
+## 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!)
@@ -58,7 +58,7 @@ See [our documentation](https://docs.all-hands.dev/modules/usage/llms) for recom
 
 ### 4. Running the application
 #### Option A: Run the Full Application
-Once the setup is complete, launching OpenHands is as simple as running a single command. This command starts both the backend and frontend servers seamlessly, allowing you to interact with OpenHands:
+Once the setup is complete, this command starts both the backend and frontend servers, allowing you to interact with OpenHands:
 ```bash
 make run
 ```
@@ -75,11 +75,11 @@ 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.
-OpenHands 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, export DEBUG=1 in the environment and restart the backend.
+OpenHands will 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 OpenHands.
+Need help or info on available targets and commands? Use the help command for all the guidance you need with OpenHands.
 ```bash
 make help
  ```
@@ -93,14 +93,14 @@ poetry run pytest ./tests/unit/test_*.py
 ```
 
 ### 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`
+1. Add your dependency in `pyproject.toml` or use `poetry add xxx`.
+2. Update the poetry.lock file via `poetry lock --no-update`.
 
 ### 9. Use existing Docker image
 To reduce build time (e.g., if no changes were made to the client-runtime component), you can use an existing Docker container image by
 setting the SANDBOX_RUNTIME_CONTAINER_IMAGE environment variable to the desired Docker image.
 
-Example: `export SANDBOX_RUNTIME_CONTAINER_IMAGE=ghcr.io/all-hands-ai/runtime:0.18-nikolaik`
+Example: `export SANDBOX_RUNTIME_CONTAINER_IMAGE=ghcr.io/all-hands-ai/runtime:0.19-nikolaik`
 
 ## Develop inside Docker container
 
@@ -110,7 +110,7 @@ TL;DR
 make docker-dev
 ```
 
-See more details [here](./containers/dev/README.md)
+See more details [here](./containers/dev/README.md).
 
 If you are just interested in running `OpenHands` without installing all the required tools on your host.
 

ISSUE_TRIAGE.md

@@ -2,8 +2,8 @@
 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.)
+* 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 or affecting single user.
@@ -11,10 +11,10 @@ These are the procedures and guidelines on how issues are triaged in this repo b
 * **Critical**: Affecting all users or potential security issues.
 
 ## Effort
-* Issues may be estimated with effort required (**small effort**, **medium effort**, **large 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**
+* 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.

README.md

@@ -43,17 +43,17 @@ See the [Installation](https://docs.all-hands.dev/modules/usage/installation) gu
 system requirements and more information.
 
 ```bash
-docker pull docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik
+docker pull docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik
 
 docker run -it --rm --pull=always \
-    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik \
+    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik \
     -e LOG_ALL_EVENTS=true \
     -v /var/run/docker.sock:/var/run/docker.sock \
     -v ~/.openhands-state:/.openhands-state \
     -p 3000:3000 \
     --add-host host.docker.internal:host-gateway \
     --name openhands-app \
-    docker.all-hands.dev/all-hands-ai/openhands:0.18
+    docker.all-hands.dev/all-hands-ai/openhands:0.19
 ```
 
 You'll find OpenHands running at [http://localhost:3000](http://localhost:3000)!

compose.yml

@@ -7,7 +7,7 @@ services:
     image: openhands:latest
     container_name: openhands-app-${DATE:-}
     environment:
-      - SANDBOX_RUNTIME_CONTAINER_IMAGE=${SANDBOX_RUNTIME_CONTAINER_IMAGE:-ghcr.io/all-hands-ai/runtime:0.18-nikolaik}
+      - SANDBOX_RUNTIME_CONTAINER_IMAGE=${SANDBOX_RUNTIME_CONTAINER_IMAGE:-ghcr.io/all-hands-ai/runtime:0.19-nikolaik}
       - SANDBOX_USER_ID=${SANDBOX_USER_ID:-1234}
       - WORKSPACE_MOUNT_PATH=${WORKSPACE_BASE:-$PWD/workspace}
     ports:

containers/dev/compose.yml

@@ -11,7 +11,7 @@ services:
       - BACKEND_HOST=${BACKEND_HOST:-"0.0.0.0"}
       - SANDBOX_API_HOSTNAME=host.docker.internal
       #
-      - SANDBOX_RUNTIME_CONTAINER_IMAGE=${SANDBOX_RUNTIME_CONTAINER_IMAGE:-ghcr.io/all-hands-ai/runtime:0.18-nikolaik}
+      - SANDBOX_RUNTIME_CONTAINER_IMAGE=${SANDBOX_RUNTIME_CONTAINER_IMAGE:-ghcr.io/all-hands-ai/runtime:0.19-nikolaik}
       - SANDBOX_USER_ID=${SANDBOX_USER_ID:-1234}
       - WORKSPACE_MOUNT_PATH=${WORKSPACE_BASE:-$PWD/workspace}
     ports:

docs/i18n/fr/docusaurus-plugin-content-docs/current/usage/how-to/cli-mode.md

@@ -52,7 +52,7 @@ LLM_API_KEY="sk_test_12345"
 ```bash
 docker run -it \
     --pull=always \
-    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik \
+    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik \
     -e SANDBOX_USER_ID=$(id -u) \
     -e WORKSPACE_MOUNT_PATH=$WORKSPACE_BASE \
     -e LLM_API_KEY=$LLM_API_KEY \
@@ -61,7 +61,7 @@ docker run -it \
     -v /var/run/docker.sock:/var/run/docker.sock \
     --add-host host.docker.internal:host-gateway \
     --name openhands-app-$(date +%Y%m%d%H%M%S) \
-    docker.all-hands.dev/all-hands-ai/openhands:0.18 \
+    docker.all-hands.dev/all-hands-ai/openhands:0.19 \
     python -m openhands.core.cli
 ```
 

docs/i18n/fr/docusaurus-plugin-content-docs/current/usage/how-to/headless-mode.md

@@ -46,7 +46,7 @@ LLM_API_KEY="sk_test_12345"
 ```bash
 docker run -it \
     --pull=always \
-    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik \
+    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik \
     -e SANDBOX_USER_ID=$(id -u) \
     -e WORKSPACE_MOUNT_PATH=$WORKSPACE_BASE \
     -e LLM_API_KEY=$LLM_API_KEY \
@@ -56,6 +56,6 @@ docker run -it \
     -v /var/run/docker.sock:/var/run/docker.sock \
     --add-host host.docker.internal:host-gateway \
     --name openhands-app-$(date +%Y%m%d%H%M%S) \
-    docker.all-hands.dev/all-hands-ai/openhands:0.18 \
+    docker.all-hands.dev/all-hands-ai/openhands:0.19 \
     python -m openhands.core.main -t "write a bash script that prints hi" --no-auto-continue
 ```

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

@@ -13,16 +13,16 @@
 La façon la plus simple d'exécuter OpenHands est avec Docker.
 
 ```bash
-docker pull docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik
+docker pull docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik
 
 docker run -it --rm --pull=always \
-    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik \
+    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik \
     -e LOG_ALL_EVENTS=true \
     -v /var/run/docker.sock:/var/run/docker.sock \
     -p 3000:3000 \
     --add-host host.docker.internal:host-gateway \
     --name openhands-app \
-    docker.all-hands.dev/all-hands-ai/openhands:0.18
+    docker.all-hands.dev/all-hands-ai/openhands:0.19
 ```
 
 Vous pouvez également exécuter OpenHands en mode [headless scriptable](https://docs.all-hands.dev/modules/usage/how-to/headless-mode), en tant que [CLI interactive](https://docs.all-hands.dev/modules/usage/how-to/cli-mode), ou en utilisant l'[Action GitHub OpenHands](https://docs.all-hands.dev/modules/usage/how-to/github-action).

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

@@ -13,7 +13,7 @@ C'est le Runtime par défaut qui est utilisé lorsque vous démarrez OpenHands.
 
 ```
 docker run # ...
-    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik \
+    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik \
     -v /var/run/docker.sock:/var/run/docker.sock \
     # ...
 ```

docs/i18n/zh-Hans/docusaurus-plugin-content-docs/current/usage/how-to/cli-mode.md

@@ -50,7 +50,7 @@ LLM_API_KEY="sk_test_12345"
 ```bash
 docker run -it \
     --pull=always \
-    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik \
+    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik \
     -e SANDBOX_USER_ID=$(id -u) \
     -e WORKSPACE_MOUNT_PATH=$WORKSPACE_BASE \
     -e LLM_API_KEY=$LLM_API_KEY \
@@ -59,7 +59,7 @@ docker run -it \
     -v /var/run/docker.sock:/var/run/docker.sock \
     --add-host host.docker.internal:host-gateway \
     --name openhands-app-$(date +%Y%m%d%H%M%S) \
-    docker.all-hands.dev/all-hands-ai/openhands:0.18 \
+    docker.all-hands.dev/all-hands-ai/openhands:0.19 \
     python -m openhands.core.cli
 ```
 

docs/i18n/zh-Hans/docusaurus-plugin-content-docs/current/usage/how-to/headless-mode.md

@@ -47,7 +47,7 @@ LLM_API_KEY="sk_test_12345"
 ```bash
 docker run -it \
     --pull=always \
-    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik \
+    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik \
     -e SANDBOX_USER_ID=$(id -u) \
     -e WORKSPACE_MOUNT_PATH=$WORKSPACE_BASE \
     -e LLM_API_KEY=$LLM_API_KEY \
@@ -57,6 +57,6 @@ docker run -it \
     -v /var/run/docker.sock:/var/run/docker.sock \
     --add-host host.docker.internal:host-gateway \
     --name openhands-app-$(date +%Y%m%d%H%M%S) \
-    docker.all-hands.dev/all-hands-ai/openhands:0.18 \
+    docker.all-hands.dev/all-hands-ai/openhands:0.19 \
     python -m openhands.core.main -t "write a bash script that prints hi" --no-auto-continue
 ```

docs/i18n/zh-Hans/docusaurus-plugin-content-docs/current/usage/installation.mdx

@@ -11,16 +11,16 @@
 在 Docker 中运行 OpenHands 是最简单的方式。
 
 ```bash
-docker pull docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik
+docker pull docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik
 
 docker run -it --rm --pull=always \
-    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik \
+    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik \
     -e LOG_ALL_EVENTS=true \
     -v /var/run/docker.sock:/var/run/docker.sock \
     -p 3000:3000 \
     --add-host host.docker.internal:host-gateway \
     --name openhands-app \
-    docker.all-hands.dev/all-hands-ai/openhands:0.18
+    docker.all-hands.dev/all-hands-ai/openhands:0.19
 ```
 
 你也可以在可脚本化的[无头模式](https://docs.all-hands.dev/modules/usage/how-to/headless-mode)下运行 OpenHands，作为[交互式 CLI](https://docs.all-hands.dev/modules/usage/how-to/cli-mode)，或使用 [OpenHands GitHub Action](https://docs.all-hands.dev/modules/usage/how-to/github-action)。

docs/i18n/zh-Hans/docusaurus-plugin-content-docs/current/usage/runtimes.md

@@ -11,7 +11,7 @@
 
 ```
 docker run # ...
-    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik \
+    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik \
     -v /var/run/docker.sock:/var/run/docker.sock \
     # ...
 ```

docs/modules/usage/how-to/cli-mode.md

@@ -35,7 +35,7 @@ To run OpenHands in CLI mode with Docker:
 ```bash
 docker run -it \
     --pull=always \
-    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik \
+    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik \
     -e SANDBOX_USER_ID=$(id -u) \
     -e WORKSPACE_MOUNT_PATH=$WORKSPACE_BASE \
     -e LLM_API_KEY=$LLM_API_KEY \
@@ -45,7 +45,7 @@ docker run -it \
     -v ~/.openhands-state:/.openhands-state \
     --add-host host.docker.internal:host-gateway \
     --name openhands-app-$(date +%Y%m%d%H%M%S) \
-    docker.all-hands.dev/all-hands-ai/openhands:0.18 \
+    docker.all-hands.dev/all-hands-ai/openhands:0.19 \
     python -m openhands.core.cli
 ```
 

docs/modules/usage/how-to/headless-mode.md

@@ -32,7 +32,7 @@ To run OpenHands in Headless mode with Docker:
 ```bash
 docker run -it \
     --pull=always \
-    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik \
+    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik \
     -e SANDBOX_USER_ID=$(id -u) \
     -e WORKSPACE_MOUNT_PATH=$WORKSPACE_BASE \
     -e LLM_API_KEY=$LLM_API_KEY \
@@ -43,7 +43,7 @@ docker run -it \
     -v ~/.openhands-state:/.openhands-state \
     --add-host host.docker.internal:host-gateway \
     --name openhands-app-$(date +%Y%m%d%H%M%S) \
-    docker.all-hands.dev/all-hands-ai/openhands:0.18 \
+    docker.all-hands.dev/all-hands-ai/openhands:0.19 \
     python -m openhands.core.main -t "write a bash script that prints hi"
 ```
 

docs/modules/usage/installation.mdx

@@ -11,17 +11,17 @@
 The easiest way to run OpenHands is in Docker.
 
 ```bash
-docker pull docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik
+docker pull docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik
 
 docker run -it --rm --pull=always \
-    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik \
+    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik \
     -e LOG_ALL_EVENTS=true \
     -v /var/run/docker.sock:/var/run/docker.sock \
     -v ~/.openhands-state:/.openhands-state \
     -p 3000:3000 \
     --add-host host.docker.internal:host-gateway \
     --name openhands-app \
-    docker.all-hands.dev/all-hands-ai/openhands:0.18
+    docker.all-hands.dev/all-hands-ai/openhands:0.19
 ```
 
 You'll find OpenHands running at http://localhost:3000!

docs/modules/usage/prompting/microagents-public.md

@@ -1,17 +1,31 @@
-# Micro-Agents
+# Public Micro-Agents
 
-OpenHands uses specialized micro-agents to handle specific tasks and contexts efficiently. These micro-agents are small, focused components that provide specialized behavior and knowledge for particular scenarios.
+OpenHands uses specialized micro-agents to handle specific tasks and contexts efficiently. These micro-agents are small,
+focused components that provide specialized behavior and knowledge for particular scenarios.
 
 ## Overview
 
-Micro-agents are defined in markdown files under the `openhands/agenthub/codeact_agent/micro/` directory. Each micro-agent is configured with:
+Public micro-agents are defined in markdown files under the
+[`microagents/knowledge/`](https://github.com/All-Hands-AI/OpenHands/tree/main/microagents/knowledge) directory.
+Each micro-agent is configured with:
 
 - A unique name.
 - The agent type (typically CodeActAgent).
 - Trigger keywords that activate the agent.
 - Specific instructions and capabilities.
 
-## Available Micro-Agents
+### Integration
+
+Public micro-agents are automatically integrated into OpenHands' workflow. They:
+- Monitor incoming commands for their trigger words.
+- Activate when relevant triggers are detected.
+- Apply their specialized knowledge and capabilities.
+- Follow their specific guidelines and restrictions.
+
+## Available Public Micro-Agents
+
+For more information about specific micro-agents, refer to their individual documentation files in
+the [`micro-agents`](https://github.com/All-Hands-AI/OpenHands/tree/main/microagents) directory.
 
 ### GitHub Agent
 **File**: `github.md`
@@ -29,6 +43,14 @@ Key features:
 - Git configuration management
 - API-first approach for GitHub operations
 
+Usage Example:
+
+```bash
+git checkout -b feature-branch
+git commit -m "Add new feature"
+git push origin feature-branch
+```
+
 ### NPM Agent
 **File**: `npm.md`
 **Triggers**: `npm`
@@ -38,9 +60,15 @@ Specializes in handling npm package management with specific focus on:
 - Automated confirmation handling using Unix 'yes' command.
 - Package installation automation.
 
-### Custom Micro-Agents
+Usage Example:
 
-You can create your own micro-agents by adding new markdown files to the micro-agents directory.
+```bash
+yes | npm install package-name
+```
+
+### Custom Public Micro-Agents
+
+You can create your own public micro-agents by adding new markdown files to the `microagents/knowledge/` directory.
 Each file should follow this structure:
 
 ```markdown
@@ -55,43 +83,29 @@ triggers:
 Instructions and capabilities for the micro-agent...
 ```
 
-## Best Practices
+## Working With Public Micro-Agents
 
-When working with micro-agents:
+When working with public micro-agents:
 - **Use Appropriate Triggers**: Ensure your commands include the relevant trigger words to activate the correct micro-agent.
 - **Follow Agent Guidelines**: Each agent has specific instructions and limitations. Respect these for optimal results.
 - **API-First Approach**: When available, use API endpoints rather than web interfaces.
 - **Automation Friendly**: Design commands that work well in non-interactive environments.
 
-## Integration
+## Contributing a Public Micro-Agent
 
-Micro-agents are automatically integrated into OpenHands' workflow. They:
-- Monitor incoming commands for their trigger words.
-- Activate when relevant triggers are detected.
-- Apply their specialized knowledge and capabilities.
-- Follow their specific guidelines and restrictions.
+Best practices for creating public micro-agents:
 
-## Example Usage
+- **Clear Scope**: Keep the micro-agent focused on a specific domain or task.
+- **Explicit Instructions**: Provide clear, unambiguous guidelines.
+- **Useful Examples**: Include practical examples of common use cases.
+- **Safety First**: Include necessary warnings and constraints.
+- **Integration Awareness**: Consider how the micro-agent interacts with other components.
 
-```bash
-# GitHub agent example
-git checkout -b feature-branch
-git commit -m "Add new feature"
-git push origin feature-branch
+To contribute a new micro-agent to OpenHands:
 
-# NPM agent example
-yes | npm install package-name
-```
+### 1. Plan the Public Micro-Agent
 
-For more information about specific agents, refer to their individual documentation files in the micro-agents directory.
-
-## Contributing a Micro-Agent
-
-To contribute a new micro-agent to OpenHands, follow these guidelines:
-
-### 1. Planning Your Micro-Agent
-
-Before creating a micro-agent, consider:
+Before creating a public micro-agent, consider:
 - What specific problem or use case will it address?
 - What unique capabilities or knowledge should it have?
 - What trigger words make sense for activating it?
@@ -99,11 +113,11 @@ Before creating a micro-agent, consider:
 
 ### 2. File Structure
 
-Create a new markdown file in `openhands/agenthub/codeact_agent/micro/` with a descriptive name (e.g., `docker.md` for a Docker-focused agent).
+Create a new markdown file in `microagents/knowledge/` with a descriptive name (e.g., `docker.md` for a Docker-focused agent).
 
 ### 3. Required Components
 
-Your micro-agent file must include:
+The micro-agent file must include:
 
 - **Front Matter**: YAML metadata at the start of the file:
 ```markdown
@@ -133,15 +147,7 @@ Examples of usage:
 [Example 2]
 ```
 
-### 4. Best Practices for Micro-Agent Development
-
-- **Clear Scope**: Keep the agent focused on a specific domain or task.
-- **Explicit Instructions**: Provide clear, unambiguous guidelines.
-- **Useful Examples**: Include practical examples of common use cases.
-- **Safety First**: Include necessary warnings and constraints.
-- **Integration Awareness**: Consider how the agent interacts with other components.
-
-### 5. Testing Your Micro-Agent
+### 4. Testing the Public Micro-Agent
 
 Before submitting:
 - Test the agent with various prompts.
@@ -149,7 +155,14 @@ Before submitting:
 - Ensure instructions are clear and comprehensive.
 - Check for potential conflicts with existing agents.
 
-### 6. Example Implementation
+### 5. Submission Process
+
+Submit a pull request with:
+- The new micro-agent file.
+- Updated documentation if needed.
+- Description of the agent's purpose and capabilities.
+
+### Example Public Micro-Agent Implementation
 
 Here's a template for a new micro-agent:
 
@@ -197,14 +210,5 @@ Remember to:
 - Optimize for build time and image size
 ```
 
-### 7. Submission Process
-
-1. Create your micro-agent file in the correct directory.
-2. Test thoroughly.
-3. Submit a pull request with:
-   - The new micro-agent file.
-   - Updated documentation if needed.
-   - Description of the agent's purpose and capabilities.
-
 Remember that micro-agents are a powerful way to extend OpenHands' capabilities in specific domains. Well-designed
 agents can significantly improve the system's ability to handle specialized tasks.

docs/modules/usage/prompting/microagents-repo.md

@@ -1,10 +1,12 @@
-# Customizing Agent Behavior
+# Repository Micro-Agents
 
-OpenHands can be customized to work more effectively with specific repositories by providing repository-specific context and guidelines. This section explains how to optimize OpenHands for your project.
+OpenHands can be customized to work more effectively with specific repositories by providing repository-specific context
+and guidelines. This section explains how to optimize OpenHands for your project.
 
 ## Repository Configuration
 
-You can customize OpenHands' behavior for your repository by creating a `.openhands` directory in your repository's root. At minimum, it should contain the file
+You can customize OpenHands' behavior for your repository by creating a `.openhands/microagents/` directory in your repository's root.
+At minimum, it should contain the file
 `.openhands/microagents/repo.md`, which includes instructions that will
 be given to the agent every time it works with this repository.
 
@@ -39,7 +41,8 @@ Guidelines:
 
 ### Customizing Prompts
 
-When working with a repository:
+You may also add customized prompts to the `.openhands/microagents/repo.md` file when working with a repository.
+These could:
 
 - **Reference Project Standards**: Mention specific coding standards or patterns used in your project.
 - **Include Context**: Reference relevant documentation or existing implementations.
@@ -54,14 +57,10 @@ The component should use our shared styling from src/styles/components.
 
 ### Best Practices for Repository Customization
 
-- **Keep Instructions Updated**: Regularly update your `.openhands` directory as your project evolves.
+- **Keep Instructions Updated**: Regularly update your `.openhands/microagents/` directory as your project evolves.
 - **Be Specific**: Include specific paths, patterns, and requirements unique to your project.
 - **Document Dependencies**: List all tools and dependencies required for development.
 - **Include Examples**: Provide examples of good code patterns from your project.
 - **Specify Conventions**: Document naming conventions, file organization, and code style preferences.
 
 By customizing OpenHands for your repository, you'll get more accurate and consistent results that align with your project's standards and requirements.
-
-## Other Microagents
-You can create other instructions in the `.openhands/microagents/` directory
-that will be sent to the agent if a particular keyword is found, like `test`, `frontend`, or `migration`. See [Micro-Agents](microagents.md) for more information.

docs/modules/usage/runtimes.md

@@ -16,7 +16,7 @@ some flags being passed to `docker run` that make this possible:
 
 ```
 docker run # ...
-    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.18-nikolaik \
+    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.19-nikolaik \
     -v /var/run/docker.sock:/var/run/docker.sock \
     # ...
 ```

docs/package.json

@@ -15,10 +15,10 @@
     "typecheck": "tsc"
   },
   "dependencies": {
-    "@docusaurus/core": "^3.6.3",
-    "@docusaurus/plugin-content-pages": "^3.6.3",
-    "@docusaurus/preset-classic": "^3.6.3",
-    "@docusaurus/theme-mermaid": "^3.6.3",
+    "@docusaurus/core": "^3.7.0",
+    "@docusaurus/plugin-content-pages": "^3.7.0",
+    "@docusaurus/preset-classic": "^3.7.0",
+    "@docusaurus/theme-mermaid": "^3.7.0",
     "@mdx-js/react": "^3.1.0",
     "clsx": "^2.0.0",
     "prism-react-renderer": "^2.4.1",
@@ -29,7 +29,7 @@
   },
   "devDependencies": {
     "@docusaurus/module-type-aliases": "^3.5.1",
-    "@docusaurus/tsconfig": "^3.6.3",
+    "@docusaurus/tsconfig": "^3.7.0",
     "@docusaurus/types": "^3.5.1",
     "typescript": "~5.7.2"
   },

docs/sidebars.ts

@@ -23,15 +23,21 @@ const sidebars: SidebarsConfig = {
           id: 'usage/prompting/prompting-best-practices',
         },
         {
-          type: 'doc',
-          label: 'Customization',
-          id: 'usage/prompting/customization',
-        },
-        {
-          type: 'doc',
-          label: 'Microagents',
-          id: 'usage/prompting/microagents',
-        },
+          type: 'category',
+          label: 'Micro-Agents',
+          items: [
+            {
+              type: 'doc',
+              label: 'Public',
+              id: 'usage/prompting/microagents-public',
+            },
+            {
+              type: 'doc',
+              label: 'Repository',
+              id: 'usage/prompting/microagents-repo',
+            },
+          ],
+        }
       ],
     },
     {

dummy_eval_output_dir/dummy_dataset_descrption/CodeActAgent/claude-3-5-sonnet-20241022_maxiter_10_N_dummy_eval_note/metadata.json

@@ -0,0 +1 @@
+{"agent_class": "CodeActAgent", "llm_config": {"model": "claude-3-5-sonnet-20241022", "api_key": null, "base_url": null, "api_version": null, "embedding_model": "local", "embedding_base_url": null, "embedding_deployment_name": null, "aws_access_key_id": null, "aws_secret_access_key": null, "aws_region_name": null, "openrouter_site_url": "https://docs.all-hands.dev/", "openrouter_app_name": "OpenHands", "num_retries": 8, "retry_multiplier": 2, "retry_min_wait": 15, "retry_max_wait": 120, "timeout": null, "max_message_chars": 30000, "temperature": 0.0, "top_p": 1.0, "custom_llm_provider": null, "max_input_tokens": null, "max_output_tokens": null, "input_cost_per_token": null, "output_cost_per_token": null, "ollama_base_url": null, "drop_params": true, "modify_params": true, "disable_vision": null, "caching_prompt": true, "log_completions": false, "log_completions_folder": "/workspace/OpenHands/logs/completions", "draft_editor": null, "custom_tokenizer": null, "native_tool_calling": null}, "max_iterations": 10, "eval_output_dir": "./dummy_eval_output_dir/dummy_dataset_descrption/CodeActAgent/claude-3-5-sonnet-20241022_maxiter_10_N_dummy_eval_note", "start_time": "2025-01-08 18:01:01", "git_commit": "007052c8aa15ea5149fff31583a3412ea7b8625a", "dataset": "dummy_dataset_descrption", "data_split": null, "details": {}, "condenser_config": {"type": "noop"}}

evaluation/benchmarks/swe_bench/run_infer.py

@@ -15,6 +15,7 @@ from evaluation.utils.shared import (
     EvalOutput,
     assert_and_raise,
     codeact_user_response,
+    get_metrics,
     is_fatal_evaluation_error,
     make_metadata,
     prepare_dataset,
@@ -148,6 +149,7 @@ def get_config(
         codeact_enable_jupyter=False,
         codeact_enable_browsing=RUN_WITH_BROWSING,
         codeact_enable_llm_editor=False,
+        condenser=metadata.condenser_config,
     )
     config.set_agent_config(agent_config)
     return config
@@ -448,7 +450,7 @@ def process_instance(
 
     # NOTE: this is NO LONGER the event stream, but an agent history that includes delegate agent's events
     histories = [event_to_dict(event) for event in state.history]
-    metrics = state.metrics.get() if state.metrics else None
+    metrics = get_metrics(state)
 
     # Save the output
     output = EvalOutput(

evaluation/utils/shared.py

@@ -17,6 +17,10 @@ from tqdm import tqdm
 
 from openhands.controller.state.state import State
 from openhands.core.config import LLMConfig
+from openhands.core.config.condenser_config import (
+    CondenserConfig,
+    NoOpCondenserConfig,
+)
 from openhands.core.exceptions import (
     AgentRuntimeBuildError,
     AgentRuntimeDisconnectedError,
@@ -33,6 +37,7 @@ from openhands.events.action.message import MessageAction
 from openhands.events.event import Event
 from openhands.events.serialization.event import event_to_dict
 from openhands.events.utils import get_pairs_from_events
+from openhands.memory.condenser import get_condensation_metadata
 
 
 class EvalMetadata(BaseModel):
@@ -45,11 +50,17 @@ class EvalMetadata(BaseModel):
     dataset: str | None = None
     data_split: str | None = None
     details: dict[str, Any] | None = None
+    condenser_config: CondenserConfig | None = None
 
     def model_dump(self, *args, **kwargs):
         dumped_dict = super().model_dump(*args, **kwargs)
         # avoid leaking sensitive information
         dumped_dict['llm_config'] = self.llm_config.to_safe_dict()
+        if hasattr(self.condenser_config, 'llm_config'):
+            dumped_dict['condenser_config']['llm_config'] = (
+                self.condenser_config.llm_config.to_safe_dict()
+            )
+
         return dumped_dict
 
     def model_dump_json(self, *args, **kwargs):
@@ -57,6 +68,11 @@ class EvalMetadata(BaseModel):
         dumped_dict = json.loads(dumped)
         # avoid leaking sensitive information
         dumped_dict['llm_config'] = self.llm_config.to_safe_dict()
+        if hasattr(self.condenser_config, 'llm_config'):
+            dumped_dict['condenser_config']['llm_config'] = (
+                self.condenser_config.llm_config.to_safe_dict()
+            )
+
         logger.debug(f'Dumped metadata: {dumped_dict}')
         return json.dumps(dumped_dict)
 
@@ -192,6 +208,7 @@ def make_metadata(
     eval_output_dir: str,
     data_split: str | None = None,
     details: dict[str, Any] | None = None,
+    condenser_config: CondenserConfig | None = None,
 ) -> EvalMetadata:
     model_name = llm_config.model.split('/')[-1]
     model_path = model_name.replace(':', '_').replace('@', '-')
@@ -222,6 +239,9 @@ def make_metadata(
         dataset=dataset_name,
         data_split=data_split,
         details=details,
+        condenser_config=condenser_config
+        if condenser_config
+        else NoOpCondenserConfig(),
     )
     metadata_json = metadata.model_dump_json()
     logger.info(f'Metadata: {metadata_json}')
@@ -551,3 +571,10 @@ def is_fatal_evaluation_error(error: str | None) -> bool:
         return True
 
     return False
+
+
+def get_metrics(state: State) -> dict[str, Any]:
+    """Extract metrics from the state."""
+    metrics = state.metrics.get() if state.metrics else {}
+    metrics['condenser'] = get_condensation_metadata(state)
+    return metrics

frontend/__tests__/context/ws-client-provider.test.tsx

@@ -0,0 +1,30 @@
+import { describe, it, expect, vi } from "vitest";
+import { render, screen } from "@testing-library/react";
+import * as ChatSlice from "#/state/chat-slice";
+import {
+  updateStatusWhenErrorMessagePresent,
+} from "#/context/ws-client-provider";
+
+describe("Propagate error message", () => {
+  it("should do nothing when no message was passed from server", () => {
+    const addErrorMessageSpy = vi.spyOn(ChatSlice, "addErrorMessage");
+    updateStatusWhenErrorMessagePresent(null)
+    updateStatusWhenErrorMessagePresent(undefined)
+    updateStatusWhenErrorMessagePresent({})
+    updateStatusWhenErrorMessagePresent({message: null})
+
+    expect(addErrorMessageSpy).not.toHaveBeenCalled();
+  });
+
+  it("should display error to user when present", () => {
+    const message = "We have a problem!"
+    const addErrorMessageSpy = vi.spyOn(ChatSlice, "addErrorMessage")
+    updateStatusWhenErrorMessagePresent({message})
+
+    expect(addErrorMessageSpy).toHaveBeenCalledWith({
+      message,
+      status_update: true,
+      type: 'error'
+     });
+  });
+});

frontend/package.json

@@ -1,6 +1,6 @@
 {
   "name": "openhands-frontend",
-  "version": "0.18.0",
+  "version": "0.19.0",
   "private": true,
   "type": "module",
   "engines": {
@@ -8,32 +8,32 @@
   },
   "dependencies": {
     "@monaco-editor/react": "^4.7.0-rc.0",
-    "@nextui-org/react": "^2.6.10",
+    "@nextui-org/react": "^2.6.11",
     "@react-router/node": "^7.1.1",
     "@react-router/serve": "^7.1.1",
     "@react-types/shared": "^3.25.0",
     "@reduxjs/toolkit": "^2.5.0",
-    "@tanstack/react-query": "^5.62.12",
+    "@tanstack/react-query": "^5.63.0",
     "@vitejs/plugin-react": "^4.3.2",
     "@xterm/addon-fit": "^0.10.0",
     "@xterm/xterm": "^5.4.0",
     "axios": "^1.7.9",
     "clsx": "^2.1.1",
     "eslint-config-airbnb-typescript": "^18.0.0",
-    "i18next": "^24.2.0",
+    "i18next": "^24.2.1",
     "i18next-browser-languagedetector": "^8.0.2",
     "i18next-http-backend": "^3.0.1",
-    "isbot": "^5.1.19",
+    "isbot": "^5.1.20",
     "jose": "^5.9.4",
     "monaco-editor": "^0.52.2",
-    "posthog-js": "^1.203.3",
+    "posthog-js": "^1.205.0",
     "react": "^19.0.0",
     "react-dom": "^19.0.0",
     "react-highlight": "^0.15.0",
     "react-hot-toast": "^2.5.1",
     "react-i18next": "^15.4.0",
     "react-icons": "^5.4.0",
-    "react-markdown": "^9.0.1",
+    "react-markdown": "^9.0.3",
     "react-redux": "^9.2.0",
     "react-router": "^7.1.1",
     "react-syntax-highlighter": "^15.6.1",
@@ -78,13 +78,13 @@
     "@mswjs/socket.io-binding": "^0.1.1",
     "@playwright/test": "^1.49.1",
     "@react-router/dev": "^7.1.1",
-    "@tailwindcss/typography": "^0.5.15",
+    "@tailwindcss/typography": "^0.5.16",
     "@tanstack/eslint-plugin-query": "^5.62.16",
     "@testing-library/jest-dom": "^6.6.1",
     "@testing-library/react": "^16.1.0",
     "@testing-library/user-event": "^14.5.2",
     "@types/node": "^22.10.5",
-    "@types/react": "^19.0.2",
+    "@types/react": "^19.0.3",
     "@types/react-dom": "^19.0.2",
     "@types/react-highlight": "^0.12.8",
     "@types/react-syntax-highlighter": "^15.5.13",

frontend/src/context/ws-client-provider.tsx

@@ -2,7 +2,10 @@ import posthog from "posthog-js";
 import React from "react";
 import { io, Socket } from "socket.io-client";
 import EventLogger from "#/utils/event-logger";
-import { handleAssistantMessage } from "#/services/actions";
+import {
+  handleAssistantMessage,
+  handleStatusMessage,
+} from "#/services/actions";
 import { useRate } from "#/hooks/use-rate";
 import { OpenHandsParsedEvent } from "#/types/core";
 import {
@@ -64,6 +67,21 @@ interface WsClientProviderProps {
   conversationId: string;
 }
 
+export function updateStatusWhenErrorMessagePresent(data: unknown) {
+  if (
+    data &&
+    typeof data === "object" &&
+    "message" in data &&
+    typeof data.message === "string"
+  ) {
+    handleStatusMessage({
+      type: "error",
+      message: data.message,
+      status_update: true,
+    });
+  }
+}
+
 export function WsClientProvider({
   conversationId,
   children,
@@ -101,7 +119,7 @@ export function WsClientProvider({
     handleAssistantMessage(event);
   }
 
-  function handleDisconnect() {
+  function handleDisconnect(data: unknown) {
     setStatus(WsClientProviderStatus.DISCONNECTED);
     const sio = sioRef.current;
     if (!sio) {
@@ -109,11 +127,13 @@ export function WsClientProvider({
     }
     sio.io.opts.query = sio.io.opts.query || {};
     sio.io.opts.query.latest_event_id = lastEventRef.current?.id;
+    updateStatusWhenErrorMessagePresent(data);
   }
 
-  function handleError() {
-    posthog.capture("socket_error");
+  function handleError(data: unknown) {
     setStatus(WsClientProviderStatus.DISCONNECTED);
+    updateStatusWhenErrorMessagePresent(data);
+    posthog.capture("socket_error");
   }
 
   React.useEffect(() => {

frontend/src/services/actions.ts

@@ -75,6 +75,7 @@ export function handleActionMessage(message: ActionMessage) {
     if (message.args && message.args.thought) {
       store.dispatch(addAssistantMessage(message.args.thought));
     }
+    // Need to convert ActionMessage to RejectAction
     // @ts-expect-error TODO: fix
     store.dispatch(addAssistantAction(message));
   }

frontend/src/state/chat-slice.ts

@@ -73,7 +73,7 @@ export const chatSlice = createSlice({
       state.messages.push(message);
     },
 
-    addAssistantMessage(state, action: PayloadAction<string>) {
+    addAssistantMessage(state: SliceState, action: PayloadAction<string>) {
       const message: Message = {
         type: "thought",
         sender: "assistant",
@@ -85,7 +85,10 @@ export const chatSlice = createSlice({
       state.messages.push(message);
     },
 
-    addAssistantAction(state, action: PayloadAction<OpenHandsAction>) {
+    addAssistantAction(
+      state: SliceState,
+      action: PayloadAction<OpenHandsAction>,
+    ) {
       const actionID = action.payload.action;
       if (!HANDLED_ACTIONS.includes(actionID)) {
         return;
@@ -125,7 +128,7 @@ export const chatSlice = createSlice({
     },
 
     addAssistantObservation(
-      state,
+      state: SliceState,
       observation: PayloadAction<OpenHandsObservation>,
     ) {
       const observationID = observation.payload.observation;
@@ -179,7 +182,7 @@ export const chatSlice = createSlice({
     },
 
     addErrorMessage(
-      state,
+      state: SliceState,
       action: PayloadAction<{ id?: string; message: string }>,
     ) {
       const { id, message } = action.payload;
@@ -192,7 +195,7 @@ export const chatSlice = createSlice({
       });
     },
 
-    clearMessages(state) {
+    clearMessages(state: SliceState) {
       state.messages = [];
     },
   },

frontend/src/types/message.tsx

@@ -43,6 +43,6 @@ export interface ObservationMessage {
 export interface StatusMessage {
   status_update: true;
   type: string;
-  id: string;
+  id?: string;
   message: string;
 }

openhands/agenthub/codeact_agent/codeact_agent.py

@@ -24,6 +24,7 @@ from openhands.events.action import (
     MessageAction,
 )
 from openhands.events.observation import (
+    AgentCondensationObservation,
     AgentDelegateObservation,
     BrowserOutputObservation,
     CmdOutputObservation,
@@ -36,6 +37,7 @@ from openhands.events.observation.error import ErrorObservation
 from openhands.events.observation.observation import Observation
 from openhands.events.serialization.event import truncate_content
 from openhands.llm.llm import LLM
+from openhands.memory.condenser import Condenser
 from openhands.runtime.plugins import (
     AgentSkillsRequirement,
     JupyterRequirement,
@@ -115,6 +117,9 @@ class CodeActAgent(Agent):
             disabled_microagents=self.config.disabled_microagents,
         )
 
+        self.condenser = Condenser.from_config(self.config.condenser)
+        logger.debug(f'Using condenser: {self.condenser}')
+
     def get_action_message(
         self,
         action: Action,
@@ -322,6 +327,9 @@ class CodeActAgent(Agent):
             text = 'OBSERVATION:\n' + truncate_content(obs.content, max_message_chars)
             text += '\n[Last action has been rejected by the user]'
             message = Message(role='user', content=[TextContent(text=text)])
+        elif isinstance(obs, AgentCondensationObservation):
+            text = truncate_content(obs.content, max_message_chars)
+            message = Message(role='user', content=[TextContent(text=text)])
         else:
             # If an observation message is not returned, it will cause an error
             # when the LLM tries to return the next message
@@ -442,7 +450,10 @@ class CodeActAgent(Agent):
 
         pending_tool_call_action_messages: dict[str, Message] = {}
         tool_call_id_to_message: dict[str, Message] = {}
-        events = list(state.history)
+
+        # Condense the events from the state.
+        events = self.condenser.condensed_history(state)
+
         for event in events:
             # create a regular message from an event
             if isinstance(event, Action):

openhands/agenthub/codeact_agent/prompts/system_prompt.j2

@@ -4,6 +4,11 @@ You are OpenHands agent, a helpful AI assistant that can interact with a compute
 * When configuring git credentials, use "openhands" as the user.name and "openhands@all-hands.dev" as the user.email by default, unless explicitly instructed otherwise.
 * The assistant MUST NOT include comments in the code unless they are necessary to describe non-obvious behavior.
 </IMPORTANT>
+{% if github_repo %}
+<REPOSITORY_INFO>
+At the user's request, repository {{ github_repo }} has been cloned to directory {{ repo_directory }}.
+</REPOSITORY_INFO>
+{% endif %}
 {% if repo_instructions %}
 <REPOSITORY_INSTRUCTIONS>
 {{ repo_instructions }}

openhands/controller/agent_controller.py

@@ -993,10 +993,12 @@ class AgentController:
 
     def __repr__(self):
         return (
-            f'AgentController(id={self.id}, agent={self.agent!r}, '
-            f'event_stream={self.event_stream!r}, '
-            f'state={self.state!r}, '
-            f'delegate={self.delegate!r}, _pending_action={self._pending_action!r})'
+            f'AgentController(id={getattr(self, "id", "<uninitialized>")}, '
+            f'agent={getattr(self, "agent", "<uninitialized>")!r}, '
+            f'event_stream={getattr(self, "event_stream", "<uninitialized>")!r}, '
+            f'state={getattr(self, "state", "<uninitialized>")!r}, '
+            f'delegate={getattr(self, "delegate", "<uninitialized>")!r}, '
+            f'_pending_action={getattr(self, "_pending_action", "<uninitialized>")!r})'
         )
 
     def _is_awaiting_observation(self):

openhands/core/config/agent_config.py

@@ -1,5 +1,6 @@
-from dataclasses import dataclass, fields
+from dataclasses import dataclass, field, fields
 
+from openhands.core.config.condenser_config import CondenserConfig, NoOpCondenserConfig
 from openhands.core.config.config_utils import get_field_info
 
 
@@ -18,6 +19,7 @@ class AgentConfig:
         llm_config: The name of the llm config to use. If specified, this will override global llm config.
         use_microagents: Whether to use microagents at all. Default is True.
         disabled_microagents: A list of microagents to disable. Default is None.
+        condenser: Configuration for the memory condenser. Default is NoOpCondenserConfig.
     """
 
     codeact_enable_browsing: bool = True
@@ -29,6 +31,7 @@ class AgentConfig:
     llm_config: str | None = None
     use_microagents: bool = True
     disabled_microagents: list[str] | None = None
+    condenser: CondenserConfig = field(default_factory=NoOpCondenserConfig)  # type: ignore
 
     def defaults_to_dict(self) -> dict:
         """Serialize fields to a dict for the frontend, including type hints, defaults, and whether it's optional."""

openhands/core/config/condenser_config.py

@@ -0,0 +1,90 @@
+from typing import Literal
+
+from pydantic import BaseModel, Field
+
+from openhands.core.config.llm_config import LLMConfig
+
+
+class NoOpCondenserConfig(BaseModel):
+    """Configuration for NoOpCondenser."""
+
+    type: Literal['noop'] = Field('noop')
+
+
+class ObservationMaskingCondenserConfig(BaseModel):
+    """Configuration for ObservationMaskingCondenser."""
+
+    type: Literal['observation_masking'] = Field('observation_masking')
+    attention_window: int = Field(
+        default=10,
+        description='The number of most-recent events where observations will not be masked.',
+        ge=1,
+    )
+
+
+class RecentEventsCondenserConfig(BaseModel):
+    """Configuration for RecentEventsCondenser."""
+
+    type: Literal['recent'] = Field('recent')
+    keep_first: int = Field(
+        default=0,
+        description='The number of initial events to condense.',
+        ge=0,
+    )
+    max_events: int = Field(
+        default=10, description='Maximum number of events to keep.', ge=1
+    )
+
+
+class LLMSummarizingCondenserConfig(BaseModel):
+    """Configuration for LLMCondenser."""
+
+    type: Literal['llm'] = Field('llm')
+    llm_config: LLMConfig = Field(
+        ..., description='Configuration for the LLM to use for condensing.'
+    )
+
+
+class AmortizedForgettingCondenserConfig(BaseModel):
+    """Configuration for AmortizedForgettingCondenser."""
+
+    type: Literal['amortized'] = Field('amortized')
+    max_size: int = Field(
+        default=100,
+        description='Maximum size of the condensed history before triggering forgetting.',
+        ge=2,
+    )
+    keep_first: int = Field(
+        default=0,
+        description='Number of initial events to always keep in history.',
+        ge=0,
+    )
+
+
+class LLMAttentionCondenserConfig(BaseModel):
+    """Configuration for LLMAttentionCondenser."""
+
+    type: Literal['llm_attention'] = Field('llm_attention')
+    llm_config: LLMConfig = Field(
+        ..., description='Configuration for the LLM to use for attention.'
+    )
+    max_size: int = Field(
+        default=100,
+        description='Maximum size of the condensed history before triggering forgetting.',
+        ge=2,
+    )
+    keep_first: int = Field(
+        default=0,
+        description='Number of initial events to always keep in history.',
+        ge=0,
+    )
+
+
+CondenserConfig = (
+    NoOpCondenserConfig
+    | ObservationMaskingCondenserConfig
+    | RecentEventsCondenserConfig
+    | LLMSummarizingCondenserConfig
+    | AmortizedForgettingCondenserConfig
+    | LLMAttentionCondenserConfig
+)

openhands/core/config/sandbox_config.py

@@ -37,8 +37,6 @@ class SandboxConfig:
         enable_gpu: Whether to enable GPU.
         docker_runtime_kwargs: Additional keyword arguments to pass to the Docker runtime when running containers.
             This should be a JSON string that will be parsed into a dictionary.
-            Example in config.toml:
-            docker_runtime_kwargs = '{"mem_limit": "4g", "cpu_quota": 100000}'
     """
 
     remote_runtime_api_url: str = 'http://localhost:8000'

openhands/core/schema/observation.py

@@ -44,5 +44,8 @@ class ObservationTypeSchema(BaseModel):
 
     USER_REJECTED: str = Field(default='user_rejected')
 
+    CONDENSE: str = Field(default='condense')
+    """Result of a condensation operation."""
+
 
 ObservationType = ObservationTypeSchema()

openhands/events/observation/__init__.py

@@ -1,4 +1,7 @@
-from openhands.events.observation.agent import AgentStateChangedObservation
+from openhands.events.observation.agent import (
+    AgentCondensationObservation,
+    AgentStateChangedObservation,
+)
 from openhands.events.observation.browse import BrowserOutputObservation
 from openhands.events.observation.commands import (
     CmdOutputMetadata,
@@ -32,4 +35,5 @@ __all__ = [
     'AgentDelegateObservation',
     'SuccessObservation',
     'UserRejectObservation',
+    'AgentCondensationObservation',
 ]

openhands/events/observation/agent.py

@@ -14,3 +14,14 @@ class AgentStateChangedObservation(Observation):
     @property
     def message(self) -> str:
         return ''
+
+
+@dataclass
+class AgentCondensationObservation(Observation):
+    """The output of a condensation action."""
+
+    observation: str = ObservationType.CONDENSE
+
+    @property
+    def message(self) -> str:
+        return self.content

openhands/events/serialization/observation.py

@@ -1,6 +1,9 @@
 import copy
 
-from openhands.events.observation.agent import AgentStateChangedObservation
+from openhands.events.observation.agent import (
+    AgentCondensationObservation,
+    AgentStateChangedObservation,
+)
 from openhands.events.observation.browse import BrowserOutputObservation
 from openhands.events.observation.commands import (
     CmdOutputMetadata,
@@ -32,6 +35,7 @@ observations = (
     ErrorObservation,
     AgentStateChangedObservation,
     UserRejectObservation,
+    AgentCondensationObservation,
 )
 
 OBSERVATION_TYPE_TO_CLASS = {

openhands/memory/__init__.py

@@ -1,4 +1,4 @@
-from openhands.memory.condenser import MemoryCondenser
+from openhands.memory.condenser import Condenser
 from openhands.memory.memory import LongTermMemory
 
-__all__ = ['LongTermMemory', 'MemoryCondenser']
+__all__ = ['LongTermMemory', 'Condenser']

openhands/memory/condenser.py

@@ -1,24 +1,409 @@
+from __future__ import annotations
+
+from abc import ABC, abstractmethod
+from contextlib import contextmanager
+from typing import Any
+
+from litellm import supports_response_schema
+from pydantic import BaseModel
+from typing_extensions import override
+
+from openhands.controller.state.state import State
+from openhands.core.config.condenser_config import (
+    AmortizedForgettingCondenserConfig,
+    CondenserConfig,
+    LLMAttentionCondenserConfig,
+    LLMSummarizingCondenserConfig,
+    NoOpCondenserConfig,
+    ObservationMaskingCondenserConfig,
+    RecentEventsCondenserConfig,
+)
 from openhands.core.logger import openhands_logger as logger
+from openhands.events.event import Event
+from openhands.events.observation import AgentCondensationObservation, Observation
 from openhands.llm.llm import LLM
 
+CONDENSER_METADATA_KEY = 'condenser_meta'
+"""Key identifying where metadata is stored in a `State` object's `extra_data` field."""
 
-class MemoryCondenser:
-    def condense(self, summarize_prompt: str, llm: LLM):
-        """Attempts to condense the memory by using the llm
 
-        Parameters:
-        - llm (LLM): llm to be used for summarization
+def get_condensation_metadata(state: State) -> list[dict[str, Any]]:
+    """Utility function to retrieve a list of metadata batches from a `State`.
+
+    Args:
+        state: The state to retrieve metadata from.
+
+    Returns:
+        list[dict[str, Any]]: A list of metadata batches, each representing a condensation.
+    """
+    if CONDENSER_METADATA_KEY in state.extra_data:
+        return state.extra_data[CONDENSER_METADATA_KEY]
+    return []
+
+
+class Condenser(ABC):
+    """Abstract condenser interface.
+
+    Condensers take a list of `Event` objects and reduce them into a potentially smaller list.
+
+    Agents can use condensers to reduce the amount of events they need to consider when deciding which action to take. To use a condenser, agents can call the `condensed_history` method on the current `State` being considered and use the results instead of the full history.
+
+    Example usage::
+
+        condenser = Condenser.from_config(condenser_config)
+        events = condenser.condensed_history(state)
+    """
+
+    def __init__(self):
+        self._metadata_batch: dict[str, Any] = {}
+
+    def add_metadata(self, key: str, value: Any) -> None:
+        """Add information to the current metadata batch.
+
+        Any key/value pairs added to the metadata batch will be recorded in the `State` at the end of the current condensation.
+
+        Args:
+            key: The key to store the metadata under.
+
+            value: The metadata to store.
+        """
+        self._metadata_batch[key] = value
+
+    def write_metadata(self, state: State) -> None:
+        """Write the current batch of metadata to the `State`.
+
+        Resets the current metadata batch: any metadata added after this call will be stored in a new batch and written to the `State` at the end of the next condensation.
+        """
+        if CONDENSER_METADATA_KEY not in state.extra_data:
+            state.extra_data[CONDENSER_METADATA_KEY] = []
+        if self._metadata_batch:
+            state.extra_data[CONDENSER_METADATA_KEY].append(self._metadata_batch)
+
+        # Since the batch has been written, clear it for the next condensation
+        self._metadata_batch = {}
+
+    @contextmanager
+    def metadata_batch(self, state: State):
+        """Context manager to ensure batched metadata is always written to the `State`."""
+        try:
+            yield
+        finally:
+            self.write_metadata(state)
+
+    @abstractmethod
+    def condense(self, events: list[Event]) -> list[Event]:
+        """Condense a sequence of events into a potentially smaller list.
+
+        New condenser strategies should override this method to implement their own condensation logic. Call `self.add_metadata` in the implementation to record any relevant per-condensation diagnostic information.
+
+        Args:
+            events: A list of events representing the entire history of the agent.
+
+        Returns:
+            list[Event]: An event sequence representing a condensed history of the agent.
+        """
+
+    def condensed_history(self, state: State) -> list[Event]:
+        """Condense the state's history."""
+        with self.metadata_batch(state):
+            return self.condense(state.history)
+
+    @classmethod
+    def from_config(cls, config: CondenserConfig) -> Condenser:
+        """Create a condenser from a configuration object.
+
+        Args:
+            config: Configuration for the condenser.
+
+        Returns:
+            Condenser: A condenser instance.
 
         Raises:
-        - Exception: the same exception as it got from the llm or processing the response
+            ValueError: If the condenser type is not recognized.
+        """
+        match config:
+            case NoOpCondenserConfig():
+                return NoOpCondenser()
+
+            case ObservationMaskingCondenserConfig():
+                return ObservationMaskingCondenser(
+                    **config.model_dump(exclude=['type'])
+                )
+
+            case RecentEventsCondenserConfig():
+                return RecentEventsCondenser(**config.model_dump(exclude=['type']))
+
+            case LLMSummarizingCondenserConfig(llm_config=llm_config):
+                return LLMSummarizingCondenser(llm=LLM(config=llm_config))
+
+            case AmortizedForgettingCondenserConfig():
+                return AmortizedForgettingCondenser(
+                    **config.model_dump(exclude=['type'])
+                )
+
+            case LLMAttentionCondenserConfig(llm_config=llm_config):
+                return LLMAttentionCondenser(
+                    llm=LLM(config=llm_config),
+                    **config.model_dump(exclude=['type', 'llm_config']),
+                )
+
+            case _:
+                raise ValueError(f'Unknown condenser config: {config}')
+
+
+class RollingCondenser(Condenser, ABC):
+    """Base class for a specialized condenser strategy that applies condensation to a rolling history.
+
+    The rolling history is computed by appending new events to the most recent condensation. For example, the sequence of calls::
+
+        assert state.history == [event1, event2, event3]
+        condensation = condenser.condensed_history(state)
+
+        # ...new events are added to the state...
+
+        assert state.history == [event1, event2, event3, event4, event5]
+        condenser.condensed_history(state)
+
+    will result in second call to `condensed_history` passing `condensation + [event4, event5]` to the `condense` method.
+    """
+
+    def __init__(self) -> None:
+        self._condensation: list[Event] = []
+        self._last_history_length: int = 0
+
+        super().__init__()
+
+    @override
+    def condensed_history(self, state: State) -> list[Event]:
+        new_events = state.history[self._last_history_length :]
+
+        with self.metadata_batch(state):
+            results = self.condense(self._condensation + new_events)
+
+        self._condensation = results
+        self._last_history_length = len(state.history)
+
+        return results
+
+
+class NoOpCondenser(Condenser):
+    """A condenser that does nothing to the event sequence."""
+
+    def condense(self, events: list[Event]) -> list[Event]:
+        """Returns the list of events unchanged."""
+        return events
+
+
+class ObservationMaskingCondenser(Condenser):
+    """A condenser that masks the values of observations outside of a recent attention window."""
+
+    def __init__(self, attention_window: int = 5):
+        self.attention_window = attention_window
+
+        super().__init__()
+
+    def condense(self, events: list[Event]) -> list[Event]:
+        """Replace the content of observations outside of the attention window with a placeholder."""
+        results: list[Event] = []
+        for i, event in enumerate(events):
+            if (
+                isinstance(event, Observation)
+                and i < len(events) - self.attention_window
+            ):
+                results.append(AgentCondensationObservation('<MASKED>'))
+            else:
+                results.append(event)
+
+        return results
+
+
+class RecentEventsCondenser(Condenser):
+    """A condenser that only keeps a certain number of the most recent events."""
+
+    def __init__(self, keep_first: int = 0, max_events: int = 10):
+        self.keep_first = keep_first
+        self.max_events = max_events
+
+        super().__init__()
+
+    def condense(self, events: list[Event]) -> list[Event]:
+        """Keep only the most recent events (up to `max_events`)."""
+        head = events[: self.keep_first]
+        tail_length = max(0, self.max_events - len(head))
+        tail = events[-tail_length:]
+        return head + tail
+
+
+class LLMSummarizingCondenser(Condenser):
+    """A condenser that relies on a language model to summarize the event sequence as a single event."""
+
+    def __init__(self, llm: LLM):
+        self.llm = llm
+
+        super().__init__()
+
+    def condense(self, events: list[Event]) -> list[Event]:
+        """Applies an LLM to summarize the list of events.
+
+        Raises:
+            Exception: If the LLM is unable to summarize the event sequence.
         """
         try:
-            messages = [{'content': summarize_prompt, 'role': 'user'}]
-            resp = llm.completion(messages=messages)
-            summary_response = resp['choices'][0]['message']['content']
-            return summary_response
-        except Exception as e:
-            logger.error('Error condensing thoughts: %s', str(e), exc_info=False)
+            # Convert events to a format suitable for summarization
+            events_text = '\n'.join(f'{e.timestamp}: {e.message}' for e in events)
+            summarize_prompt = f'Please summarize these events:\n{events_text}'
 
-            # TODO If the llm fails with ContextWindowExceededError, we can try to condense the memory chunk by chunk
-            raise
+            resp = self.llm.completion(
+                messages=[{'content': summarize_prompt, 'role': 'user'}]
+            )
+            summary_response = resp.choices[0].message.content
+
+            # Create a new summary event with the condensed content
+            summary_event = AgentCondensationObservation(summary_response)
+
+            # Add metrics to state
+            self.add_metadata('response', resp.model_dump())
+            self.add_metadata('metrics', self.llm.metrics.get())
+
+            return [summary_event]
+
+        except Exception as e:
+            logger.error('Error condensing events: %s', str(e), exc_info=False)
+            raise e
+
+
+class AmortizedForgettingCondenser(RollingCondenser):
+    """A condenser that maintains a condensed history and forgets old events when it grows too large."""
+
+    def __init__(self, max_size: int = 100, keep_first: int = 0):
+        """Initialize the condenser.
+
+        Args:
+            max_size: Maximum size of history before forgetting.
+            keep_first: Number of initial events to always keep.
+
+        Raises:
+            ValueError: If keep_first is greater than max_size, keep_first is negative, or max_size is non-positive.
+        """
+        if keep_first >= max_size // 2:
+            raise ValueError(
+                f'keep_first ({keep_first}) must be less than half of max_size ({max_size})'
+            )
+        if keep_first < 0:
+            raise ValueError(f'keep_first ({keep_first}) cannot be negative')
+        if max_size < 1:
+            raise ValueError(f'max_size ({keep_first}) cannot be non-positive')
+
+        self.max_size = max_size
+        self.keep_first = keep_first
+
+        super().__init__()
+
+    def condense(self, events: list[Event]) -> list[Event]:
+        """Apply the amortized forgetting strategy to the given list of events."""
+        if len(events) <= self.max_size:
+            return events
+
+        target_size = self.max_size // 2
+        head = events[: self.keep_first]
+
+        events_from_tail = target_size - len(head)
+        tail = events[-events_from_tail:]
+
+        return head + tail
+
+
+class ImportantEventSelection(BaseModel):
+    """Utility class for the `LLMAttentionCondenser` that forces the LLM to return a list of integers."""
+
+    ids: list[int]
+
+
+class LLMAttentionCondenser(RollingCondenser):
+    """Rolling condenser strategy that uses an LLM to select the most important events when condensing the history."""
+
+    def __init__(self, llm: LLM, max_size: int = 100, keep_first: int = 0):
+        if keep_first >= max_size // 2:
+            raise ValueError(
+                f'keep_first ({keep_first}) must be less than half of max_size ({max_size})'
+            )
+        if keep_first < 0:
+            raise ValueError(f'keep_first ({keep_first}) cannot be negative')
+        if max_size < 1:
+            raise ValueError(f'max_size ({keep_first}) cannot be non-positive')
+
+        self.max_size = max_size
+        self.keep_first = keep_first
+        self.llm = llm
+
+        # This condenser relies on the `response_schema` feature, which is not supported by all LLMs
+        if not supports_response_schema(
+            model=self.llm.config.model,
+            custom_llm_provider=self.llm.config.custom_llm_provider,
+        ):
+            raise ValueError(
+                "The LLM model must support the 'response_schema' parameter to use the LLMAttentionCondenser."
+            )
+
+        super().__init__()
+
+    def condense(self, events: list[Event]) -> list[Event]:
+        """If the history is too long, use an LLM to select the most important events."""
+        if len(events) <= self.max_size:
+            return events
+
+        target_size = self.max_size // 2
+        head = events[: self.keep_first]
+
+        events_from_tail = target_size - len(head)
+
+        message: str = """You will be given a list of actions, observations, and thoughts from a coding agent.
+        Each item in the list has an identifier. Please sort the identifiers in order of how important the
+        contents of the item are for the next step of the coding agent's task, from most important to least
+        important."""
+
+        response = self.llm.completion(
+            messages=[
+                {'content': message, 'role': 'user'},
+                *[
+                    {
+                        'content': f'<ID>{e.id}</ID>\n<CONTENT>{e.message}</CONTENT>',
+                        'role': 'user',
+                    }
+                    for e in events
+                ],
+            ],
+            response_format={
+                'type': 'json_schema',
+                'json_schema': {
+                    'name': 'ImportantEventSelection',
+                    'schema': ImportantEventSelection.model_json_schema(),
+                },
+            },
+        )
+
+        response_ids = ImportantEventSelection.model_validate_json(
+            response.choices[0].message.content
+        ).ids
+
+        self.add_metadata('all_event_ids', [event.id for event in events])
+        self.add_metadata('response_ids', response_ids)
+        self.add_metadata('metrics', self.llm.metrics.get())
+
+        # Filter out any IDs from the head and trim the results down
+        head_ids = [event.id for event in head]
+        response_ids = [
+            response_id for response_id in response_ids if response_id not in head_ids
+        ][:events_from_tail]
+
+        # If the response IDs aren't _long_ enough, iterate backwards through the events and add any unfound IDs to the list.
+        for event in reversed(events):
+            if len(response_ids) >= events_from_tail:
+                break
+            if event.id not in response_ids:
+                response_ids.append(event.id)
+
+        # Grab the events associated with the response IDs
+        tail = [event for event in events if event.id in response_ids]
+
+        return head + tail

openhands/runtime/builder/docker.py

@@ -26,6 +26,17 @@ class DockerRuntimeBuilder(RuntimeBuilder):
 
         self.rolling_logger = RollingLogger(max_lines=10)
 
+    @staticmethod
+    def check_buildx():
+        """Check if Docker Buildx is available"""
+        try:
+            result = subprocess.run(
+                ['docker', 'buildx', 'version'], capture_output=True, text=True
+            )
+            return result.returncode == 0
+        except FileNotFoundError:
+            return False
+
     def build(
         self,
         path: str,
@@ -62,6 +73,38 @@ class DockerRuntimeBuilder(RuntimeBuilder):
                 'Docker server version must be >= 18.09 to use BuildKit'
             )
 
+        if not DockerRuntimeBuilder.check_buildx():
+            # when running openhands in a container, there might not be a "docker"
+            # binary available, in which case we need to download docker binary.
+            # since the official openhands app image is built from debian, we use
+            # debian way to install docker binary
+            logger.info(
+                'No docker binary available inside openhands-app container, trying to download online...'
+            )
+            commands = [
+                'apt-get update',
+                'apt-get install -y ca-certificates curl gnupg',
+                'install -m 0755 -d /etc/apt/keyrings',
+                'curl -fsSL https://download.docker.com/linux/debian/gpg -o /etc/apt/keyrings/docker.asc',
+                'chmod a+r /etc/apt/keyrings/docker.asc',
+                'echo \
+                  "deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/docker.asc] https://download.docker.com/linux/debian \
+                  $(. /etc/os-release && echo "$VERSION_CODENAME") stable" | \
+                  tee /etc/apt/sources.list.d/docker.list > /dev/null',
+                'apt-get update',
+                'apt-get install -y docker-ce docker-ce-cli containerd.io docker-buildx-plugin docker-compose-plugin',
+            ]
+            for cmd in commands:
+                try:
+                    subprocess.run(
+                        cmd, shell=True, check=True, stdout=subprocess.DEVNULL
+                    )
+                except subprocess.CalledProcessError as e:
+                    logger.error(f'Image build failed:\n{e}')
+                    logger.error(f'Command output:\n{e.output}')
+                    raise
+            logger.info('Downloaded and installed docker binary')
+
         target_image_hash_name = tags[0]
         target_image_repo, target_image_source_tag = target_image_hash_name.split(':')
         target_image_tag = tags[1].split(':')[1] if len(tags) > 1 else None

openhands/server/session/agent_session.py

@@ -204,8 +204,9 @@ class AgentSession:
                 )
             return
 
+        repo_directory = None
         if selected_repository:
-            await call_sync_from_async(
+            repo_directory = await call_sync_from_async(
                 self.runtime.clone_repo, github_token, selected_repository
             )
         if agent.prompt_manager:
@@ -213,6 +214,10 @@ class AgentSession:
                 self.runtime.get_microagents_from_selected_repo, selected_repository
             )
             agent.prompt_manager.load_microagents(microagents)
+            # Pass GitHub repository information to the prompt manager
+            agent.prompt_manager.set_repository_info(
+                selected_repository, repo_directory
+            )
 
         logger.debug(
             f'Runtime initialized with plugins: {[plugin.name for plugin in self.runtime.plugins]}'
@@ -274,27 +279,25 @@ class AgentSession:
             confirmation_mode=confirmation_mode,
             headless_mode=False,
             status_callback=self._status_callback,
+            initial_state=self._maybe_restore_state(),
         )
 
-        # Note: We now attempt to restore the state from session here,
-        # but if it fails, we fall back to None and still initialize the controller
-        # with a fresh state. That way, the controller will always load events from the event stream
-        # even if the state file was corrupt.
+        return controller
 
+    def _maybe_restore_state(self) -> State | None:
+        """Helper method to handle state restore logic."""
         restored_state = None
+
+        # Attempt to restore the state from session.
+        # Use a heuristic to figure out if we should have a state:
+        # if we have events in the stream.
         try:
             restored_state = State.restore_from_session(self.sid, self.file_store)
+            logger.debug(f'Restored state from session, sid: {self.sid}')
         except Exception as e:
             if self.event_stream.get_latest_event_id() > 0:
                 # if we have events, we should have a state
                 logger.warning(f'State could not be restored: {e}')
-
-        # Set the initial state through the controller.
-        controller.set_initial_state(restored_state, max_iterations, confirmation_mode)
-        if restored_state:
-            logger.debug(f'Restored agent state from session, sid: {self.sid}')
-        else:
-            logger.debug('New session state created.')
-
-        logger.debug('Agent controller initialized.')
-        return controller
+            else:
+                logger.debug('No events found, no state to restore')
+        return restored_state

openhands/storage/google_cloud.py

@@ -21,7 +21,8 @@ class GoogleCloudFileStore(FileStore):
 
     def write(self, path: str, contents: str | bytes) -> None:
         blob = self.bucket.blob(path)
-        with blob.open('w') as f:
+        mode = 'wb' if isinstance(contents, bytes) else 'w'
+        with blob.open(mode) as f:
             f.write(contents)
 
     def read(self, path: str) -> str:

openhands/utils/prompt.py

@@ -1,4 +1,5 @@
 import os
+from dataclasses import dataclass
 from itertools import islice
 
 from jinja2 import Template
@@ -13,6 +14,14 @@ from openhands.microagent import (
 )
 
 
+@dataclass
+class RepositoryInfo:
+    """Information about a GitHub repository that has been cloned."""
+
+    repo_name: str | None = None
+    repo_directory: str | None = None
+
+
 class PromptManager:
     """
     Manages prompt templates and micro-agents for AI interactions.
@@ -32,9 +41,14 @@ class PromptManager:
         prompt_dir: str,
         microagent_dir: str | None = None,
         disabled_microagents: list[str] | None = None,
+        github_repo: str | None = None,
+        repo_directory: str | None = None,
     ):
         self.disabled_microagents: list[str] = disabled_microagents or []
         self.prompt_dir: str = prompt_dir
+        self.repository_info = RepositoryInfo()
+        if github_repo:
+            self.set_repository_info(github_repo, repo_directory)
 
         self.system_template: Template = self._load_template('system_prompt')
         self.user_template: Template = self._load_template('user_prompt')
@@ -91,7 +105,24 @@ class PromptManager:
             if repo_instructions:
                 repo_instructions += '\n\n'
             repo_instructions += microagent.content
-        return self.system_template.render(repo_instructions=repo_instructions).strip()
+
+        return self.system_template.render(
+            repo_instructions=repo_instructions,
+            github_repo=self.repository_info.repo_name,
+            repo_directory=self.repository_info.repo_directory,
+        ).strip()
+
+    def set_repository_info(
+        self, repo_name: str | None, repo_directory: str | None = None
+    ) -> None:
+        """Sets information about the GitHub repository that has been cloned.
+
+        Args:
+            repo_name: The name of the GitHub repository (e.g. 'owner/repo')
+            repo_directory: The directory where the repository has been cloned
+        """
+        self.repository_info.repo_name = repo_name
+        self.repository_info.repo_directory = repo_directory
 
     def get_example_user_message(self) -> str:
         """This is the initial user message provided to the agent

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "openhands-ai"
-version = "0.18.0a0"
+version = "0.19.0"
 description = "OpenHands: Code Less, Make More"
 authors = ["OpenHands"]
 license = "MIT"
@@ -14,7 +14,7 @@ packages = [
 python = "^3.12"
 datasets = "*"
 pandas = "*"
-litellm = "^1.54.1"
+litellm = "^1.55.4"
 google-generativeai = "*" # To use litellm with Gemini Pro API
 google-api-python-client = "*" # For Google Sheets API
 google-auth-httplib2 = "*" # For Google Sheets authentication
@@ -61,7 +61,7 @@ protobuf = "^4.21.6,<5.0.0" # chromadb currently fails on 5.0+
 opentelemetry-api = "1.25.0"
 opentelemetry-exporter-otlp-proto-grpc = "1.25.0"
 modal = ">=0.66.26,<0.72.0"
-runloop-api-client = "0.11.0"
+runloop-api-client = "0.12.0"
 libtmux = ">=0.37,<0.40"
 pygithub = "^2.5.0"
 joblib = "*"
@@ -101,7 +101,6 @@ reportlab = "*"
 [tool.coverage.run]
 concurrency = ["gevent"]
 
-
 [tool.poetry.group.runtime.dependencies]
 jupyterlab = "*"
 notebook = "*"
@@ -130,7 +129,6 @@ ignore = ["D1"]
 [tool.ruff.lint.pydocstyle]
 convention = "google"
 
-
 [tool.poetry.group.evaluation.dependencies]
 streamlit = "*"
 whatthepatch = "*"

tests/test_fileops.py

@@ -10,36 +10,36 @@ WORKSPACE_BASE = 'workspace'
 
 def test_resolve_path():
     assert (
-        files.resolve_path('test.txt', '/workspace')
+        files.resolve_path('test.txt', '/workspace', WORKSPACE_BASE, SANDBOX_PATH_PREFIX)
         == Path(WORKSPACE_BASE) / 'test.txt'
     )
     assert (
-        files.resolve_path('subdir/test.txt', '/workspace')
+        files.resolve_path('subdir/test.txt', '/workspace', WORKSPACE_BASE, SANDBOX_PATH_PREFIX)
         == Path(WORKSPACE_BASE) / 'subdir' / 'test.txt'
     )
     assert (
-        files.resolve_path(Path(SANDBOX_PATH_PREFIX) / 'test.txt', '/workspace')
+        files.resolve_path(Path(SANDBOX_PATH_PREFIX) / 'test.txt', '/workspace', WORKSPACE_BASE, SANDBOX_PATH_PREFIX)
         == Path(WORKSPACE_BASE) / 'test.txt'
     )
     assert (
         files.resolve_path(
-            Path(SANDBOX_PATH_PREFIX) / 'subdir' / 'test.txt', '/workspace'
+            Path(SANDBOX_PATH_PREFIX) / 'subdir' / 'test.txt', '/workspace', WORKSPACE_BASE, SANDBOX_PATH_PREFIX
         )
         == Path(WORKSPACE_BASE) / 'subdir' / 'test.txt'
     )
     assert (
         files.resolve_path(
-            Path(SANDBOX_PATH_PREFIX) / 'subdir' / '..' / 'test.txt', '/workspace'
+            Path(SANDBOX_PATH_PREFIX) / 'subdir' / '..' / 'test.txt', '/workspace', WORKSPACE_BASE, SANDBOX_PATH_PREFIX
         )
         == Path(WORKSPACE_BASE) / 'test.txt'
     )
     with pytest.raises(PermissionError):
-        files.resolve_path(Path(SANDBOX_PATH_PREFIX) / '..' / 'test.txt', '/workspace')
+        files.resolve_path(Path(SANDBOX_PATH_PREFIX) / '..' / 'test.txt', '/workspace', WORKSPACE_BASE, SANDBOX_PATH_PREFIX)
     with pytest.raises(PermissionError):
-        files.resolve_path(Path('..') / 'test.txt', '/workspace')
+        files.resolve_path(Path('..') / 'test.txt', '/workspace', WORKSPACE_BASE, SANDBOX_PATH_PREFIX)
     with pytest.raises(PermissionError):
-        files.resolve_path(Path('/') / 'test.txt', '/workspace')
+        files.resolve_path(Path('/') / 'test.txt', '/workspace', WORKSPACE_BASE, SANDBOX_PATH_PREFIX)
     assert (
-        files.resolve_path('test.txt', '/workspace/test')
+        files.resolve_path('test.txt', '/workspace/test', WORKSPACE_BASE, SANDBOX_PATH_PREFIX)
         == Path(WORKSPACE_BASE) / 'test' / 'test.txt'
     )

tests/unit/test_agent_session.py

@@ -0,0 +1,186 @@
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+from openhands.controller.agent import Agent
+from openhands.controller.agent_controller import AgentController
+from openhands.controller.state.state import State
+from openhands.core.config import AppConfig, LLMConfig
+from openhands.events import EventStream, EventStreamSubscriber
+from openhands.llm import LLM
+from openhands.llm.metrics import Metrics
+from openhands.runtime.base import Runtime
+from openhands.server.session.agent_session import AgentSession
+from openhands.storage.memory import InMemoryFileStore
+
+
+@pytest.fixture
+def mock_agent():
+    """Create a properly configured mock agent with all required nested attributes"""
+    # Create the base mocks
+    agent = MagicMock(spec=Agent)
+    llm = MagicMock(spec=LLM)
+    metrics = MagicMock(spec=Metrics)
+    llm_config = MagicMock(spec=LLMConfig)
+
+    # Configure the LLM config
+    llm_config.model = 'test-model'
+    llm_config.base_url = 'http://test'
+    llm_config.draft_editor = None
+    llm_config.max_message_chars = 1000
+
+    # Set up the chain of mocks
+    llm.metrics = metrics
+    llm.config = llm_config
+    agent.llm = llm
+    agent.name = 'test-agent'
+    agent.sandbox_plugins = []
+
+    return agent
+
+
+@pytest.mark.asyncio
+async def test_agent_session_start_with_no_state(mock_agent):
+    """Test that AgentSession.start() works correctly when there's no state to restore"""
+
+    # Setup
+    file_store = InMemoryFileStore({})
+    session = AgentSession(sid='test-session', file_store=file_store)
+
+    # Create a mock runtime and set it up
+    mock_runtime = MagicMock(spec=Runtime)
+
+    # Mock the runtime creation to set up the runtime attribute
+    async def mock_create_runtime(*args, **kwargs):
+        session.runtime = mock_runtime
+
+    session._create_runtime = AsyncMock(side_effect=mock_create_runtime)
+
+    # Create a mock EventStream with no events
+    mock_event_stream = MagicMock(spec=EventStream)
+    mock_event_stream.get_events.return_value = []
+    mock_event_stream.subscribe = MagicMock()
+    mock_event_stream.get_latest_event_id.return_value = 0
+
+    # Inject the mock event stream into the session
+    session.event_stream = mock_event_stream
+
+    # Create a spy on set_initial_state
+    class SpyAgentController(AgentController):
+        set_initial_state_call_count = 0
+        test_initial_state = None
+
+        def set_initial_state(self, *args, state=None, **kwargs):
+            self.set_initial_state_call_count += 1
+            self.test_initial_state = state
+            super().set_initial_state(*args, state=state, **kwargs)
+
+    # Patch AgentController and State.restore_from_session to fail
+    with patch(
+        'openhands.server.session.agent_session.AgentController', SpyAgentController
+    ), patch(
+        'openhands.server.session.agent_session.EventStream',
+        return_value=mock_event_stream,
+    ), patch(
+        'openhands.controller.state.state.State.restore_from_session',
+        side_effect=Exception('No state found'),
+    ):
+        await session.start(
+            runtime_name='test-runtime',
+            config=AppConfig(),
+            agent=mock_agent,
+            max_iterations=10,
+        )
+
+        # Verify EventStream.subscribe was called with correct parameters
+        mock_event_stream.subscribe.assert_called_with(
+            EventStreamSubscriber.AGENT_CONTROLLER,
+            session.controller.on_event,
+            session.controller.id,
+        )
+
+        # Verify set_initial_state was called once with None as state
+        assert session.controller.set_initial_state_call_count == 1
+        assert session.controller.test_initial_state is None
+        assert session.controller.state.max_iterations == 10
+        assert session.controller.agent.name == 'test-agent'
+        assert session.controller.state.start_id == 0
+        assert session.controller.state.end_id == -1
+        assert session.controller.state.truncation_id == -1
+
+
+@pytest.mark.asyncio
+async def test_agent_session_start_with_restored_state(mock_agent):
+    """Test that AgentSession.start() works correctly when there's a state to restore"""
+
+    # Setup
+    file_store = InMemoryFileStore({})
+    session = AgentSession(sid='test-session', file_store=file_store)
+
+    # Create a mock runtime and set it up
+    mock_runtime = MagicMock(spec=Runtime)
+
+    # Mock the runtime creation to set up the runtime attribute
+    async def mock_create_runtime(*args, **kwargs):
+        session.runtime = mock_runtime
+
+    session._create_runtime = AsyncMock(side_effect=mock_create_runtime)
+
+    # Create a mock EventStream with some events
+    mock_event_stream = MagicMock(spec=EventStream)
+    mock_event_stream.get_events.return_value = []
+    mock_event_stream.subscribe = MagicMock()
+    mock_event_stream.get_latest_event_id.return_value = 5  # Indicate some events exist
+
+    # Inject the mock event stream into the session
+    session.event_stream = mock_event_stream
+
+    # Create a mock restored state
+    mock_restored_state = MagicMock(spec=State)
+    mock_restored_state.start_id = -1
+    mock_restored_state.end_id = -1
+    mock_restored_state.truncation_id = -1
+    mock_restored_state.max_iterations = 5
+
+    # Create a spy on set_initial_state by subclassing AgentController
+    class SpyAgentController(AgentController):
+        set_initial_state_call_count = 0
+        test_initial_state = None
+
+        def set_initial_state(self, *args, state=None, **kwargs):
+            self.set_initial_state_call_count += 1
+            self.test_initial_state = state
+            super().set_initial_state(*args, state=state, **kwargs)
+
+    # Patch AgentController and State.restore_from_session to succeed
+    with patch(
+        'openhands.server.session.agent_session.AgentController', SpyAgentController
+    ), patch(
+        'openhands.server.session.agent_session.EventStream',
+        return_value=mock_event_stream,
+    ), patch(
+        'openhands.controller.state.state.State.restore_from_session',
+        return_value=mock_restored_state,
+    ):
+        await session.start(
+            runtime_name='test-runtime',
+            config=AppConfig(),
+            agent=mock_agent,
+            max_iterations=10,
+        )
+
+        # Verify set_initial_state was called once with the restored state
+        assert session.controller.set_initial_state_call_count == 1
+
+        # Verify EventStream.subscribe was called with correct parameters
+        mock_event_stream.subscribe.assert_called_with(
+            EventStreamSubscriber.AGENT_CONTROLLER,
+            session.controller.on_event,
+            session.controller.id,
+        )
+        assert session.controller.test_initial_state is mock_restored_state
+        assert session.controller.state is mock_restored_state
+        assert session.controller.state.max_iterations == 5
+        assert session.controller.state.start_id == 0
+        assert session.controller.state.end_id == -1
+        assert session.controller.state.truncation_id == -1

tests/unit/test_codeact_agent.py

@@ -1,6 +1,7 @@
 from unittest.mock import Mock
 
 import pytest
+from litellm import ChatCompletionMessageToolCall
 
 from openhands.agenthub.codeact_agent.codeact_agent import CodeActAgent
 from openhands.agenthub.codeact_agent.function_calling import (
@@ -15,6 +16,7 @@ from openhands.agenthub.codeact_agent.function_calling import (
     get_tools,
     response_to_actions,
 )
+from openhands.controller.state.state import State
 from openhands.core.config import AgentConfig, LLMConfig
 from openhands.core.exceptions import FunctionCallNotExistsError
 from openhands.core.message import ImageContent, TextContent
@@ -48,6 +50,15 @@ def agent() -> CodeActAgent:
     return agent
 
 
+@pytest.fixture
+def mock_state() -> State:
+    state = Mock(spec=State)
+    state.history = []
+    state.extra_data = {}
+
+    return state
+
+
 def test_cmd_output_observation_message(agent: CodeActAgent):
     agent.config.function_calling = False
     obs = CmdOutputObservation(
@@ -481,7 +492,7 @@ def test_response_to_actions_invalid_tool():
         response_to_actions(mock_response)
 
 
-def test_step_with_no_pending_actions():
+def test_step_with_no_pending_actions(mock_state: State):
     # Mock the LLM response
     mock_response = Mock()
     mock_response.id = 'mock_id'
@@ -502,16 +513,68 @@ def test_step_with_no_pending_actions():
     agent = CodeActAgent(llm=llm, config=config)
 
     # Test step with no pending actions
-    state = Mock()
-    state.history = []
-    state.latest_user_message = None
-    state.latest_user_message_id = None
-    state.latest_user_message_timestamp = None
-    state.latest_user_message_cause = None
-    state.latest_user_message_timeout = None
-    state.latest_user_message_llm_metrics = None
-    state.latest_user_message_tool_call_metadata = None
+    mock_state.latest_user_message = None
+    mock_state.latest_user_message_id = None
+    mock_state.latest_user_message_timestamp = None
+    mock_state.latest_user_message_cause = None
+    mock_state.latest_user_message_timeout = None
+    mock_state.latest_user_message_llm_metrics = None
+    mock_state.latest_user_message_tool_call_metadata = None
 
-    action = agent.step(state)
+    action = agent.step(mock_state)
     assert isinstance(action, MessageAction)
     assert action.content == 'Task completed'
+
+
+def test_mismatched_tool_call_events(mock_state: State):
+    """Tests that the agent can convert mismatched tool call events (i.e., an observation with no corresponding action) into messages."""
+    agent = CodeActAgent(llm=LLM(LLMConfig()), config=AgentConfig())
+
+    tool_call_metadata = Mock(
+        spec=ToolCallMetadata,
+        model_response=Mock(
+            id='model_response_0',
+            choices=[
+                Mock(
+                    message=Mock(
+                        role='assistant',
+                        content='',
+                        tool_calls=[
+                            Mock(spec=ChatCompletionMessageToolCall, id='tool_call_0')
+                        ],
+                    )
+                )
+            ],
+        ),
+        tool_call_id='tool_call_0',
+        function_name='foo',
+    )
+
+    action = CmdRunAction('foo')
+    action._source = 'agent'
+    action.tool_call_metadata = tool_call_metadata
+
+    observation = CmdOutputObservation(content='', command_id=0, command='foo')
+    observation.tool_call_metadata = tool_call_metadata
+
+    # When both events are provided, the agent should get three messages:
+    # 1. The system message,
+    # 2. The action message, and
+    # 3. The observation message
+    mock_state.history = [action, observation]
+    messages = agent._get_messages(mock_state)
+    assert len(messages) == 3
+
+    # The same should hold if the events are presented out-of-order
+    mock_state.history = [observation, action]
+    messages = agent._get_messages(mock_state)
+    assert len(messages) == 3
+
+    # If only one of the two events is present, then we should just get the system message
+    mock_state.history = [action]
+    messages = agent._get_messages(mock_state)
+    assert len(messages) == 1
+
+    mock_state.history = [observation]
+    messages = agent._get_messages(mock_state)
+    assert len(messages) == 1

tests/unit/test_condenser.py

@@ -1,44 +1,520 @@
-from unittest.mock import Mock, patch
+from datetime import datetime
+from typing import Any
+from unittest.mock import MagicMock
 
 import pytest
 
-from openhands.core.exceptions import LLMResponseError
-from openhands.llm.llm import LLM
-from openhands.memory.condenser import MemoryCondenser
+from openhands.controller.state.state import State
+from openhands.core.config.condenser_config import (
+    AmortizedForgettingCondenserConfig,
+    LLMAttentionCondenserConfig,
+    LLMSummarizingCondenserConfig,
+    NoOpCondenserConfig,
+    ObservationMaskingCondenserConfig,
+    RecentEventsCondenserConfig,
+)
+from openhands.core.config.llm_config import LLMConfig
+from openhands.events.event import Event, EventSource
+from openhands.events.observation.observation import Observation
+from openhands.llm import LLM
+from openhands.memory.condenser import (
+    AmortizedForgettingCondenser,
+    Condenser,
+    ImportantEventSelection,
+    LLMAttentionCondenser,
+    LLMSummarizingCondenser,
+    NoOpCondenser,
+    ObservationMaskingCondenser,
+    RecentEventsCondenser,
+)
+
+
+def create_test_event(
+    message: str, timestamp: datetime | None = None, id: int | None = None
+) -> Event:
+    """Create a simple test event."""
+    event = Event()
+    event._message = message
+    event.timestamp = timestamp if timestamp else datetime.now()
+    if id:
+        event._id = id
+    event._source = EventSource.USER
+    return event
 
 
 @pytest.fixture
-def memory_condenser():
-    return MemoryCondenser()
+def mock_llm() -> LLM:
+    """Mocks an LLM object with a utility function for setting and resetting response contents in unit tests."""
+    # Create a MagicMock for the LLM object
+    mock_llm = MagicMock(
+        spec=LLM,
+        config=MagicMock(
+            spec=LLMConfig, model='gpt-4o', api_key='test_key', custom_llm_provider=None
+        ),
+        metrics=MagicMock(),
+    )
+    _mock_content = None
+
+    # Set a mock message with the mocked content
+    mock_message = MagicMock()
+    mock_message.content = _mock_content
+
+    def set_mock_response_content(content: Any):
+        """Set the mock response for the LLM."""
+        nonlocal mock_message
+        mock_message.content = content
+
+    mock_choice = MagicMock()
+    mock_choice.message = mock_message
+
+    mock_response = MagicMock()
+    mock_response.choices = [mock_choice]
+
+    mock_llm.completion.return_value = mock_response
+
+    # Attach helper methods to the mock object
+    mock_llm.set_mock_response_content = set_mock_response_content
+
+    return mock_llm
 
 
 @pytest.fixture
-def mock_llm():
-    return Mock(spec=LLM)
+def mock_state() -> State:
+    """Mocks a State object with the only parameters needed for testing condensers: history and extra_data."""
+    mock_state = MagicMock(spec=State)
+    mock_state.history = []
+    mock_state.extra_data = {}
+
+    return mock_state
 
 
-def test_condense_success(memory_condenser, mock_llm):
-    mock_llm.completion.return_value = {
-        'choices': [{'message': {'content': 'Condensed memory'}}]
-    }
-    result = memory_condenser.condense('Summarize this', mock_llm)
-    assert result == 'Condensed memory'
-    mock_llm.completion.assert_called_once_with(
-        messages=[{'content': 'Summarize this', 'role': 'user'}]
+def test_noop_condenser_from_config():
+    """Test that the NoOpCondenser objects can be made from config."""
+    config = NoOpCondenserConfig()
+    condenser = Condenser.from_config(config)
+
+    assert isinstance(condenser, NoOpCondenser)
+
+
+def test_noop_condenser():
+    """Test that NoOpCondensers preserve their input events."""
+    events = [
+        create_test_event('Event 1'),
+        create_test_event('Event 2'),
+        create_test_event('Event 3'),
+    ]
+
+    mock_state = MagicMock()
+    mock_state.history = events
+
+    condenser = NoOpCondenser()
+    result = condenser.condensed_history(mock_state)
+
+    assert result == events
+
+
+def test_observation_masking_condenser_from_config():
+    """Test that ObservationMaskingCondenser objects can be made from config."""
+    attention_window = 5
+    config = ObservationMaskingCondenserConfig(attention_window=attention_window)
+    condenser = Condenser.from_config(config)
+
+    assert isinstance(condenser, ObservationMaskingCondenser)
+    assert condenser.attention_window == attention_window
+
+
+def test_observation_masking_condenser_respects_attention_window(mock_state):
+    """Test that ObservationMaskingCondenser only masks events outside the attention window."""
+    attention_window = 3
+    condenser = ObservationMaskingCondenser(attention_window=attention_window)
+
+    events = [
+        create_test_event('Event 1'),
+        Observation('Observation 1'),
+        create_test_event('Event 3'),
+        create_test_event('Event 4'),
+        Observation('Observation 2'),
+    ]
+
+    mock_state.history = events
+    result = condenser.condensed_history(mock_state)
+
+    assert len(result) == len(events)
+
+    for index, (event, condensed_event) in enumerate(zip(events, result)):
+        # If we're outside the attention window, observations should be masked.
+        if index < len(events) - attention_window:
+            if isinstance(event, Observation):
+                assert '<MASKED>' in str(condensed_event)
+
+        # If we're within the attention window, events are unchanged.
+        else:
+            assert event == condensed_event
+
+
+def test_recent_events_condenser_from_config():
+    """Test that RecentEventsCondenser objects can be made from config."""
+    max_events = 5
+    keep_first = True
+    config = RecentEventsCondenserConfig(keep_first=keep_first, max_events=max_events)
+    condenser = Condenser.from_config(config)
+
+    assert isinstance(condenser, RecentEventsCondenser)
+    assert condenser.max_events == max_events
+    assert condenser.keep_first == keep_first
+
+
+def test_recent_events_condenser():
+    """Test that RecentEventsCondensers keep just the most recent events."""
+    events = [
+        create_test_event('Event 1'),
+        create_test_event('Event 2'),
+        create_test_event('Event 3'),
+        create_test_event('Event 4'),
+        create_test_event('Event 5'),
+    ]
+
+    mock_state = MagicMock()
+    mock_state.history = events
+
+    # If the max_events are larger than the number of events, equivalent to a NoOpCondenser.
+    condenser = RecentEventsCondenser(max_events=len(events))
+    result = condenser.condensed_history(mock_state)
+
+    assert result == events
+
+    # If the max_events are smaller than the number of events, only keep the last few.
+    max_events = 2
+    condenser = RecentEventsCondenser(max_events=max_events)
+    result = condenser.condensed_history(mock_state)
+
+    assert len(result) == max_events
+    assert result[0]._message == 'Event 4'
+    assert result[1]._message == 'Event 5'
+
+    # If the keep_first flag is set, the first event will always be present.
+    keep_first = 1
+    max_events = 2
+    condenser = RecentEventsCondenser(keep_first=keep_first, max_events=max_events)
+    result = condenser.condensed_history(mock_state)
+
+    assert len(result) == max_events
+    assert result[0]._message == 'Event 1'
+    assert result[1]._message == 'Event 5'
+
+    # We should be able to keep more of the initial events.
+    keep_first = 2
+    max_events = 3
+    condenser = RecentEventsCondenser(keep_first=keep_first, max_events=max_events)
+    result = condenser.condensed_history(mock_state)
+
+    assert len(result) == max_events
+    assert result[0]._message == 'Event 1'
+    assert result[1]._message == 'Event 2'
+    assert result[2]._message == 'Event 5'
+
+
+def test_llm_condenser_from_config():
+    """Test that LLMCondensers can be made from config."""
+    config = LLMSummarizingCondenserConfig(
+        llm_config=LLMConfig(
+            model='gpt-4o',
+            api_key='test_key',
+        )
+    )
+    condenser = Condenser.from_config(config)
+
+    assert isinstance(condenser, LLMSummarizingCondenser)
+    assert condenser.llm.config.model == 'gpt-4o'
+    assert condenser.llm.config.api_key == 'test_key'
+
+
+def test_llm_condenser(mock_llm, mock_state):
+    """Test that LLMCondensers use the LLM to generate a summary event."""
+    events = [
+        create_test_event('Event 1'),
+        create_test_event('Event 2'),
+    ]
+    mock_state.history = events
+
+    mock_llm.metrics = MagicMock()
+    mock_llm.metrics.get.return_value = {'test_metric': 1.0}
+
+    mock_llm.set_mock_response_content('Summary of events')
+
+    condenser = LLMSummarizingCondenser(llm=mock_llm)
+    result = condenser.condensed_history(mock_state)
+
+    assert len(result) == 1
+    assert result[0].content == 'Summary of events'
+
+    # Verify LLM was called with correct prompt.
+    mock_llm.completion.assert_called_once()
+    call_args = mock_llm.completion.call_args[1]
+    assert 'messages' in call_args
+    assert len(call_args['messages']) == 1
+    assert 'Event 1' in call_args['messages'][0]['content']
+    assert 'Event 2' in call_args['messages'][0]['content']
+
+    # Verify metrics were added to state
+    assert 'condenser_meta' in mock_state.extra_data
+    assert len(mock_state.extra_data['condenser_meta']) == 1
+    assert mock_state.extra_data['condenser_meta'][0]['metrics'] == {'test_metric': 1.0}
+
+
+def test_llm_condenser_error():
+    """Test that LLM errors are propagated during condensation."""
+    events = [create_test_event('Event 1', datetime(2024, 1, 1, 10, 0))]
+
+    mock_state = MagicMock()
+    mock_state.history = events
+
+    mock_llm = MagicMock()
+    mock_llm.completion.side_effect = Exception('LLM error')
+
+    condenser = LLMSummarizingCondenser(llm=mock_llm)
+
+    try:
+        condenser.condensed_history(mock_state)
+        raise AssertionError('Expected exception was not raised.')
+    except Exception as e:
+        assert str(e) == 'LLM error'
+
+
+def test_amortized_forgetting_condenser_from_config():
+    """Test that AmortizedForgettingCondenser objects can be made from config."""
+    max_size = 50
+    keep_first = 10
+    config = AmortizedForgettingCondenserConfig(
+        max_size=max_size, keep_first=keep_first
+    )
+    condenser = Condenser.from_config(config)
+
+    assert isinstance(condenser, AmortizedForgettingCondenser)
+    assert condenser.max_size == max_size
+    assert condenser.keep_first == keep_first
+
+
+def test_amortized_forgetting_condenser_invalid_config():
+    """Test that AmortizedForgettingCondenser raises error when keep_first > max_size."""
+    pytest.raises(ValueError, AmortizedForgettingCondenser, max_size=4, keep_first=2)
+    pytest.raises(ValueError, AmortizedForgettingCondenser, max_size=0)
+    pytest.raises(ValueError, AmortizedForgettingCondenser, keep_first=-1)
+
+
+def test_amortized_forgetting_condenser_grows_to_max_size():
+    """Test that AmortizedForgettingCondenser correctly maintains an event context up to max size."""
+    max_size = 15
+    condenser = AmortizedForgettingCondenser(max_size=max_size)
+
+    mock_state = MagicMock()
+    mock_state.extra_data = {}
+    mock_state.history = []
+
+    for i in range(max_size):
+        event = create_test_event(f'Event {i}')
+        mock_state.history.append(event)
+        results = condenser.condensed_history(mock_state)
+        assert len(results) == i + 1
+
+
+def test_amortized_forgetting_condenser_forgets_when_larger_than_max_size():
+    """Test that the AmortizedForgettingCondenser forgets events when the context grows too large."""
+    max_size = 2
+    condenser = AmortizedForgettingCondenser(max_size=max_size)
+
+    mock_state = MagicMock()
+    mock_state.extra_data = {}
+    mock_state.history = []
+
+    for i in range(max_size * 10):
+        event = create_test_event(f'Event {i}')
+        mock_state.history.append(event)
+        results = condenser.condensed_history(mock_state)
+
+        # The last event in the results is always the event we just added.
+        assert results[-1] == event
+
+        # The number of results should bounce back and forth between 1, 2, 1, 2, ...
+        assert len(results) == (i % 2) + 1
+
+
+def test_amortized_forgetting_condenser_keeps_first_events():
+    """Test that the AmortizedForgettingCondenser keeps the right number of initial events when forgetting."""
+    max_size = 4
+    keep_first = 1
+    condenser = AmortizedForgettingCondenser(max_size=max_size, keep_first=keep_first)
+
+    first_event = create_test_event('Event 0')
+
+    mock_state = MagicMock()
+    mock_state.extra_data = {}
+    mock_state.history = [first_event]
+
+    for i in range(max_size * 10):
+        event = create_test_event(f'Event {i+1}', datetime(2024, 1, 1, 10, i + 1))
+        mock_state.history.append(event)
+        results = condenser.condensed_history(mock_state)
+
+        # The last event is always the event we just added.
+        assert results[-1] == event
+
+        # The first event is always the first event.
+        assert results[0] == first_event
+
+        # The number of results should bounce back between 2, 3, 4, 2, 3, 4, ...
+        print(len(results))
+        assert len(results) == (i % 3) + 2
+
+
+def test_llm_attention_condenser_from_config():
+    """Test that LLMAttentionCondenser objects can be made from config."""
+    config = LLMAttentionCondenserConfig(
+        max_size=50,
+        keep_first=10,
+        llm_config=LLMConfig(
+            model='gpt-4o',
+            api_key='test_key',
+        ),
+    )
+    condenser = Condenser.from_config(config)
+
+    assert isinstance(condenser, LLMAttentionCondenser)
+    assert condenser.llm.config.model == 'gpt-4o'
+    assert condenser.llm.config.api_key == 'test_key'
+    assert condenser.max_size == 50
+    assert condenser.keep_first == 10
+
+
+def test_llm_attention_condenser_invalid_config():
+    """Test that LLMAttentionCondenser raises an error if the configured LLM doesn't support response schema."""
+    config = LLMAttentionCondenserConfig(
+        max_size=50,
+        keep_first=10,
+        llm_config=LLMConfig(
+            model='claude-2',  # Older model that doesn't support response schema
+            api_key='test_key',
+        ),
     )
 
-
-def test_condense_exception(memory_condenser, mock_llm):
-    mock_llm.completion.side_effect = LLMResponseError('LLM error')
-    with pytest.raises(LLMResponseError, match='LLM error'):
-        memory_condenser.condense('Summarize this', mock_llm)
+    pytest.raises(ValueError, LLMAttentionCondenser.from_config, config)
 
 
-@patch('openhands.memory.condenser.logger')
-def test_condense_logs_error(mock_logger, memory_condenser, mock_llm):
-    mock_llm.completion.side_effect = LLMResponseError('LLM error')
-    with pytest.raises(LLMResponseError):
-        memory_condenser.condense('Summarize this', mock_llm)
-    mock_logger.error.assert_called_once_with(
-        'Error condensing thoughts: %s', 'LLM error', exc_info=False
-    )
+def test_llm_attention_condenser_keeps_first_events(mock_llm, mock_state):
+    """Test that the LLMAttentionCondenser keeps the right number of initial events when forgetting."""
+    max_size = 4
+    condenser = LLMAttentionCondenser(max_size=max_size, keep_first=1, llm=mock_llm)
+
+    first_event = create_test_event('Event 0', id=0)
+    mock_state.history.append(first_event)
+
+    for i in range(max_size * 10):
+        event = create_test_event(f'Event {i+1}', id=i + 1)
+        mock_state.history.append(event)
+
+        mock_llm.set_mock_response_content(
+            ImportantEventSelection(
+                ids=[event.id for event in mock_state.history]
+            ).model_dump_json()
+        )
+        results = condenser.condensed_history(mock_state)
+
+        # The first event is always the first event.
+        assert results[0] == first_event
+
+
+def test_llm_attention_condenser_grows_to_max_size(mock_llm, mock_state):
+    """Test that LLMAttentionCondenser correctly maintains an event context up to max size."""
+    max_size = 15
+    condenser = LLMAttentionCondenser(max_size=max_size, llm=mock_llm)
+
+    for i in range(max_size):
+        event = create_test_event(f'Event {i}')
+        mock_state.history.append(event)
+        mock_llm.set_mock_response_content(
+            ImportantEventSelection(ids=[event.id for event in mock_state.history])
+        )
+        results = condenser.condensed_history(mock_state)
+        assert len(results) == i + 1
+
+
+def test_llm_attention_condenser_forgets_when_larger_than_max_size(
+    mock_llm, mock_state
+):
+    """Test that the LLMAttentionCondenser forgets events when the context grows too large."""
+    max_size = 2
+    condenser = LLMAttentionCondenser(max_size=max_size, llm=mock_llm)
+
+    for i in range(max_size * 10):
+        event = create_test_event(f'Event {i}', id=i)
+        mock_state.history.append(event)
+
+        mock_llm.set_mock_response_content(
+            ImportantEventSelection(
+                ids=[event.id for event in mock_state.history]
+            ).model_dump_json()
+        )
+
+        results = condenser.condensed_history(mock_state)
+
+        # The number of results should bounce back and forth between 1, 2, 1, 2, ...
+        assert len(results) == (i % 2) + 1
+
+
+def test_llm_attention_condenser_handles_events_outside_history(mock_llm, mock_state):
+    """Test that the LLMAttentionCondenser handles event IDs that aren't from the event history."""
+    max_size = 2
+    condenser = LLMAttentionCondenser(max_size=max_size, llm=mock_llm)
+
+    for i in range(max_size * 10):
+        event = create_test_event(f'Event {i}', id=i)
+        mock_state.history.append(event)
+
+        mock_llm.set_mock_response_content(
+            ImportantEventSelection(
+                ids=[event.id for event in mock_state.history] + [-1, -2, -3, -4]
+            ).model_dump_json()
+        )
+        results = condenser.condensed_history(mock_state)
+
+        # The number of results should bounce back and forth between 1, 2, 1, 2, ...
+        assert len(results) == (i % 2) + 1
+
+
+def test_llm_attention_condenser_handles_too_many_events(mock_llm, mock_state):
+    """Test that the LLMAttentionCondenser handles when the response contains too many event IDs."""
+    max_size = 2
+    condenser = LLMAttentionCondenser(max_size=max_size, llm=mock_llm)
+
+    for i in range(max_size * 10):
+        event = create_test_event(f'Event {i}', id=i)
+        mock_state.history.append(event)
+        mock_llm.set_mock_response_content(
+            ImportantEventSelection(
+                ids=[event.id for event in mock_state.history]
+                + [event.id for event in mock_state.history]
+            ).model_dump_json()
+        )
+        results = condenser.condensed_history(mock_state)
+
+        # The number of results should bounce back and forth between 1, 2, 1, 2, ...
+        assert len(results) == (i % 2) + 1
+
+
+def test_llm_attention_condenser_handles_too_few_events(mock_llm, mock_state):
+    """Test that the LLMAttentionCondenser handles when the response contains too few event IDs."""
+    max_size = 2
+    condenser = LLMAttentionCondenser(max_size=max_size, llm=mock_llm)
+
+    for i in range(max_size * 10):
+        event = create_test_event(f'Event {i}', id=i)
+        mock_state.history.append(event)
+
+        mock_llm.set_mock_response_content(
+            ImportantEventSelection(ids=[]).model_dump_json()
+        )
+
+        results = condenser.condensed_history(mock_state)
+
+        # The number of results should bounce back and forth between 1, 2, 1, 2, ...
+        assert len(results) == (i % 2) + 1

tests/unit/test_config.py

@@ -13,6 +13,9 @@ from openhands.core.config import (
     load_from_env,
     load_from_toml,
 )
+from openhands.core.config.condenser_config import (
+    NoOpCondenserConfig,
+)
 from openhands.core.logger import openhands_logger
 
 
@@ -618,6 +621,13 @@ def test_cache_dir_creation(default_config, tmpdir):
     assert os.path.exists(default_config.cache_dir)
 
 
+def test_agent_config_condenser_default():
+    """Test that default agent condenser is NoOpCondenser."""
+    config = AppConfig()
+    agent_config = config.get_agent_config()
+    assert isinstance(agent_config.condenser, NoOpCondenserConfig)
+
+
 def test_api_keys_repr_str():
     # Test LLMConfig
     llm_config = LLMConfig(

tests/unit/test_prompt_caching.py

@@ -75,7 +75,7 @@ def test_get_messages(codeact_agent: CodeActAgent):
 
     codeact_agent.reset()
     messages = codeact_agent._get_messages(
-        Mock(history=history, max_iterations=5, iteration=0)
+        Mock(history=history, max_iterations=5, iteration=0, extra_data={})
     )
 
     assert (
@@ -111,7 +111,7 @@ def test_get_messages_prompt_caching(codeact_agent: CodeActAgent):
 
     codeact_agent.reset()
     messages = codeact_agent._get_messages(
-        Mock(history=history, max_iterations=10, iteration=5)
+        Mock(history=history, max_iterations=10, iteration=5, extra_data={})
     )
 
     # Check that only the last two user messages have cache_prompt=True
@@ -144,6 +144,7 @@ def test_prompt_caching_headers(codeact_agent: CodeActAgent):
     mock_state.history = history
     mock_state.max_iterations = 5
     mock_state.iteration = 0
+    mock_state.extra_data = {}
 
     codeact_agent.reset()
 

tests/unit/test_prompt_manager.py

@@ -5,7 +5,7 @@ import pytest
 
 from openhands.core.message import Message, TextContent
 from openhands.microagent import BaseMicroAgent
-from openhands.utils.prompt import PromptManager
+from openhands.utils.prompt import PromptManager, RepositoryInfo
 
 
 @pytest.fixture
@@ -39,6 +39,7 @@ only respond with a message telling them how smart they are
     with open(os.path.join(prompt_dir, 'micro', f'{microagent_name}.md'), 'w') as f:
         f.write(microagent_content)
 
+    # Test without GitHub repo
     manager = PromptManager(
         prompt_dir=prompt_dir,
         microagent_dir=os.path.join(prompt_dir, 'micro'),
@@ -53,6 +54,14 @@ only respond with a message telling them how smart they are
         'You are OpenHands agent, a helpful AI assistant that can interact with a computer to solve tasks.'
         in manager.get_system_message()
     )
+    assert '<REPOSITORY_INFO>' not in manager.get_system_message()
+
+    # Test with GitHub repo
+    manager.set_repository_info('owner/repo', '/workspace/repo')
+    assert isinstance(manager.get_system_message(), str)
+    assert '<REPOSITORY_INFO>' in manager.get_system_message()
+    assert 'owner/repo' in manager.get_system_message()
+    assert '/workspace/repo' in manager.get_system_message()
 
     assert isinstance(manager.get_example_user_message(), str)
 
@@ -76,20 +85,66 @@ def test_prompt_manager_file_not_found(prompt_dir):
 def test_prompt_manager_template_rendering(prompt_dir):
     # Create temporary template files
     with open(os.path.join(prompt_dir, 'system_prompt.j2'), 'w') as f:
-        f.write('System prompt: bar')
+        f.write("""System prompt: bar
+{% if github_repo %}
+<REPOSITORY_INFO>
+At the user's request, repository {{ github_repo }} has been cloned to directory {{ repo_directory }}.
+</REPOSITORY_INFO>
+{% endif %}
+{{ repo_instructions }}""")
     with open(os.path.join(prompt_dir, 'user_prompt.j2'), 'w') as f:
         f.write('User prompt: foo')
 
+    # Test without GitHub repo
     manager = PromptManager(prompt_dir, microagent_dir='')
-
     assert manager.get_system_message() == 'System prompt: bar'
     assert manager.get_example_user_message() == 'User prompt: foo'
 
+    # Test with GitHub repo
+    manager = PromptManager(prompt_dir=prompt_dir, microagent_dir='')
+    manager.set_repository_info('owner/repo', '/workspace/repo')
+    system_msg = manager.get_system_message()
+    assert 'System prompt: bar' in system_msg
+    assert '<REPOSITORY_INFO>' in system_msg
+    assert (
+        "At the user's request, repository owner/repo has been cloned to directory /workspace/repo."
+        in system_msg
+    )
+    assert '</REPOSITORY_INFO>' in system_msg
+    assert manager.get_example_user_message() == 'User prompt: foo'
+
     # Clean up temporary files
     os.remove(os.path.join(prompt_dir, 'system_prompt.j2'))
     os.remove(os.path.join(prompt_dir, 'user_prompt.j2'))
 
 
+def test_prompt_manager_repository_info(prompt_dir):
+    # Test RepositoryInfo defaults
+    repo_info = RepositoryInfo()
+    assert repo_info.repo_name is None
+    assert repo_info.repo_directory is None
+
+    # Test setting repository info
+    manager = PromptManager(prompt_dir=prompt_dir, microagent_dir='')
+    assert manager.repository_info.repo_name is None
+    assert manager.repository_info.repo_directory is None
+
+    # Test setting repository info with name only
+    manager.set_repository_info('owner/repo')
+    assert manager.repository_info.repo_name == 'owner/repo'
+    assert manager.repository_info.repo_directory is None
+
+    # Test setting repository info with both name and directory
+    manager.set_repository_info('owner/repo2', '/workspace/repo2')
+    assert manager.repository_info.repo_name == 'owner/repo2'
+    assert manager.repository_info.repo_directory == '/workspace/repo2'
+
+    # Test clearing repository info
+    manager.set_repository_info(None)
+    assert manager.repository_info.repo_name is None
+    assert manager.repository_info.repo_directory is None
+
+
 def test_prompt_manager_disabled_microagents(prompt_dir):
     # Create test microagent files
     microagent1_name = 'test_microagent1'