fix: after_llm_call hooks no longer break native tool execution

[lucasgomide]

Registering any after_llm_call hook broke native tool execution: the executor runs _setup_after_llm_call_hooks on the intermediate response that carries the model's tool calls, the structured payload got stringified for the response-rewrite pass, and the executor then treated that string as the final answer instead of executing the tools. Structured payloads (neither str nor BaseModel) now pass through untouched, matching the guard the direct-call path already applies in _invoke_after_llm_call_hooks; hooks still fire on the follow-up textual response.

Fixes #6529

---

Note

Low Risk
Small, targeted guard in hook setup with a regression test; behavior change only affects the tool-call intermediate response when after hooks are registered.

Overview
Fixes a regression where any registered after_llm_call hook broke native tool execution: _setup_after_llm_call_hooks ran on the LLM’s intermediate tool-call payload, stringified it for the hook rewrite path, and the executor then treated that string as the final answer instead of running tools (#6529).

_setup_after_llm_call_hooks now returns structured responses unchanged when the answer is neither str nor BaseModel (e.g. native tool-call lists). Hooks still run on normal text and Pydantic responses. This mirrors the guard the direct llm.call() path already uses via _invoke_after_llm_call_hooks.

Adds a regression test ensuring tool-call payloads pass through untouched while hooks still observe textual responses.

^{Reviewed by Cursor Bugbot for commit 12356b4. Bugbot is set up for automated code reviews on this repo. Configure here.}

[coderabbitai]

📝 Walkthrough

Walkthrough

Changes

Tool-call hook preservation

Layer / File(s)	Summary
After-LLM response guard and regression coverage lib/crewai/src/crewai/utilities/agent_utils.py, lib/crewai/tests/hooks/test_llm_hooks.py	The after-LLM hook pipeline bypasses rewriting for non-string, non-BaseModel responses, while regression coverage verifies structured tool-call passthrough and textual hook execution.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	The changes address #6529 by letting structured tool-call payloads pass through unchanged so tools can execute normally.
Out of Scope Changes check	✅ Passed	The PR stays focused on the hook serialization fix and its regression test, with no obvious unrelated changes.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Title check	✅ Passed	The title clearly describes the main fix: after_llm_call hooks no longer interfere with native tool execution.
Description check	✅ Passed	The description matches the changeset and explains the hook/tool-call regression, the guard, and the regression test.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch luzk/fix-after-hook-tool-calls

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

🧹 Nitpick comments (1)

lib/crewai/src/crewai/utilities/agent_utils.py (1)

1754-1760: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Guard logic is correct; type annotations need widening.

The early-return guard correctly prevents stringification of structured tool-call payloads. However, the function signature at lines 1736 and 1739 still types answer as str | BaseModel and the return as str | BaseModel, even though the guard now allows arbitrary objects (e.g., list[Mock()] in the test) to pass through unchanged. Update the annotations to reflect reality.

♻️ Proposed type annotation update

 def _setup_after_llm_call_hooks(
     executor_context: CrewAgentExecutor | AgentExecutor | LiteAgent | None,
-    answer: str | BaseModel,
+    answer: str | BaseModel | Any,
     printer: Printer,
     verbose: bool = True,
-) -> str | BaseModel:
+) -> str | BaseModel | Any:

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@lib/crewai/src/crewai/utilities/agent_utils.py` around lines 1754 - 1760,
Widen the answer parameter and return annotations of the function containing the
new isinstance guard to include arbitrary object values, matching its
passthrough behavior for non-str/BaseModel responses. Keep the existing guard
and textual/BaseModel handling unchanged, and use the project’s established
broad object type annotation.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@lib/crewai/src/crewai/utilities/agent_utils.py`:
- Around line 1754-1760: Widen the answer parameter and return annotations of
the function containing the new isinstance guard to include arbitrary object
values, matching its passthrough behavior for non-str/BaseModel responses. Keep
the existing guard and textual/BaseModel handling unchanged, and use the
project’s established broad object type annotation.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: bf19ddef-bd25-401f-af1d-d582a0e8367e

📥 Commits

Reviewing files that changed from the base of the PR and between 9d72e26 and 3bdd042.

📒 Files selected for processing (2)

• lib/crewai/src/crewai/utilities/agent_utils.py
• lib/crewai/tests/hooks/test_llm_hooks.py