fix(ci): repair dev after the 2.65.0 round merges

[lidge-jun]

Summary

Fixes the three failures on the dev Cross-platform CI run 36028905040 (tip 9cf3a06889) left by this round's merges:

• File-size ratchet (test 2/4, windows 5/9, macos 2/2): fix(usage): split the attempt row on account rotation, for every pool #5752 grew src/server/request-log.ts from 1,996 to 2,025 lines, over the 2,000-line limit for files without a baseline row (NEW_OVERSIZED). The account-rotation logic from fix(usage): split the attempt row on account rotation, for every pool #5752 (attemptAccountChanged and sealRequestAttemptIdentity) now lives in src/server/request-log-account-rotation.ts with its full explanation; request-log.ts re-exports sealRequestAttemptIdentity, so callers are unchanged. request-log.ts is 1,983 lines. Behavior is identical.
• Provider icon gate (gates): feat(provider): add a Claude Code CLI subscription provider #5712's claude-cli provider had a committed claude-color.svg but no PROVIDER_ICON_ALIASES row; added.
• DeepSeek terminal repair (windows 4/9): tests/providers/deepseek-inbound-wire.test.ts still asserted that a custom-named provider at the official DeepSeek endpoint gets no terminal repair. fix(providers): fallback to transport destination for registry model wire defaults #5697 intentionally made an alias at an exact canonical destination take that registry row's policy (the same contract pinned for OpenCode Go), so the assertion now expects { graceMs: 5000 }.

Verification

• bun run typecheck: pass. bun run structure:check: pass.
• bun test tests/providers/deepseek-inbound-wire.test.ts tests/ci-workflows/file-size-ratchet.test.ts tests/usage/key-attribution.test.ts tests/usage/request-log-attempt-identity.test.ts: 63 pass, 0 fail.
• cd gui && bun test tests/provider-icons.test.ts: 6 pass, 0 fail.
• Full suite left to the dev Cross-platform CI run dispatched after merge.

Checklist

• Scope stays focused and avoids unrelated cleanup.
• Docs or release notes were updated when needed. (No user-facing change.)
• Security-sensitive changes were reviewed for secrets, auth, and unsafe defaults. (None touched.)

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

Configuration used: Repository: lidge-jun/opencodex/.coderabbit.yaml

Review profile: ASSERTIVE

Plan: Advanced

Run ID: 27260098-f811-45b5-a194-344abd0b0d80

📥 Commits

Reviewing files that changed from the base of the PR and between 9cf3a06 and 14e228c.

📒 Files selected for processing (4)

• gui/src/provider-icons.ts
• src/server/request-log-account-rotation.ts
• src/server/request-log.ts
• tests/providers/deepseek-inbound-wire.test.ts

 _____________________________________________________________________________________
< All idioms must be learned. Good idioms only need to be learned once. - Alan Cooper >
 -------------------------------------------------------------------------------------
  \
   \   \
        \ /\
        ( )
      .( o ).

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

🧪 Generate unit tests (beta)

• Commit to this branch
• Create a new PR

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

✅ Deterministic PR hygiene checks passed.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-24T16:54:14.243854Z	14e228c	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.