fix: append trailing user turn in native Gemini provider

[hvlcrs]

Summary

Fixes #6984.

Gemini's generateContent API rejects a request whose history ends on a model turn (400 'Requests ending with a model turn are not supported.'). CrewAI's own agent loop (handle_max_iterations_exceeded, task guardrail/guardrail_max_retries retries) can produce exactly that history. This covers both places a Gemini call can be dispatched from:

• Native provider: GeminiCompletion._format_messages_for_gemini maps assistant messages to Gemini's model role (in both the plain-text and tool-call branches) but never checked whether the resulting contents list ends on model. Now appends a trailing user types.Content when it does.
• LiteLLM fallback: reached when google-genai isn't installed, or the model name isn't recognized as native (custom deployments, new previews, fine-tunes). LLM._format_messages_for_provider already guards this exact case for Mistral and Ollama, but had no Gemini branch — and litellm's own Vertex/Gemini transformation doesn't guard it either (checked litellm/llms/vertex_ai/gemini/transformation.py). Added a Gemini branch mirroring the existing Mistral one.

Test plan

• uv run --package crewai pytest lib/crewai/tests/llms/google/test_google.py -v — 46 passed, 1 skipped, 1 pre-existing unrelated failure (test_gemini_raises_error_when_model_not_supported, fails identically on main without this change; requires the litellm extra which isn't installed for that test module).
• uv run --package crewai pytest lib/crewai/tests/test_llm.py — 66 passed, 3 skipped, no failures.
• Added test_gemini_message_formatting_appends_user_turn_after_trailing_model_turn and test_gemini_message_formatting_leaves_user_terminated_history_unchanged in tests/llms/google/test_google.py (native provider).
• Added test_gemini_litellm_appends_user_message_when_last_is_assistant and test_gemini_litellm_does_not_modify_when_last_is_user in tests/test_llm.py (LiteLLM fallback), mirroring the existing Ollama tests.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 40888763-4645-42e8-a868-cb2df6f11d1b

📥 Commits

Reviewing files that changed from the base of the PR and between 5765912 and 3a4a52a.

📒 Files selected for processing (2)

• lib/crewai/src/crewai/llms/providers/gemini/completion.py
• lib/crewai/tests/llms/google/test_google.py

🚧 Files skipped from review as they are similar to previous changes (1)

• lib/crewai/src/crewai/llms/providers/gemini/completion.py

---

📝 Walkthrough

Walkthrough

Gemini formatters now append a synthetic user continuation message after assistant or model-ending history. The native formatter raises a ValueError for unresolved function calls. Tests cover native Gemini and LiteLLM formatting.

Changes

Gemini message formatting

Layer / File(s)	Summary
Add trailing user continuation and function-call validation lib/crewai/src/crewai/llms/providers/gemini/completion.py, lib/crewai/src/crewai/llm.py	The formatters append a user message containing “Please continue.” after assistant or model-ending history. The native Gemini formatter raises a ValueError when the final model turn contains an unresolved function call.
Validate Gemini history termination lib/crewai/tests/llms/google/test_google.py, lib/crewai/tests/test_llm.py	Tests verify continuation messages, unchanged user-terminated histories, input preservation, and unresolved function call errors.

Suggested reviewers: lorenzejay

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	The changes satisfy issue #6972 by handling trailing model turns in native Gemini and LiteLLM paths with regression tests.
Out of Scope Changes check	✅ Passed	All code and test changes support Gemini trailing-turn handling or safe unresolved function-call behavior.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Title check	✅ Passed	The title clearly identifies the primary change: appending a trailing user turn for the native Gemini provider.
Description check	✅ Passed	The description directly explains the Gemini formatting fix, both provider paths, the error case, and the regression tests.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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]

Actionable comments posted: 2

🤖 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.

Inline comments:
In `@lib/crewai/src/crewai/llms/providers/gemini/completion.py`:
- Around line 675-684: Update the model-ending handling around the contents
construction to inspect the final model Content’s parts before appending the
synthetic “Please continue.” turn. If it contains an unresolved function_call,
route it through the existing function-response handling or raise a targeted
error instead; only append the continuation prompt for model turns without
pending calls. Add a regression test covering an assistant tool_calls message
with no subsequent tool response.

In `@lib/crewai/tests/llms/google/test_google.py`:
- Around line 517-519: Strengthen the assertions around
_format_messages_for_gemini by verifying the complete formatted message
sequence, including each expected role and exact text such as "Please continue."
Confirm that the original test_messages retain their original roles and "Hello"
content after formatting, covering both relevant assertion blocks without
relying only on implementation details.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: f7698c15-9527-4436-8012-722c9b983ebe

📥 Commits

Reviewing files that changed from the base of the PR and between 28d868c and a74770e.

📒 Files selected for processing (2)

• lib/crewai/src/crewai/llms/providers/gemini/completion.py
• lib/crewai/tests/llms/google/test_google.py