fix(provider): widen Kimi anthropic-SDK thinking match to K2.x by Astro-Han · Pull Request #739 · Astro-Han/pawwork

Astro-Han · 2026-05-18T13:22:09Z

Summary

Widen the transform.ts substring match that enables thinking on Kimi models served over the Anthropic SDK path, so Kimi K2.6 (k2p6) is covered the same way K2.5 already is. Cherry-pick from upstream anomalyco/opencode commit 58232d896 (PR #23696, 2026-04-21).

Why

Upstream widened the Anthropic-SDK Kimi match from K2.5-only to all k2.x ids a month ago. PawWork missed that sync, so on the Anthropic-SDK Kimi path (kimi-for-coding provider), K2.6 requests went out without the thinking block that Moonshot's coding endpoint expects on reasoning models. This is a latent protocol-level gap on a paid Kimi tier.

Note on scope: a user-reported "death loop on Kimi for Coding K2.6" arrived at the same time, and this PR was initially expected to fix it. End-to-end testing on a real Kimi for Coding plan account showed that the death-loop symptom is actually HTTP 429 + rate_limit_error from Moonshot side (a quota / rate-limit issue, not a protocol issue). Builds on this branch send the correct thinking block but still hit 429 when the account is rate-limited. So this PR fixes a real protocol gap but is not the fix for the rate-limit report; that needs separate work on retry behavior and error messaging.

Related Issue

No GitHub issue. The rate-limit UX follow-up will be tracked separately.

Human Review Status

Pending. A human should make the final merge decision after reviewing the final diff and verification evidence.

Review Focus

The widened substring matches mirror upstream 58232d896 exactly (k2p5 -> k2p, kimi-k2.5 -> kimi-k2.). The intent is to stay sync-friendly, not to micro-optimize the match shape.
The updated negative-case fixtures (kimi-k2-preview, kimi-k2, kimi-k2-thinking) are the only K2-family ids in our snapshot that should still be excluded.

Risk Notes

Behavior change only affects models on the Anthropic SDK code path. Today that is kimi-for-coding per models-snapshot.js. OpenAI-compatible Kimi paths (volcengine plan, moonshot.cn, moonshot.ai) are untouched.
kimi-k2-thinking returns reasoning content by design and intentionally stays out of this branch; covered by the negative test.

How To Verify

Targeted local checks from the worktree:

bun --cwd packages/opencode test test/provider/transform.test.ts:
  201 pass, 0 fail, 385 expect() calls
bun --cwd packages/opencode test test/provider/provider.test.ts:
  95 pass, 0 fail, 252 expect() calls
bun turbo typecheck:
  5/5 successful (opencode + 4 cached)

Manual desktop verification on a real Kimi for Coding K2.6 account confirmed the request body now includes thinking: { type: enabled, budget_tokens: 16000 } on this branch. Server-side rate-limit behavior is out of scope for this PR.

Screenshots or Recordings

Not applicable; provider-layer change with no UI surface.

Checklist

Cherry-pick from upstream anomalyco/opencode #23696 (commit 58232d896, 2026-04-21). Without this, Kimi for Coding K2.6 (k2p6) on the Anthropic SDK path never gets `thinking` enabled, the request is rejected by Moonshot's coding endpoint, and the user is stuck in a 10-attempt retry loop showing "The engine is currently overloaded, please try again later". Root cause: - `kimi-for-coding` uses `@ai-sdk/anthropic` but Kimi is not an Anthropic model. Moonshot's coding endpoint requires `thinking: enabled` on K2.x reasoning models or it returns a misleading overload-style error. - The existing transform only matched K2.5 ids (`k2p5`, `kimi-k2.5`, `kimi-k2p5`). K2.6 (`k2p6`) had no transform and went out unmodified. Change boundary: - transform.ts: substring match widened from `k2p5` -> `k2p` and from `kimi-k2.5` -> `kimi-k2.` in two places (variants() and options()). Now covers K2.5, K2.6, and future K2.x without further edits. - transform.test.ts: update the existing "Kimi anthropic thinking" block to assert K2.5 and K2.6 both get thinking enabled, while `kimi-k2-preview`, `kimi-k2`, and `kimi-k2-thinking` still don't. Out of scope: - This PR does not touch the OpenAI-compatible Kimi paths (volcengine plan, moonshot.cn, moonshot.ai). Those already have their own reasoning_content handling.

coderabbitai · 2026-05-18T13:22:18Z

Warning

Rate limit exceeded

@Astro-Han has exceeded the limit for the number of commits that can be reviewed per hour. Please wait 39 minutes and 40 seconds before requesting another review.

You’ve run out of usage credits. Purchase more in the billing tab.

⌛ How to resolve this issue?

After the wait time has elapsed, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

We recommend that you space out your commits to avoid hitting the rate limit.

🚦 How do rate limits work?

CodeRabbit enforces hourly rate limits for each developer per organization.

Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout.

Please see our FAQ for further information.

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 4f083c27-6013-4756-b3e1-a52f50716b6f

📥 Commits

Reviewing files that changed from the base of the PR and between 18b9429 and 0bd0af3.

📒 Files selected for processing (2)

packages/opencode/src/provider/transform.ts
packages/opencode/test/provider/transform.test.ts

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests
Commit unit tests in branch claude/kimi-anthropic-sdk-k2x

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

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

github-actions

Suggested priority: P2 (includes non-doc, non-test paths outside the low-risk bucket).

P1/P0 are reserved for maintainer confirmation. Please relabel manually if this is a release blocker, security issue, data-loss risk, or updater/runtime failure.

gemini-code-assist

Code Review

This pull request generalizes the matching logic for Kimi K2 models by replacing specific version strings like 'k2p5' with broader patterns such as 'k2p' and 'kimi-k2.', and updates the test suite to include additional K2.x variants. The reviewer recommends using explicit regular expressions instead of broad string inclusion checks to ensure more robust model identification and avoid unintended matches.

Astro-Han · 2026-05-18T14:25:08Z

Re Gemini review on transform.ts:686 and transform.ts:1161:

These two hunks are an intentional verbatim cherry-pick of upstream anomalyco/opencode commit 58232d896 (PR #23696, 2026-04-21). Keeping the same includes(...) substring style — not a regex — is the deliberate trade-off: PawWork is already 4269 commits behind upstream on this layer, and any local micro-optimization here just adds another conflict point on the next sync.

The false-positive surface Gemini is worried about is bounded by what models-snapshot.js actually exposes: the only k2p* ids today are k2p5 and k2p6 under kimi-for-coding, both of which we want matched. If a future model id legitimately contains k2p but should NOT get this branch, that's a model-registry fix at that time, not a precondition for this backport.

Not actioning the suggestion. Resolving the threads.

github-actions Bot added harness Model harness, prompts, tool descriptions, and session mechanics P2 Medium priority labels May 18, 2026

github-actions Bot reviewed May 18, 2026

View reviewed changes

Astro-Han added the bug Something isn't working label May 18, 2026

gemini-code-assist Bot reviewed May 18, 2026

View reviewed changes

Comment thread packages/opencode/src/provider/transform.ts

Comment thread packages/opencode/src/provider/transform.ts

Astro-Han merged commit d2cf1c8 into dev May 18, 2026
28 of 29 checks passed

Astro-Han deleted the claude/kimi-anthropic-sdk-k2x branch May 18, 2026 14:27

This was referenced May 18, 2026

Kimi for Coding rate-limit 429 surfaces as misleading 'engine overloaded' death loop #740

Closed

chore(release): bump desktop version to v2026.5.19 #742

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(provider): widen Kimi anthropic-SDK thinking match to K2.x#739

fix(provider): widen Kimi anthropic-SDK thinking match to K2.x#739
Astro-Han merged 1 commit into
devfrom
claude/kimi-anthropic-sdk-k2x

Astro-Han commented May 18, 2026 •

edited

Loading

Uh oh!

coderabbitai Bot commented May 18, 2026

Rate limit exceeded

Uh oh!

github-actions Bot left a comment

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

Uh oh!

Uh oh!

Astro-Han commented May 18, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

Astro-Han commented May 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Why

Related Issue

Human Review Status

Review Focus

Risk Notes

How To Verify

Screenshots or Recordings

Checklist

Uh oh!

coderabbitai Bot commented May 18, 2026

Rate limit exceeded

Uh oh!

github-actions Bot left a comment

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

Uh oh!

Astro-Han commented May 18, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Astro-Han commented May 18, 2026 •

edited

Loading