Skip to content

fix(provider): replay reasoning_content on v4 deepseek turns - #1369

Merged
TheGreatAxios merged 1 commit into
mainfrom
cl-10223-reasoning-content-replay
Oct 9, 2026
Merged

TheGreatAxios merged 1 commit into
mainfrom
cl-10223-reasoning-content-replay

Conversation

@TheGreatAxios

Copy link
Copy Markdown
Collaborator

Summary

  • Decouples DeepSeek testing from reasoning_content stripping: pre-V4 DeepSeek models strip reasoning_content from replayed input messages, while V4 keeps each turn's reasoning on replay.
  • Routes V4 models past the early return so their request settings always reach the wire.
  • Adds byte-identical guards proving non-V4/non-DeepSeek models and pre-V4 DeepSeek ids are unchanged.

Verification

  • bun run check passes on this branch (8451 pass / 1 pre-existing environmental fail, matching the approved combined-tree baseline for the unrelated shell-guard cwd test).
  • src/provider/openai-compatible-adapter.test.ts passes (52 pass / 0 fail).

Fixes CL-10223

@linear-code

linear-code Bot commented Oct 9, 2026

Copy link
Copy Markdown

CL-10223

@TheGreatAxios
TheGreatAxios force-pushed the cl-10223-reasoning-content-replay branch 2 times, most recently from 33e1eea to 4cab808 Compare October 9, 2026 04:09
@TheGreatAxios
TheGreatAxios force-pushed the cl-10223-reasoning-content-replay branch from 4cab808 to a5836e3 Compare October 9, 2026 04:40
@TheGreatAxios
TheGreatAxios merged commit 822108e into main Oct 9, 2026
13 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant