Skip to content

[Klaud Cold] Signoff verifier: Check 3 recipe link is N/A for disagg/multi-node PRs#2030

Merged
functionstackx merged 1 commit into
mainfrom
klaud/signoff-check3-disagg-na
Jul 4, 2026
Merged

[Klaud Cold] Signoff verifier: Check 3 recipe link is N/A for disagg/multi-node PRs#2030
functionstackx merged 1 commit into
mainfrom
klaud/signoff-check3-disagg-na

Conversation

@functionstackx

Copy link
Copy Markdown
Collaborator

Summary

  • Scopes the sign-off verifier's Check 3 (recipe linked + complete) to single-node recipes only: the official upstream recipe sources (vLLM recipes, SGLang cookbook) publish single-node serve commands, so disaggregated / multi-node submissions (dynamo-trt/dynamo-sglang, ATOM/ATOMesh, sglang-disagg, vLLM disagg, benchmarks/multi_node/** incl. srt-slurm-recipes/**, master-config entries with multinode: true/disagg: true) now report ➖ N/A instead of ❌ FAIL.
  • A sign-off note like "this is a disagg submission, no recipe update required" is treated as a legitimate statement of that fact.
  • Mixed PRs (single-node + multi-node changes) still get the recipe-link requirement applied to the single-node portion.
  • Fixes false rejections like Add DSV4 FP4 B200 Dynamo-vLLM point-specific disagg recipes #1963 (comment).

Test plan

  • YAML validated locally.
  • After merge: throwaway disagg-only test PR + sign-off with no recipe link → expect Check 3 = N/A (not FAIL).

🤖 Generated with Claude Code

Upstream recipe sources (vLLM recipes, SGLang cookbook) publish
single-node serve commands, so the recipe-link requirement only applies
to single-node recipes. Disagg/multi-node PRs (dynamo, ATOM/ATOMesh,
sglang/vllm disagg, srt-slurm-recipes) now report N/A instead of FAIL,
fixing false rejections like PR #1963.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
@functionstackx functionstackx requested a review from a team July 4, 2026 02:51
@functionstackx functionstackx merged commit 32047a3 into main Jul 4, 2026
3 checks passed
@functionstackx functionstackx deleted the klaud/signoff-check3-disagg-na branch July 4, 2026 02:51

@claude claude Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM — prompt-only tweak to scope Check 3 to single-node recipes; no workflow logic or permission changes.

Extended reasoning...

Overview

This PR modifies only the Claude prompt text inside .github/workflows/codeowner-signoff-verify.yml. Specifically, it updates the description of Check 3 (recipe linked + complete) to explicitly scope the requirement to single-node recipes and report disaggregated / multi-node submissions as N/A rather than FAIL. It also updates the header comment block accordingly.

Security risks

None. The change touches only natural-language instructions given to Claude within the workflow; it does not alter permissions, secrets handling, the trusted-branch checkout policy, the MCP config, the concurrency group, or any of the surrounding shell/JavaScript logic. The security-critical property of this workflow — that it never checks out untrusted PR head code — is preserved.

Level of scrutiny

Low. This is a prompt-clarification change to a workflow that gates merges, but the modification narrows a false-positive failure mode (disagg/multi-node PRs getting FAIL on Check 3 despite having no upstream recipe source) and does not weaken any of the eight substantive gates. The listed disagg indicators (multinode: true, disagg: true, dynamo-trt/dynamo-sglang, ATOM/ATOMesh, sglang-disagg, benchmarks/multi_node/**, srt-slurm-recipes/**) match established framework/config conventions in the repo.

Other factors

Bug hunting system found no bugs. Mixed PRs still get single-node recipe enforcement per the added explicit instruction. The change is small, self-contained, and cleanly reversible.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

Development

Successfully merging this pull request may close these issues.

1 participant