fix(agent): run response hooks before yielding llm_result in skills_like fallback (#9788)

[aniruddhaadak80]

What Problem This Solves

In the skills_like tool requery fallback path, the runner yields llm_result (sending the response to the user) before calling _complete_with_assistant_response, which triggers on_agent_done hooks. Plugins doing content safety review, sanitization, or rewriting cannot affect text that has already been sent.

Why This Change Was Made

When tool_schema_mode=skills_like and the model initially selects a tool but the schema requery returns a plain assistant response (no tool calls), the fallback branch at tool_loop_agent_runner.py:934 yields the response to the user before the response hooks run. The normal no-tool path already has the correct order (complete first, then yield); this fallback branch was missed.

User Impact

Response hooks (content safety, sanitization, rewriting, audit) now run before the response reaches the user in the skills_like fallback path, matching the behavior of the normal no-tool path.

Evidence

• Issue: [Bug] skills_like 无工具重查询退化分支在 response hook 前发送响应 #9788
• Root cause confirmed at tool_loop_agent_runner.py:934-962 — yields before _complete_with_assistant_response
• _complete_with_assistant_response calls on_agent_done hook at line 201
• Fix: moved await self._complete_with_assistant_response(llm_resp) before the yield statements
• 1 file changed, 4 insertions, 1 deletion

Summary by Sourcery

Bug Fixes:

• Run response-completion hooks before yielding assistant responses in the skills_like fallback path, allowing safety, sanitization, rewriting, and audit hooks to affect user-visible content.

[sourcery-ai]

Hey - I've reviewed your changes and they look great!

Sourcery assessment

Approved.

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}