Add a built-in tool that lets the agent inspect and modify its own
runtime state (model, iterations, context window, etc.).
Key features:
- inspect: view current config, usage stats, and subagent status
- modify: adjust parameters at runtime (protected by type/range validation)
- Subagent observability: inspect running subagent tasks (phase,
iteration, tool events, errors) — subagents are no longer a black box
- Watchdog corrects out-of-bounds values on each iteration
- Enabled by default in read-only mode (self_modify: false)
- All changes are in-memory only; restart restores defaults
- Comprehensive test suite (90 tests)
Includes a self-awareness skill (always-on) with progressive disclosure:
SKILL.md for core rules, references/examples.md for detailed scenarios.
- Added Jinja2 template support for various agent responses, including identity, skills, and memory consolidation.
- Introduced new templates for evaluating notifications, handling subagent announcements, and managing platform policies.
- Updated the agent context and memory modules to utilize the new templating system for improved readability and maintainability.
- Added a new dependency on Jinja2 in pyproject.toml.
Introduce a CompositeHook that fans out lifecycle callbacks to an
ordered list of AgentHook instances with per-hook error isolation.
Extract the nested _LoopHook and _SubagentHook to module scope as
public LoopHook / SubagentHook so downstream users can subclass or
compose them. Add `hooks` parameter to AgentLoop.__init__ for
registering custom hooks at construction time.
Closes#2603
- Rename WebSearchConfig references to the new WebToolsConfig root struct that wraps both search config and global proxy settings
- Add 'enable' flag to WebToolsConfig to allow fully disabling all web-related tools (WebSearch, WebFetch) at runtime
- Update AgentLoop and SubagentManager to receive the full web config object instead of separate web_search_config/web_proxy parameters
- Update CLI command initialization to pass the consolidated web config struct instead of split fields
- Change default web search provider from brave to duckduckgo for better out-of-the-box usability (no API key required)
Keep multimodal tool outputs on the native content-block path while
restoring redirect SSRF checks for web_fetch image responses. Also share
image block construction, simplify persisted history sanitization, and
add regression tests for image reads and blocked private redirects.
Made-with: Cursor
Instead of adding a separate load_skill tool to bypass workspace restrictions,
extend ReadFileTool with extra_allowed_dirs so it can read builtin skill paths
while keeping write/edit tools locked to the workspace. Fixes the original issue
for both main agent and subagents.
Made-with: Cursor
When restrictToWorkspace is enabled, the agent cannot read builtin skill
files via read_file since they live outside the workspace. This adds a
dedicated load_skill tool that reads skills by name through the SkillsLoader,
which accesses files directly via Python without the workspace restriction.
- Add LoadSkillTool to filesystem tools
- Register it in the agent loop
- Update system prompt to instruct agent to use load_skill instead of read_file
- Remove raw filesystem paths from skills summary
Share assistant message construction between the main agent and subagents, and add a regression test to keep reasoning_content and thinking_blocks in follow-up tool rounds.
Subagent's _run_subagent() was dropping reasoning_content and
thinking_blocks when building assistant messages for the conversation
history. Providers like Deepseek Reasoner require reasoning_content on
every assistant message when thinking mode is active, causing a 400
BadRequestError on the second LLM round-trip.
Align with the main AgentLoop which already preserves these fields via
ContextBuilder.add_assistant_message().
Closes#1834
- Remove trailing whitespace and normalize blank lines
- Unify string quotes and line breaks for long lines
- Sort imports alphabetically across modules
- cancel_by_session: use asyncio.gather for parallel cancellation
instead of sequential await per task
- _dispatch: register in _active_tasks before acquiring lock so /stop
can find queued tasks (synced from #1179)
- SubagentManager tracks _session_tasks: session_key -> {task_id, ...}
- cancel_by_session() cancels all subagents for a session
- SpawnTool passes session_key through to SubagentManager
- /stop response reports subagent cancellation count
- Cleanup callback removes from both _running_tasks and _session_tasks
Builds on #1179
xzq.xuandXubin Ren
Mariano CampoandXubin Ren
chengyongruandXubin Ren
aiguozhi123456andXubin Ren
Xubin Ren
Jack Lu
zhangxiaoyu.yorkandXubin Ren
sontianyeandXubin Ren
Shiniese
vandazia
kinchahoy
BenandXubin Ren
WhalerO
lailoo
JK_Lu
coldxiangyu
aiguozhi123456
Nikolas de Hor
chtangwin