empty commit

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

evaluation/EDA/run_infer.py

@@ -35,7 +35,8 @@ def codeact_user_response_eda(state: State) -> str:
 
     # retrieve the latest model message from history
     if state.history:
-        model_guess = state.get_last_agent_message()
+        last_agent_message = state.get_last_agent_message()
+        model_guess = last_agent_message.content if last_agent_message else ''
 
     assert game is not None, 'Game is not initialized.'
     msg = game.generate_user_response(model_guess)
@@ -140,7 +141,8 @@ def process_instance(
     if state is None:
         raise ValueError('State should not be None.')
 
-    final_message = state.get_last_agent_message()
+    last_agent_message = state.get_last_agent_message()
+    final_message = last_agent_message.content if last_agent_message else ''
 
     logger.info(f'Final message: {final_message} | Ground truth: {instance["text"]}')
     test_result = game.reward()

evaluation/gorilla/run_infer.py

@@ -102,7 +102,8 @@ def process_instance(
         raise ValueError('State should not be None.')
 
     # retrieve the last message from the agent
-    model_answer_raw = state.get_last_agent_message()
+    last_agent_message = state.get_last_agent_message()
+    model_answer_raw = last_agent_message.content if last_agent_message else ''
 
     # attempt to parse model_answer
     ast_eval_fn = instance['ast_eval']

evaluation/swe_bench/eval_infer.py

@@ -83,6 +83,7 @@ def get_config(instance: pd.Series) -> AppConfig:
             timeout=1800,
             api_key=os.environ.get('ALLHANDS_API_KEY', None),
             remote_runtime_api_url=os.environ.get('SANDBOX_REMOTE_RUNTIME_API_URL'),
+            remote_runtime_init_timeout=1800,
         ),
         # do not mount workspace
         workspace_base=None,

evaluation/swe_bench/run_infer.py

@@ -146,6 +146,7 @@ def get_config(
             api_key=os.environ.get('ALLHANDS_API_KEY', None),
             remote_runtime_api_url=os.environ.get('SANDBOX_REMOTE_RUNTIME_API_URL'),
             keep_remote_runtime_alive=False,
+            remote_runtime_init_timeout=1800,
         ),
         # do not mount workspace
         workspace_base=None,

evaluation/toolqa/run_infer.py

@@ -127,7 +127,8 @@ def process_instance(instance: Any, metadata: EvalMetadata, reset_logger: bool =
         raise ValueError('State should not be None.')
 
     # retrieve the last message from the agent
-    model_answer_raw = state.get_last_agent_message()
+    last_agent_message = state.get_last_agent_message()
+    model_answer_raw = last_agent_message.content if last_agent_message else ''
 
     # attempt to parse model_answer
     correct = eval_answer(str(model_answer_raw), str(answer))

frontend/src/components/project-menu/ProjectMenuCard.tsx

@@ -43,10 +43,7 @@ export function ProjectMenuCard({
     posthog.capture("push_to_github_button_clicked");
     const rawEvent = {
       content: `
-Let's push the code to GitHub.
-If we're currently on the openhands-workspace branch, please create a new branch with a descriptive name.
-Commit any changes and push them to the remote repository.
-Finally, open up a pull request using the GitHub API and the token in the GITHUB_TOKEN environment variable, then show me the URL of the pull request.
+Please push the changes to GitHub and open a pull request.
 `,
       imageUrls: [],
       timestamp: new Date().toISOString(),

frontend/src/services/api.ts

@@ -63,7 +63,7 @@ export async function request(
   } catch (e) {
     onFail(`Error fetching ${url}`);
   }
-  if (response?.status === 401) {
+  if (response?.status === 401 && !url.startsWith("/api/authenticate")) {
     await request(
       "/api/authenticate",
       {

frontend/src/utils/verified-models.ts

@@ -1,10 +1,6 @@
 // Here are the list of verified models and providers that we know work well with OpenHands.
 export const VERIFIED_PROVIDERS = ["openai", "azure", "anthropic"];
-export const VERIFIED_MODELS = [
-  "gpt-4o",
-  "claude-3-5-sonnet-20240620",
-  "claude-3-5-sonnet-20241022",
-];
+export const VERIFIED_MODELS = ["gpt-4o", "claude-3-5-sonnet-20241022"];
 
 // LiteLLM does not return OpenAI models with the provider, so we list them here to set them ourselves for consistency
 // (e.g., they return `gpt-4o` instead of `openai/gpt-4o`)
@@ -23,11 +19,8 @@ export const VERIFIED_OPENAI_MODELS = [
 export const VERIFIED_ANTHROPIC_MODELS = [
   "claude-2",
   "claude-2.1",
-  "claude-3-5-sonnet-20241022",
   "claude-3-5-sonnet-20240620",
   "claude-3-haiku-20240307",
   "claude-3-opus-20240229",
   "claude-3-sonnet-20240229",
-  "claude-instant-1",
-  "claude-instant-1.2",
 ];

openhands/agenthub/codeact_agent/codeact_agent.py

@@ -39,7 +39,6 @@ from openhands.runtime.plugins import (
     JupyterRequirement,
     PluginRequirement,
 )
-from openhands.utils.microagent import MicroAgent
 from openhands.utils.prompt import PromptManager
 
 
@@ -86,16 +85,6 @@ class CodeActAgent(Agent):
         super().__init__(llm, config)
         self.reset()
 
-        self.micro_agent = (
-            MicroAgent(
-                os.path.join(
-                    os.path.dirname(__file__), 'micro', f'{config.micro_agent_name}.md'
-                )
-            )
-            if config.micro_agent_name
-            else None
-        )
-
         self.function_calling_active = self.config.function_calling
         if self.function_calling_active and not self.llm.is_function_calling_active():
             logger.warning(
@@ -105,7 +94,6 @@ class CodeActAgent(Agent):
             self.function_calling_active = False
 
         if self.function_calling_active:
-            # Function calling mode
             self.tools = codeact_function_calling.get_tools(
                 codeact_enable_browsing=self.config.codeact_enable_browsing,
                 codeact_enable_jupyter=self.config.codeact_enable_jupyter,
@@ -114,18 +102,17 @@ class CodeActAgent(Agent):
             logger.debug(
                 f'TOOLS loaded for CodeActAgent: {json.dumps(self.tools, indent=2)}'
             )
-            self.system_prompt = codeact_function_calling.SYSTEM_PROMPT
-            self.initial_user_message = None
+            self.prompt_manager = PromptManager(
+                microagent_dir=os.path.join(os.path.dirname(__file__), 'micro'),
+                prompt_dir=os.path.join(os.path.dirname(__file__), 'prompts', 'tools'),
+            )
         else:
-            # Non-function-calling mode
             self.action_parser = CodeActResponseParser()
             self.prompt_manager = PromptManager(
-                prompt_dir=os.path.join(os.path.dirname(__file__)),
+                microagent_dir=os.path.join(os.path.dirname(__file__), 'micro'),
+                prompt_dir=os.path.join(os.path.dirname(__file__), 'prompts', 'default'),
                 agent_skills_docs=AgentSkillsRequirement.documentation,
-                micro_agent=self.micro_agent,
             )
-            self.system_prompt = self.prompt_manager.system_message
-            self.initial_user_message = self.prompt_manager.initial_user_message
 
         self.pending_actions: deque[Action] = deque()
 
@@ -337,8 +324,8 @@ class CodeActAgent(Agent):
             return self.pending_actions.popleft()
 
         # if we're done, go back
-        last_user_message = state.get_last_user_message()
-        if last_user_message and last_user_message.strip() == '/exit':
+        latest_user_message = state.get_last_user_message()
+        if latest_user_message and latest_user_message.content.strip() == '/exit':
             return AgentFinishAction()
 
         # prepare what we want to send to the LLM
@@ -403,17 +390,19 @@ class CodeActAgent(Agent):
                 role='system',
                 content=[
                     TextContent(
-                        text=self.system_prompt,
-                        cache_prompt=self.llm.is_caching_prompt_active(),  # Cache system prompt
+                        text=self.prompt_manager.get_system_message(),
+                        cache_prompt=self.llm.is_caching_prompt_active(),
                     )
                 ],
             )
         ]
-        if self.initial_user_message:
+        example_message = self.prompt_manager.get_example_user_message()
+        if example_message:
             messages.append(
                 Message(
                     role='user',
-                    content=[TextContent(text=self.initial_user_message)],
+                    content=[TextContent(text=example_message)],
+                    cache_prompt=self.llm.is_caching_prompt_active(),
                 )
             )
 
@@ -462,8 +451,9 @@ class CodeActAgent(Agent):
                 pending_tool_call_action_messages.pop(response_id)
 
             for message in messages_to_add:
-                # add regular message
                 if message:
+                    if message.role == 'user':
+                        self.prompt_manager.enhance_message(message)
                     # handle error if the message is the SAME role as the previous message
                     # litellm.exceptions.BadRequestError: litellm.BadRequestError: OpenAIException - Error code: 400 - {'detail': 'Only supports u/a/u/a/u...'}
                     # there shouldn't be two consecutive messages from the same role
@@ -493,23 +483,6 @@ class CodeActAgent(Agent):
                         break
 
         if not self.function_calling_active:
-            # The latest user message is important:
-            # we want to remind the agent of the environment constraints
-            latest_user_message = next(
-                islice(
-                    (
-                        m
-                        for m in reversed(messages)
-                        if m.role == 'user'
-                        and any(isinstance(c, TextContent) for c in m.content)
-                    ),
-                    1,
-                ),
-                None,
-            )
-            # do not add this for function calling
-            if latest_user_message:
-                reminder_text = f'\n\nENVIRONMENT REMINDER: You have {state.max_iterations - state.iteration} turns left to complete the task. When finished reply with <finish></finish>.'
-                latest_user_message.content.append(TextContent(text=reminder_text))
+            self.prompt_manager.add_turns_left_reminder(messages, state)
 
         return messages

openhands/agenthub/codeact_agent/function_calling.py

@@ -25,14 +25,6 @@ from openhands.events.action import (
 )
 from openhands.events.tool import ToolCallMetadata
 
-SYSTEM_PROMPT = """You are OpenHands agent, a helpful AI assistant that can interact with a computer to solve tasks.
-<IMPORTANT>
-* If user provides a path, you should NOT assume it's relative to the current working directory. Instead, you should explore the file system to find the file before working on it.
-* 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>
-"""
-
 _BASH_DESCRIPTION = """Execute a bash command in the terminal.
 * Long running commands: For commands that may run indefinitely, it should be run in the background and the output should be redirected to a file, e.g. command = `python3 app.py > server.log 2>&1 &`.
 * Interactive: If a bash command returns exit code `-1`, this means the process is not yet finished. The assistant must then send a second call to terminal with an empty `command` (which will retrieve any additional logs), or it can send additional text (set `command` to the text) to STDIN of the running process, or it can send command=`ctrl+c` to interrupt the process.

openhands/agenthub/codeact_agent/micro/flarglebargle.md

@@ -0,0 +1,9 @@
+---
+name: flarglebargle
+agent: CodeActAgent
+triggers:
+- flarglebargle
+---
+
+IMPORTANT! The user has said the magic word "flarglebargle". You must
+only respond with a message telling them how smart they are

openhands/agenthub/codeact_agent/micro/github.md

@@ -1,69 +1,31 @@
 ---
 name: github
 agent: CodeActAgent
-require_env_var:
-    SANDBOX_ENV_GITHUB_TOKEN: "Create a GitHub Personal Access Token (https://docs.github.com/en/authentication/keeping-your-account-and-data-secure/managing-your-personal-access-tokens) and set it as SANDBOX_GITHUB_TOKEN in your environment variables."
+triggers:
+- github
+- git
 ---
 
-# How to Interact with Github
+You have access to an environment variable, `GITHUB_TOKEN`, which allows you to interact with
+the GitHub API.
 
-## Environment Variable Available
+You can use `curl` with the `GITHUB_TOKEN` to interact with GitHub's API.
+ALWAYS use the GitHub API for operations instead of a web browser.
 
-- `GITHUB_TOKEN`: A read-only token for Github.
-
-## Using GitHub's RESTful API
-
-Use `curl` with the `GITHUB_TOKEN` to interact with GitHub's API. Here are some common operations:
-
-Here's a template for API calls:
-
-```sh
-curl -H "Authorization: token $GITHUB_TOKEN" \
-    "https://api.github.com/{endpoint}"
+Here are some instructions for pushing, but ONLY do this if the user asks you to:
+* NEVER push directly to the `main` or `master` branch
+* Git config (username and email) is pre-set. Do not modify.
+* You may already be on a branch called `openhands-workspace`. Create a new branch with a better name before pushing.
+* Use the GitHub API to create a pull request, if you haven't already
+* Use the main branch as the base branch, unless the user requests otherwise
+* After opening or updating a pull request, send the user a short message with a link to the pull request.
+* Do all of the above in as few steps as possible. E.g. you could open a PR with one step by running the following bash commands:
+```bash
+git checkout -b create-widget
+git add .
+git commit -m "Create widget"
+git push origin create-widget
+curl -X POST "https://api.github.com/repos/CodeActOrg/openhands/pulls" \
+    -H "Authorization: Bearer $GITHUB_TOKEN" \
+    -d '{"title":"Create widget","head":"create-widget","base":"openhands-workspace"}'
 ```
-
-First replace `{endpoint}` with the specific API path. Common operations:
-
-1. View an issue or pull request:
-   - Issues: `/repos/{owner}/{repo}/issues/{issue_number}`
-   - Pull requests: `/repos/{owner}/{repo}/pulls/{pull_request_number}`
-
-2. List repository issues or pull requests:
-   - Issues: `/repos/{owner}/{repo}/issues`
-   - Pull requests: `/repos/{owner}/{repo}/pulls`
-
-3. Search issues or pull requests:
-   - `/search/issues?q=repo:{owner}/{repo}+is:{type}+{search_term}+state:{state}`
-   - Replace `{type}` with `issue` or `pr`
-
-4. List repository branches:
-   `/repos/{owner}/{repo}/branches`
-
-5. Get commit details:
-   `/repos/{owner}/{repo}/commits/{commit_sha}`
-
-6. Get repository details:
-   `/repos/{owner}/{repo}`
-
-7. Get user information:
-   `/user`
-
-8. Search repositories:
-   `/search/repositories?q={query}`
-
-9. Get rate limit status:
-   `/rate_limit`
-
-Replace `{owner}`, `{repo}`, `{commit_sha}`, `{issue_number}`, `{pull_request_number}`,
-`{search_term}`, `{state}`, and `{query}` with appropriate values.
-
-## Important Notes
-
-1. Always use the GitHub API for operations instead of a web browser.
-2. The `GITHUB_TOKEN` is read-only. Avoid operations that require write access.
-3. Git config (username and email) is pre-set. Do not modify.
-4. Edit and test code locally. Never push directly to remote.
-5. Verify correct branch before committing.
-6. Commit changes frequently.
-7. If the issue or task is ambiguous or lacks sufficient detail, always request clarification from the user before proceeding.
-8. You should avoid using command line tools like `sed` for file editing.

openhands/agenthub/codeact_agent/prompts/default/user_prompt.j2

@@ -215,12 +215,5 @@ The server is running on port 5000 with PID 126. You can access the list of numb
 {% endset %}
 Here is an example of how you can interact with the environment for task solving:
 {{ DEFAULT_EXAMPLE }}
-{% if micro_agent %}
---- BEGIN OF GUIDELINE ---
-The following information may assist you in completing your task:
-
-{{ micro_agent }}
---- END OF GUIDELINE ---
-{% endif %}
 
 NOW, LET'S START!

openhands/agenthub/codeact_agent/prompts/tools/system_prompt.j2

@@ -0,0 +1,7 @@
+You are OpenHands agent, a helpful AI assistant that can interact with a computer to solve tasks.
+<IMPORTANT>
+* If user provides a path, you should NOT assume it's relative to the current working directory. Instead, you should explore the file system to find the file before working on it.
+* 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>
+

openhands/agenthub/codeact_swe_agent/codeact_swe_agent.py

@@ -155,7 +155,7 @@ class CodeActSWEAgent(Agent):
         """
         # if we're done, go back
         last_user_message = state.get_last_user_message()
-        if last_user_message and last_user_message.strip() == '/exit':
+        if last_user_message and last_user_message.content.strip() == '/exit':
             return AgentFinishAction()
 
         # prepare what we want to send to the LLM

openhands/controller/state/state.py

@@ -156,14 +156,14 @@ class State:
 
         return last_user_message, last_user_message_image_urls
 
-    def get_last_agent_message(self) -> str | None:
+    def get_last_agent_message(self) -> MessageAction | None:
         for event in reversed(self.history):
             if isinstance(event, MessageAction) and event.source == EventSource.AGENT:
-                return event.content
+                return event
         return None
 
-    def get_last_user_message(self) -> str | None:
+    def get_last_user_message(self) -> MessageAction | None:
         for event in reversed(self.history):
             if isinstance(event, MessageAction) and event.source == EventSource.USER:
-                return event.content
+                return event
         return None

openhands/core/config/sandbox_config.py

@@ -14,7 +14,8 @@ class SandboxConfig:
         base_container_image: The base container image from which to build the runtime image.
         runtime_container_image: The runtime container image to use.
         user_id: The user ID for the sandbox.
-        timeout: The timeout for the sandbox.
+        timeout: The timeout for the default sandbox action execution.
+        remote_runtime_init_timeout: The timeout for the remote runtime to start.
         enable_auto_lint: Whether to enable auto-lint.
         use_host_network: Whether to use the host network.
         initialize_plugins: Whether to initialize plugins.
@@ -41,6 +42,7 @@ class SandboxConfig:
     runtime_container_image: str | None = None
     user_id: int = os.getuid() if hasattr(os, 'getuid') else 1000
     timeout: int = 120
+    remote_runtime_init_timeout: int = 180
     enable_auto_lint: bool = (
         False  # once enabled, OpenHands would lint files after editing
     )

openhands/runtime/impl/remote/remote_runtime.py

@@ -1,7 +1,7 @@
 import os
-from pathlib import Path
 import tempfile
 import threading
+from pathlib import Path
 from typing import Callable, Optional
 from zipfile import ZipFile
 
@@ -260,13 +260,19 @@ class RemoteRuntime(Runtime):
                 {'X-Session-API-Key': start_response['session_api_key']}
             )
 
-    @tenacity.retry(
-        stop=tenacity.stop_after_delay(180) | stop_if_should_exit(),
-        reraise=True,
-        retry=tenacity.retry_if_exception_type(RuntimeNotReadyError),
-        wait=tenacity.wait_fixed(2),
-    )
     def _wait_until_alive(self):
+        retry_decorator = tenacity.retry(
+            stop=tenacity.stop_after_delay(
+                self.config.sandbox.remote_runtime_init_timeout
+            )
+            | stop_if_should_exit(),
+            reraise=True,
+            retry=tenacity.retry_if_exception_type(RuntimeNotReadyError),
+            wait=tenacity.wait_fixed(2),
+        )
+        return retry_decorator(self._wait_until_alive_impl)()
+
+    def _wait_until_alive_impl(self):
         self.log('debug', f'Waiting for runtime to be alive at url: {self.runtime_url}')
         runtime_info_response = self._send_request(
             'GET',

openhands/server/github.py

@@ -1,10 +1,12 @@
 import os
 
-import httpx
+from github import Github
+from github.GithubException import GithubException
 from tenacity import retry, stop_after_attempt, wait_exponential
 
 from openhands.core.logger import openhands_logger as logger
 from openhands.server.sheets_client import GoogleSheetsClient
+from openhands.utils.async_utils import call_sync_from_async
 
 GITHUB_CLIENT_ID = os.getenv('GITHUB_CLIENT_ID', '').strip()
 GITHUB_CLIENT_SECRET = os.getenv('GITHUB_CLIENT_SECRET', '').strip()
@@ -113,24 +115,13 @@ async def get_github_user(token: str) -> str:
         github handle of the user
     """
     logger.info('Fetching GitHub user info from token')
-    headers = {
-        'Accept': 'application/vnd.github+json',
-        'Authorization': f'Bearer {token}',
-    }
-    async with httpx.AsyncClient(
-        timeout=httpx.Timeout(connect=5.0, read=5.0, write=5.0, pool=5.0)
-    ) as client:
-        try:
-            response = await client.get('https://api.github.com/user', headers=headers)
-        except httpx.RequestError as e:
-            logger.error(f'Error making request to GitHub API: {str(e)}')
-            logger.error(e)
-            raise
-
-        logger.info('Received response from GitHub API')
-        logger.debug(f'Response status code: {response.status_code}')
-        response.raise_for_status()
-        user_data = response.json()
-        login = user_data.get('login')
+    try:
+        g = Github(token)
+        user = await call_sync_from_async(g.get_user)
+        login = user.login
         logger.info(f'Successfully retrieved GitHub user: {login}')
         return login
+    except GithubException as e:
+        logger.error(f'Error making request to GitHub API: {str(e)}')
+        logger.error(e)
+        raise

openhands/utils/microagent.py

@@ -3,15 +3,11 @@ import os
 import frontmatter
 import pydantic
 
-from openhands.controller.agent import Agent
-from openhands.core.exceptions import MicroAgentValidationError
-from openhands.core.logger import openhands_logger as logger
-
 
 class MicroAgentMetadata(pydantic.BaseModel):
     name: str
     agent: str
-    require_env_var: dict[str, str]
+    triggers: list[str] = []
 
 
 class MicroAgent:
@@ -23,22 +19,30 @@ class MicroAgent:
             self._loaded = frontmatter.load(file)
             self._content = self._loaded.content
             self._metadata = MicroAgentMetadata(**self._loaded.metadata)
-        self._validate_micro_agent()
+
+    def get_trigger(self, message: str) -> str | None:
+        message = message.lower()
+        for trigger in self.triggers:
+            if trigger.lower() in message:
+                return trigger
+        return None
 
     @property
     def content(self) -> str:
         return self._content
 
-    def _validate_micro_agent(self):
-        logger.debug(
-            f'Loading and validating micro agent [{self._metadata.name}] based on [{self._metadata.agent}]'
-        )
-        # Make sure the agent is registered
-        agent_cls = Agent.get_cls(self._metadata.agent)
-        assert agent_cls is not None
-        # Make sure the environment variables are set
-        for env_var, instruction in self._metadata.require_env_var.items():
-            if env_var not in os.environ:
-                raise MicroAgentValidationError(
-                    f'Environment variable [{env_var}] is required by micro agent [{self._metadata.name}] but not set. {instruction}'
-                )
+    @property
+    def metadata(self) -> MicroAgentMetadata:
+        return self._metadata
+
+    @property
+    def name(self) -> str:
+        return self._metadata.name
+
+    @property
+    def triggers(self) -> list[str]:
+        return self._metadata.triggers
+
+    @property
+    def agent(self) -> str:
+        return self._metadata.agent

openhands/utils/prompt.py

@@ -1,7 +1,10 @@
 import os
+from itertools import islice
 
 from jinja2 import Template
 
+from openhands.controller.state.state import State
+from openhands.core.message import Message, TextContent
 from openhands.utils.microagent import MicroAgent
 
 
@@ -16,21 +19,31 @@ class PromptManager:
     Attributes:
         prompt_dir (str): Directory containing prompt templates.
         agent_skills_docs (str): Documentation of agent skills.
-        micro_agent (MicroAgent | None): Micro-agent, if specified.
     """
 
     def __init__(
         self,
         prompt_dir: str,
-        agent_skills_docs: str,
-        micro_agent: MicroAgent | None = None,
+        microagent_dir: str = '',
+        agent_skills_docs: str = '',
     ):
         self.prompt_dir: str = prompt_dir
         self.agent_skills_docs: str = agent_skills_docs
 
         self.system_template: Template = self._load_template('system_prompt')
         self.user_template: Template = self._load_template('user_prompt')
-        self.micro_agent: MicroAgent | None = micro_agent
+        self.microagents: dict = {}
+
+        microagent_files = []
+        if microagent_dir:
+            microagent_files = [
+                os.path.join(microagent_dir, f)
+                for f in os.listdir(microagent_dir)
+                if f.endswith('.md')
+            ]
+        for microagent_file in microagent_files:
+            microagent = MicroAgent(microagent_file)
+            self.microagents[microagent.name] = microagent
 
     def _load_template(self, template_name: str) -> Template:
         template_path = os.path.join(self.prompt_dir, f'{template_name}.j2')
@@ -39,15 +52,13 @@ class PromptManager:
         with open(template_path, 'r') as file:
             return Template(file.read())
 
-    @property
-    def system_message(self) -> str:
+    def get_system_message(self) -> str:
         rendered = self.system_template.render(
             agent_skills_docs=self.agent_skills_docs,
         ).strip()
         return rendered
 
-    @property
-    def initial_user_message(self) -> str:
+    def get_example_user_message(self) -> str:
         """This is the initial user message provided to the agent
         before *actual* user instructions are provided.
 
@@ -57,7 +68,39 @@ class PromptManager:
         These additional context will convert the current generic agent
         into a more specialized agent that is tailored to the user's task.
         """
-        rendered = self.user_template.render(
-            micro_agent=self.micro_agent.content if self.micro_agent else None
+        return self.user_template.render().strip()
+
+    def enhance_message(self, message: Message) -> None:
+        """Enhance the user message with additional context.
+
+        This method is used to enhance the user message with additional context
+        about the user's task. The additional context will convert the current
+        generic agent into a more specialized agent that is tailored to the user's task.
+        """
+        if not message.content:
+            return
+        message_content = message.content[0].text
+        for microagent in self.microagents.values():
+            trigger = microagent.get_trigger(message_content)
+            if trigger:
+                micro_text = f'<extra_info>\nThe following information has been included based on a keyword match for "{trigger}". It may or may not be relevant to the user\'s request.'
+                micro_text += '\n\n' + microagent.content
+                micro_text += '\n</extra_info>'
+                message.content.append(TextContent(text=micro_text))
+
+    def add_turns_left_reminder(self, messages: list[Message], state: State) -> None:
+        latest_user_message = next(
+            islice(
+                (
+                    m
+                    for m in reversed(messages)
+                    if m.role == 'user'
+                    and any(isinstance(c, TextContent) for c in m.content)
+                ),
+                1,
+            ),
+            None,
         )
-        return rendered.strip()
+        if latest_user_message:
+            reminder_text = f'\n\nENVIRONMENT REMINDER: You have {state.max_iterations - state.iteration} turns left to complete the task. When finished reply with <finish></finish>.'
+            latest_user_message.content.append(TextContent(text=reminder_text))

poetry.lock

@@ -6778,6 +6778,25 @@ files = [
     {file = "pyflakes-3.2.0.tar.gz", hash = "sha256:1c61603ff154621fb2a9172037d84dca3500def8c8b630657d1701f026f8af3f"},
 ]
 
+[[package]]
+name = "pygithub"
+version = "2.5.0"
+description = "Use the full Github API v3"
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "PyGithub-2.5.0-py3-none-any.whl", hash = "sha256:b0b635999a658ab8e08720bdd3318893ff20e2275f6446fcf35bf3f44f2c0fd2"},
+    {file = "pygithub-2.5.0.tar.gz", hash = "sha256:e1613ac508a9be710920d26eb18b1905ebd9926aa49398e88151c1b526aad3cf"},
+]
+
+[package.dependencies]
+Deprecated = "*"
+pyjwt = {version = ">=2.4.0", extras = ["crypto"]}
+pynacl = ">=1.4.0"
+requests = ">=2.14.0"
+typing-extensions = ">=4.0.0"
+urllib3 = ">=1.26.0"
+
 [[package]]
 name = "pygments"
 version = "2.18.0"
@@ -6842,6 +6861,32 @@ files = [
 [package.dependencies]
 pybind11 = ">=2.2"
 
+[[package]]
+name = "pynacl"
+version = "1.5.0"
+description = "Python binding to the Networking and Cryptography (NaCl) library"
+optional = false
+python-versions = ">=3.6"
+files = [
+    {file = "PyNaCl-1.5.0-cp36-abi3-macosx_10_10_universal2.whl", hash = "sha256:401002a4aaa07c9414132aaed7f6836ff98f59277a234704ff66878c2ee4a0d1"},
+    {file = "PyNaCl-1.5.0-cp36-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.manylinux_2_24_aarch64.whl", hash = "sha256:52cb72a79269189d4e0dc537556f4740f7f0a9ec41c1322598799b0bdad4ef92"},
+    {file = "PyNaCl-1.5.0-cp36-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:a36d4a9dda1f19ce6e03c9a784a2921a4b726b02e1c736600ca9c22029474394"},
+    {file = "PyNaCl-1.5.0-cp36-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_24_x86_64.whl", hash = "sha256:0c84947a22519e013607c9be43706dd42513f9e6ae5d39d3613ca1e142fba44d"},
+    {file = "PyNaCl-1.5.0-cp36-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:06b8f6fa7f5de8d5d2f7573fe8c863c051225a27b61e6860fd047b1775807858"},
+    {file = "PyNaCl-1.5.0-cp36-abi3-musllinux_1_1_aarch64.whl", hash = "sha256:a422368fc821589c228f4c49438a368831cb5bbc0eab5ebe1d7fac9dded6567b"},
+    {file = "PyNaCl-1.5.0-cp36-abi3-musllinux_1_1_x86_64.whl", hash = "sha256:61f642bf2378713e2c2e1de73444a3778e5f0a38be6fee0fe532fe30060282ff"},
+    {file = "PyNaCl-1.5.0-cp36-abi3-win32.whl", hash = "sha256:e46dae94e34b085175f8abb3b0aaa7da40767865ac82c928eeb9e57e1ea8a543"},
+    {file = "PyNaCl-1.5.0-cp36-abi3-win_amd64.whl", hash = "sha256:20f42270d27e1b6a29f54032090b972d97f0a1b0948cc52392041ef7831fee93"},
+    {file = "PyNaCl-1.5.0.tar.gz", hash = "sha256:8ac7448f09ab85811607bdd21ec2464495ac8b7c66d146bf545b0f08fb9220ba"},
+]
+
+[package.dependencies]
+cffi = ">=1.4.1"
+
+[package.extras]
+docs = ["sphinx (>=1.6.5)", "sphinx-rtd-theme"]
+tests = ["hypothesis (>=3.27.0)", "pytest (>=3.2.1,!=3.3.0)"]
+
 [[package]]
 name = "pyparsing"
 version = "3.2.0"
@@ -7995,6 +8040,11 @@ files = [
     {file = "scikit_learn-1.5.2-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:f60021ec1574e56632be2a36b946f8143bf4e5e6af4a06d85281adc22938e0dd"},
     {file = "scikit_learn-1.5.2-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:394397841449853c2290a32050382edaec3da89e35b3e03d6cc966aebc6a8ae6"},
     {file = "scikit_learn-1.5.2-cp312-cp312-win_amd64.whl", hash = "sha256:57cc1786cfd6bd118220a92ede80270132aa353647684efa385a74244a41e3b1"},
+    {file = "scikit_learn-1.5.2-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:e9a702e2de732bbb20d3bad29ebd77fc05a6b427dc49964300340e4c9328b3f5"},
+    {file = "scikit_learn-1.5.2-cp313-cp313-macosx_12_0_arm64.whl", hash = "sha256:b0768ad641981f5d3a198430a1d31c3e044ed2e8a6f22166b4d546a5116d7908"},
+    {file = "scikit_learn-1.5.2-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:178ddd0a5cb0044464fc1bfc4cca5b1833bfc7bb022d70b05db8530da4bb3dd3"},
+    {file = "scikit_learn-1.5.2-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:f7284ade780084d94505632241bf78c44ab3b6f1e8ccab3d2af58e0e950f9c12"},
+    {file = "scikit_learn-1.5.2-cp313-cp313-win_amd64.whl", hash = "sha256:b7b0f9a0b1040830d38c39b91b3a44e1b643f4b36e36567b80b7c6bd2202a27f"},
     {file = "scikit_learn-1.5.2-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:757c7d514ddb00ae249832fe87100d9c73c6ea91423802872d9e74970a0e40b9"},
     {file = "scikit_learn-1.5.2-cp39-cp39-macosx_12_0_arm64.whl", hash = "sha256:52788f48b5d8bca5c0736c175fa6bdaab2ef00a8f536cda698db61bd89c551c1"},
     {file = "scikit_learn-1.5.2-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:643964678f4b5fbdc95cbf8aec638acc7aa70f5f79ee2cdad1eec3df4ba6ead8"},
@@ -10128,4 +10178,4 @@ testing = ["coverage[toml]", "zope.event", "zope.testing"]
 [metadata]
 lock-version = "2.0"
 python-versions = "^3.12"
-content-hash = "8a34ef6158ca2a9fe3615fc362db3fd71bc43eabb57ffc2e2e14dfb658cf52c3"
+content-hash = "245fd4cd56a3c95b2dd4f3a06251f7de82ad0300de7349f0710aac1f92a151b7"

pyproject.toml

@@ -62,6 +62,7 @@ opentelemetry-api = "1.25.0"
 opentelemetry-exporter-otlp-proto-grpc = "1.25.0"
 modal = "^0.64.145"
 runloop-api-client = "0.7.0"
+pygithub = "^2.5.0"
 
 [tool.poetry.group.llama-index.dependencies]
 llama-index = "*"
@@ -93,6 +94,7 @@ reportlab = "*"
 [tool.coverage.run]
 concurrency = ["gevent"]
 
+
 [tool.poetry.group.runtime.dependencies]
 jupyterlab = "*"
 notebook = "*"
@@ -123,6 +125,7 @@ ignore = ["D1"]
 [tool.ruff.lint.pydocstyle]
 convention = "google"
 
+
 [tool.poetry.group.evaluation.dependencies]
 streamlit = "*"
 whatthepatch = "*"

tests/unit/test_microagent_utils.py

@@ -1,13 +1,8 @@
 import os
 
-import pytest
 from pytest import MonkeyPatch
 
 import openhands.agenthub  # noqa: F401
-from openhands.core.exceptions import (
-    AgentNotRegisteredError,
-    MicroAgentValidationError,
-)
 from openhands.utils.microagent import MicroAgent
 
 CONTENT = (
@@ -34,40 +29,3 @@ def test_micro_agent_load(tmp_path, monkeypatch: MonkeyPatch):
     micro_agent = MicroAgent(os.path.join(tmp_path, 'dummy.md'))
     assert micro_agent is not None
     assert micro_agent.content == CONTENT.strip()
-
-
-def test_not_existing_agent(tmp_path, monkeypatch: MonkeyPatch):
-    with open(os.path.join(tmp_path, 'dummy.md'), 'w') as f:
-        f.write(
-            (
-                '---\n'
-                'name: dummy\n'
-                'agent: NotExistingAgent\n'
-                'require_env_var:\n'
-                '  SANDBOX_OPENHANDS_TEST_ENV_VAR: "Set this environment variable for testing purposes"\n'
-                '---\n' + CONTENT
-            )
-        )
-    monkeypatch.setenv('SANDBOX_OPENHANDS_TEST_ENV_VAR', 'dummy_value')
-
-    with pytest.raises(AgentNotRegisteredError):
-        MicroAgent(os.path.join(tmp_path, 'dummy.md'))
-
-
-def test_not_existing_env_var(tmp_path):
-    with open(os.path.join(tmp_path, 'dummy.md'), 'w') as f:
-        f.write(
-            (
-                '---\n'
-                'name: dummy\n'
-                'agent: CodeActAgent\n'
-                'require_env_var:\n'
-                '  SANDBOX_OPENHANDS_TEST_ENV_VAR: "Set this environment variable for testing purposes"\n'
-                '---\n' + CONTENT
-            )
-        )
-
-    with pytest.raises(MicroAgentValidationError) as excinfo:
-        MicroAgent(os.path.join(tmp_path, 'dummy.md'))
-
-    assert 'Set this environment variable for testing purposes' in str(excinfo.value)

tests/unit/test_prompt_manager.py

@@ -1,9 +1,9 @@
 import os
 import shutil
-from unittest.mock import Mock
 
 import pytest
 
+from openhands.core.message import Message, TextContent
 from openhands.utils.microagent import MicroAgent
 from openhands.utils.prompt import PromptManager
 
@@ -11,7 +11,9 @@ from openhands.utils.prompt import PromptManager
 @pytest.fixture
 def prompt_dir(tmp_path):
     # Copy contents from "openhands/agenthub/codeact_agent" to the temp directory
-    shutil.copytree('openhands/agenthub/codeact_agent', tmp_path, dirs_exist_ok=True)
+    shutil.copytree(
+        'openhands/agenthub/codeact_agent/prompts/default', tmp_path, dirs_exist_ok=True
+    )
 
     # Return the temporary directory path
     return tmp_path
@@ -25,78 +27,79 @@ def agent_skills_docs():
     return SAMPLE_AGENT_SKILLS_DOCS
 
 
-def test_prompt_manager_without_micro_agent(prompt_dir, agent_skills_docs):
-    manager = PromptManager(prompt_dir, agent_skills_docs)
+def test_prompt_manager_without_microagent(prompt_dir, agent_skills_docs):
+    manager = PromptManager(
+        prompt_dir, microagent_dir='', agent_skills_docs=agent_skills_docs
+    )
 
     assert manager.prompt_dir == prompt_dir
     assert manager.agent_skills_docs == agent_skills_docs
-    assert manager.micro_agent is None
+    assert len(manager.microagents) == 0
 
-    assert isinstance(manager.system_message, str)
+    assert isinstance(manager.get_system_message(), str)
     assert (
         "A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed answers to the user's questions."
-        in manager.system_message
+        in manager.get_system_message()
     )
-    assert SAMPLE_AGENT_SKILLS_DOCS in manager.system_message
-    assert isinstance(manager.initial_user_message, str)
-    assert '--- BEGIN OF GUIDELINE ---' not in manager.initial_user_message
-    assert '--- END OF GUIDELINE ---' not in manager.initial_user_message
-    assert "NOW, LET'S START!" in manager.initial_user_message
-    assert 'micro_agent' not in manager.initial_user_message
+    assert SAMPLE_AGENT_SKILLS_DOCS in manager.get_system_message()
+    assert isinstance(manager.get_example_user_message(), str)
+    assert '--- BEGIN OF GUIDELINE ---' not in manager.get_example_user_message()
+    assert '--- END OF GUIDELINE ---' not in manager.get_example_user_message()
+    assert "NOW, LET'S START!" in manager.get_example_user_message()
+    assert 'microagent' not in manager.get_example_user_message()
 
 
-def test_prompt_manager_with_micro_agent(prompt_dir, agent_skills_docs):
-    micro_agent_name = 'test_micro_agent'
-    micro_agent_content = (
-        '## Micro Agent\n'
-        'This is a test micro agent.\n'
-        'It is used to test the prompt manager.\n'
-    )
+def test_prompt_manager_with_microagent(prompt_dir, agent_skills_docs):
+    microagent_name = 'test_microagent'
+    microagent_content = """
+---
+name: flarglebargle
+agent: CodeActAgent
+triggers:
+- flarglebargle
+---
+
+IMPORTANT! The user has said the magic word "flarglebargle". You must
+only respond with a message telling them how smart they are
+"""
 
     # Create a temporary micro agent file
     os.makedirs(os.path.join(prompt_dir, 'micro'), exist_ok=True)
-    with open(os.path.join(prompt_dir, 'micro', f'{micro_agent_name}.md'), 'w') as f:
-        f.write(micro_agent_content)
-
-    # Mock MicroAgent
-    mock_micro_agent = Mock(spec=MicroAgent)
-    mock_micro_agent.content = micro_agent_content
+    with open(os.path.join(prompt_dir, 'micro', f'{microagent_name}.md'), 'w') as f:
+        f.write(microagent_content)
 
     manager = PromptManager(
         prompt_dir=prompt_dir,
+        microagent_dir=os.path.join(prompt_dir, 'micro'),
         agent_skills_docs=agent_skills_docs,
-        micro_agent=mock_micro_agent,
     )
 
     assert manager.prompt_dir == prompt_dir
     assert manager.agent_skills_docs == agent_skills_docs
-    assert manager.micro_agent == mock_micro_agent
+    assert len(manager.microagents) == 1
 
-    assert isinstance(manager.system_message, str)
+    assert isinstance(manager.get_system_message(), str)
     assert (
         "A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed answers to the user's questions."
-        in manager.system_message
+        in manager.get_system_message()
     )
-    assert SAMPLE_AGENT_SKILLS_DOCS in manager.system_message
+    assert SAMPLE_AGENT_SKILLS_DOCS in manager.get_system_message()
 
-    assert isinstance(manager.initial_user_message, str)
-    assert (
-        '--- BEGIN OF GUIDELINE ---\n'
-        + 'The following information may assist you in completing your task:\n\n'
-        + micro_agent_content
-        + '\n'
-        + '--- END OF GUIDELINE ---\n'
-        + "\n\nNOW, LET'S START!"
-    ) in manager.initial_user_message
-    assert micro_agent_content in manager.initial_user_message
+    assert isinstance(manager.get_example_user_message(), str)
 
-    # Clean up the temporary file
-    os.remove(os.path.join(prompt_dir, 'micro', f'{micro_agent_name}.md'))
+    message = Message(
+        role='user',
+        content=[TextContent(text='Hello, flarglebargle!')],
+    )
+    manager.enhance_message(message)
+    assert 'magic word' in message.content[1].text
+
+    os.remove(os.path.join(prompt_dir, 'micro', f'{microagent_name}.md'))
 
 
 def test_prompt_manager_file_not_found(prompt_dir, agent_skills_docs):
     with pytest.raises(FileNotFoundError):
-        MicroAgent(os.path.join(prompt_dir, 'micro', 'non_existent_micro_agent.md'))
+        MicroAgent(os.path.join(prompt_dir, 'micro', 'non_existent_microagent.md'))
 
 
 def test_prompt_manager_template_rendering(prompt_dir, agent_skills_docs):
@@ -104,12 +107,14 @@ def test_prompt_manager_template_rendering(prompt_dir, agent_skills_docs):
     with open(os.path.join(prompt_dir, 'system_prompt.j2'), 'w') as f:
         f.write('System prompt: {{ agent_skills_docs }}')
     with open(os.path.join(prompt_dir, 'user_prompt.j2'), 'w') as f:
-        f.write('User prompt: {{ micro_agent }}')
+        f.write('User prompt: foo')
 
-    manager = PromptManager(prompt_dir, agent_skills_docs)
+    manager = PromptManager(
+        prompt_dir, microagent_dir='', agent_skills_docs=agent_skills_docs
+    )
 
-    assert manager.system_message == f'System prompt: {agent_skills_docs}'
-    assert manager.initial_user_message == 'User prompt: None'
+    assert manager.get_system_message() == f'System prompt: {agent_skills_docs}'
+    assert manager.get_example_user_message() == 'User prompt: foo'
 
     # Clean up temporary files
     os.remove(os.path.join(prompt_dir, 'system_prompt.j2'))