fix(opencode): Error 400 missing reasoning_content – transform reasoning according to DeepSeek API and aditional test cases

[dixoxib]

Issue for this PR

Closes #17523, #9397, #8934, #5577

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

1. Fixes DeepSeek API thinking mode handling by ensuring assistant messages after the last user message receive an empty reasoning_content field when thinking mode is enabled.
2. Only current turn's reasoning is provided for the request.

Problem: When using DeepSeek models with thinking mode (via reasoner model IDs or thinking option), assistant messages after the last user message were missing the reasoning_content field. The DeepSeek API expects this field to be present (even if empty) for proper thinking mode operation in multi-turn conversations. The official DeepSeek API only utilizes current turn reasoning, no need for previous reasoning to be provided.

Changes made:

1. Modified packages/opencode/src/provider/transform.ts to:

   • Detect when thinking mode is enabled for DeepSeek models (checking capabilities.reasoning, model ID containing "deepseek-reasoner", or thinking: { type: "enabled" } option)
   • Find the index of the last user message in the conversation
   • Set empty reasoning_content for all assistant messages after that index when thinking mode is active
   • Preserve existing behavior for actual reasoning content (keeps reasoning text when present)

2. Added tests in packages/opencode/test/provider/transform.test.ts covering:

   • DeepSeek with empty reasoning_content for assistant messages after last user message
   • DeepSeek with reasoner model ID
   • DeepSeek with thinking option enabled
   • DeepSeek without thinking mode (should NOT set empty reasoning_content)
   • Non-DeepSeek providers unaffected

Why it works: The DeepSeek API documentation indicates that thinking mode requires the reasoning_content field to be present on assistant messages. By setting it to an empty string for assistant messages after the last user message, we comply with API expectations while maintaining backward compatibility for non-thinking mode usage.

How did you verify your code works?

1. Ran all existing tests: bun test passes
2. Added and ran new specific tests for DeepSeek thinking mode scenarios
3. Manual verification with DeepSeek API would show proper thinking mode behavior in multi-turn conversations
4. Verified non-DeepSeek providers are unaffected by checking test cases for other providers
5. live test on the official API

Screenshots / recordings

N/A - This is a backend fix with no UI changes.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

If you do not follow this template your PR will be automatically rejected.

[github-actions]

Hey! Your PR title transform to DeepSeek API fix and aditional test cases doesn't follow conventional commit format.

Please update it to start with one of:

• feat: or feat(scope): new feature
• fix: or fix(scope): bug fix
• docs: or docs(scope): documentation changes
• chore: or chore(scope): maintenance tasks
• refactor: or refactor(scope): code refactoring
• test: or test(scope): adding or updating tests

Where scope is the package name (e.g., app, desktop, opencode).

See CONTRIBUTING.md for details.

[github-actions]

Thanks for updating your PR! It now meets our contributing guidelines. 👍