Skip to content

pr-review: a reviewer skip on a draft PR fails the run (exit 1) after two model calls #2545

Open

Description

@don-petry

What happens

When pr-review's single-confirmation reviewer answers skip with reason draft, the log says the skip is "treated as a no-op", and then the job fails with exit code 1. A draft PR that clears every gate therefore turns each pr-review run red without posting anything.

Evidence

Three runs on 2026-11-14, all for #2319 ("Potential fix for code scanning alert no. 40", a draft PR), on pr-review/v1-next at 8a4e373:

Each ends the "Review each PR (cascade)" step like this:

[maintainer-gate] no undispositioned PR issue comments
[maintainer-review-gate] no unaddressed maintainer review threads
    [tier1] escalate=false risk=MEDIUM signals=[]
    [approve] triage cleared — running single confirmation (claude-opus-5-5)
    reviewer returned skip (draft) — treating as no-op
    [done]  https://github.com/combsmith/.github-private/pull/2319
##[error]Process completed with exit code 1.

The message comes from scripts/post-pr-review.sh (the DECISION = skip branch, which exits 100). Where the 100 becomes a failing exit 1 has not been traced.

This is not new in v1.57.0: none of post-pr-review.sh, review-one-pr.sh or the batch runner changed between pr-review/v1-stable (8917aa7) and 8a4e373. #2319 only reached the reviewer today because its earlier gates cleared.

Why it matters

  • Each run spends a triage call and a confirmation call on a PR that cannot be approved.
  • A red pr-review run counts as a failed run in canary ring evidence, so a draft PR can hold a rollout for a reason unrelated to the candidate.

Expected

  • A draft PR is skipped before any model call, with a skip decision line and reason draft.
  • A reviewer skip ends the run green.

Acceptance criteria

  • A draft PR is skipped by a gate before tier 1 triage, with a decision line naming draft.
  • A reviewer verdict of skip exits the batch successfully; a test covers it.
  • No pr-review run fails solely because the PR is a draft.

Activity

  1. github-actions commented on Nov 14, 2026

    @github-actions
    Contributor

    Thanks for the detailed report, and for including the run links and the exact log excerpt. That made the problem easy to follow. The evidence already covers expected versus actual behaviour. The one thing still missing is where the reviewer's skip (exit 100) gets turned into exit 1. Do you know whether review-one-pr.sh or the batch runner treats any non-zero code as a failure, or have you only seen it in the Actions logs so far?

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

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugBug reports

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions