feat: Toolset Context Managers

[monoxgas]

Added automatic context manager support for Toolset classes with re-entrancy protection. Agents now automatically enter/exit tool context managers before execution using AsyncExitStack, preventing duplicate setup/teardown even when manually managed by users.

class AsyncCMToolSet(Toolset):
    """
    Scenario 1: A standard, async-native Toolset.
    Tests that __aenter__/__aexit__ are called correctly and only once.
    """

    enter_count: int = 0
    exit_count: int = 0

    async def __aenter__(self) -> "AsyncCMToolSet":
        event_log.append("async_tool_enter_start")
        await asyncio.sleep(0.01)  # Simulate async work
        self.enter_count += 1
        event_log.append("async_tool_enter_end")
        return self

    async def __aexit__(self, *args: object) -> None:
        event_log.append("async_tool_exit_start")
        await asyncio.sleep(0.01)
        self.exit_count += 1
        event_log.append("async_tool_exit_end")

    @tool_method
    async def do_work(self) -> str:
        """A sample method for the agent to call."""
        event_log.append("async_tool_method_called")
        return "async work done"

---

Generated Summary:

• Added support for async context management in the Agent and Toolset classes, enhancing concurrency capabilities.
• Implemented automatic re-entrancy handling for context managers in Toolset, ensuring safe entry/exit behavior.
• Updated API documentation to replace deprecated "slug" field with a new "identifier" field for organization models.
• Introduced comprehensive test suite for agent lifecycle management and toolset behavior, verifying context management and idempotency.
• Minor documentation corrections in model descriptions for clarity.

This summary was generated with ❤️ by rigging

[Copilot]

Pull Request Overview

This PR adds automatic context manager support for Toolset classes with re-entrancy protection. Agents now automatically enter/exit tool context managers before execution using AsyncExitStack, preventing duplicate setup/teardown even when manually managed by users.

• Implemented __init_subclass__ hook in Toolset to wrap __aenter__/__aexit__ (and sync variants) with re-entrancy protection
• Modified Agent.stream() to automatically enter context managers for all tools before execution
• Fixed spelling error in Organization model documentation ("identifer" → "identifier")

Reviewed Changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/test_agent_lifecycle.py	Comprehensive test suite covering async/sync context managers, re-entrancy, and return value handling
dreadnode/api/models.py	Fixed typo in Organization.key field documentation
dreadnode/agent/tools/base.py	Implemented re-entrancy wrapper logic in Toolset.init_subclass using reference counting
dreadnode/agent/agent.py	Added AsyncExitStack to automatically enter tool context managers in stream() method
docs/sdk/task.mdx	Fixed logic for output_object_hash initialization and conditional checks
docs/sdk/api.mdx	Updated Organization documentation to rename "slug" to "identifier"
docs/sdk/agent_tools.mdx	Added documentation for new async context management support

Comments suppressed due to low confidence (1)

docs/sdk/api.mdx:1794

• The documentation refers to this field as "identifier" but the actual field name in the Organization model (line 504 in dreadnode/api/models.py) is "key". The documentation should match the actual field name:

### key

```python
key: str

URL-friendly identifier for the organization.

Unique identifier for the organization.

### is\_active

```python

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.