fix: preserve reasoning.text type in consolidateReasoningDetails for Gemini models

[roomote-v0]

Related GitHub Issue

Closes: #11629

Description

This PR attempts to address Issue #11629. Feedback and guidance are welcome.

Two bugs in consolidateReasoningDetails() in src/api/transform/openai-format.ts cause assistant content[] to be stripped on Turn 2 for google/gemini-3.1-pro-preview via OpenRouter:

Bug 1 (secondary, root cause): Shared type variable mislabels reasoning.text as reasoning.encrypted

When a group contains both a reasoning.text entry and a reasoning.encrypted entry at the same index, the consolidation loop used a single shared type variable. The last entry's type overwrote earlier ones, so the text entry was emitted with type: "reasoning.encrypted" instead of type: "reasoning.text".

Fix: Track textType separately from encrypted entry types. When emitting the consolidated text entry, use the type derived from text-bearing entries only, and correct any mislabeled reasoning.encrypted back to reasoning.text.

Bug 2 (primary, cascading): Corrupted block filter drops mislabeled text entries

The filter at line 53 dropped any entry with type === "reasoning.encrypted" && !detail.data. This caught the mislabeled text entries from Bug 1, which had a text field but no data field. With all reasoning_details dropped, sanitizeGeminiMessages() saw hasReasoningDetails=false and dropped all tool_calls, producing the empty content[] that triggers the "no assistant messages" error.

Fix: Only drop entries that have neither data nor text -- truly corrupted blocks with no usable content.

Test Procedure

• Added 3 new tests to consolidateReasoningDetails covering:
  • Mixed reasoning.text + reasoning.encrypted at same index preserves correct types
  • Mislabeled text entries (type=encrypted, has text, no data) are preserved, not dropped
  • Truly corrupted blocks (no data AND no text) are still dropped
• Added 1 new regression test to sanitizeGeminiMessages reproducing the exact issue scenario
• All 48 tests pass: cd src && pnpm vitest run api/transform/__tests__/openai-format.spec.ts

Pre-Submission Checklist

• Issue Linked: This PR is linked to an approved GitHub Issue (see "Related GitHub Issue" above).
• Scope: My changes are focused on the linked issue (one major feature/fix per PR).
• Self-Review: I have performed a thorough self-review of my code.
• Testing: New and/or updated tests have been added to cover my changes.
• Documentation Impact: No documentation updates are required.
• Contribution Guidelines: I have read and agree to the Contributor Guidelines.

Documentation Updates

• No documentation updates are required.

Additional Notes

Files modified:

• src/api/transform/openai-format.ts -- fix consolidateReasoningDetails() type tracking and corrupted block filter
• src/api/transform/__tests__/openai-format.spec.ts -- 4 new regression tests

Interactively review PR in Roo Code Cloud

[sewasti]

please review this fix to enable gemini 3.1 pro via openrouter

[jongleeat]

@roomote
This PR is now on "Review required". Please assign a reviewer to merge this pull request.

[roomote-v0]

Fixaroo   See task

Marked this PR as ready for review. All 48 tests pass locally (cd src && pnpm vitest run api/transform/__tests__/openai-format.spec.ts).

[jongleeat]

Hello @cte, I know you must be busy, but I would really appreciate it if you could take a look at this PR when you have some free time. I'm happy to address any feedback or provide more context. Thank you!