Skip to content

fix(provider): tighten deepseek-v4 effort policy - #1383

Closed
TheGreatAxios wants to merge 1 commit into
mainfrom
cl-10311-deepseek-v4-effort-policy
Closed

TheGreatAxios wants to merge 1 commit into
mainfrom
cl-10311-deepseek-v4-effort-policy

Conversation

@TheGreatAxios

Copy link
Copy Markdown
Collaborator

Summary

  • Coder spawns at max V4 reasoning effort, matching the other orchestrators, instead of the cheapest low/none rung (CL-10311).
  • Off-ladder V4 reasoning efforts (anything but none/xhigh/max) arriving on a manual, exec, or config path are rejected with a clear error rather than coerced or passed through to the wire (CL-10312).
  • The DeepSeek family predicate matches the whole deepseek line (deepseek, deepseek-v4, deepseek-chat, deepseek-coder, deepseek-v3, and org-qualified ids), so the adapter router, effort routing, and model-family key treat pre-V4 and V4 DeepSeek consistently (CL-10313).

Verification

  • bun run check passes: lint, typecheck, check:dead-exports, and build all green; guarded tests 8571 pass / 1 fail.
  • The single test failure is the known pre-existing src/shell/literal-path-arguments.test.ts "allows: cat .ENV" case, which also fails on clean main and is unrelated to this change.

Fixes CL-10311
Fixes CL-10312
Fixes CL-10313

@linear-code

linear-code Bot commented Oct 9, 2026 •

Copy link
Copy Markdown

CL-10312

CL-10313

CL-10311

@TheGreatAxios
TheGreatAxios deleted the cl-10311-deepseek-v4-effort-policy branch October 9, 2026 15:59
@TheGreatAxios

Copy link
Copy Markdown
Collaborator Author

Superseded by #1384 (head branch renamed to the Linear gitBranchName).

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