Summary

• What changed: Mistral requests now get a provider-aware outbound copy of chat history with reasoning_content removed from messages before the request is sent.
• Why it changed: strict Mistral-compatible endpoints reject that non-standard message field after a user switches from a thinking/reasoning provider, but the shared in-memory history still needs to preserve it for providers such as DeepSeek.
• Reviewer focus: confirm this is scoped to Mistral detection (official base URL plus model-name fallback) and that the original history object is not mutated.

Validation

• Commands run:

  npm ci --ignore-scripts
  cd packages/core && npx vitest run src/core/openaiContentGenerator/provider/mistral.test.ts
  cd packages/core && npx vitest run src/core/openaiContentGenerator/provider/mistral.test.ts src/core/openaiContentGenerator/provider/deepseek.test.ts src/core/openaiContentGenerator/provider/minimax.test.ts
  cd packages/core && npx vitest run src/core/openaiContentGenerator/provider
  npm run build
  npm run typecheck --workspace=packages/core
  npx prettier --check packages/core/src/core/openaiContentGenerator/index.ts packages/core/src/core/openaiContentGenerator/provider/index.ts packages/core/src/core/openaiContentGenerator/provider/mistral.ts packages/core/src/core/openaiContentGenerator/provider/mistral.test.ts
  npx eslint packages/core/src/core/openaiContentGenerator/index.ts packages/core/src/core/openaiContentGenerator/provider/index.ts packages/core/src/core/openaiContentGenerator/provider/mistral.ts packages/core/src/core/openaiContentGenerator/provider/mistral.test.ts
  git diff --check

• Prompts / inputs used: provider-unit regression with prior assistant history containing reasoning_content, then an outbound Mistral request.
• Expected result: Mistral-bound request messages omit reasoning_content; non-Mistral providers keep it; the source history still contains it.
• Observed result: matched expected; provider test directory passes (116 tests).
• Quickest reviewer verification path: cd packages/core && npx vitest run src/core/openaiContentGenerator/provider/mistral.test.ts
• Evidence (output, logs, screenshots, video, JSON, before/after, etc.): local Vitest output showed src/core/openaiContentGenerator/provider passing with 116/116 tests.

Scope / Risk

• Main risk or tradeoff: model-name fallback may catch custom Mistral-family aliases behind strict OpenAI-compatible proxies, which is intended for this compatibility quirk.
• Not covered / not validated: live Mistral API smoke test with real credentials was not run.
• Breaking changes / migration notes: none; this only changes the outbound request copy for Mistral-compatible providers.

Testing Matrix

Testing matrix notes:

• Verified locally on Linux only; no sandbox/container surface was changed.

Linked Issues / Bugs

Fixes #3304