Conversation
…in gate 1 Gate 1 in commands/release.md stated, as settled fact, that "this repository is one instance of exactly that shape" (a reduced push matrix, full coverage reserved for a `workflow_dispatch` input, #1246) -- inside a document every managed repo release reads identically. Read against claude-supertool at commit c5e3044, whose push trigger already runs the full matrix and defines no such workflow_dispatch input, that assertion is simply false there, and a releaser following the gate literally waits forever on a dispatched run that will never appear (release_ci_wait.py's PENDING state, spun with --require-event workflow_dispatch and nothing to satisfy it). Gate 1 now frames the reduced/full split as something to derive per repo -- by reading that repo own .github/workflows/*.yml at release time -- rather than a fact asserted for one named repo, per CLAUDE.md own governing rule: a fact about one repository must never live in shared code. The post-push CI-wait section (#1266, #1324) carried the same unconditional assumption in its own worked example and additionally never stated what to do when a repo genuinely lacks this shape; it now says explicitly to skip the second dispatch entirely when the push trigger already runs full coverage, closing the gap that produced the eternal-PENDING failure mode. Added tests/test_release_gate1_repo_specific_1375.py: red before the fix (the hardcoded phrase matched), green after (removed, while the per-repo derivation instruction is still present). No existing test covered commands/*.md for repo-specific claims -- test_content_invariants.py's HARDCODED sweep only reaches skills/agents/phase files, not commands/. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QYm4rkixzFWu3MXm2r9hhG
…t itself Two findings from the Explore self-review spawn against de3fa2b: 1. The post-push section's new "skip the rest of this section entirely" instruction, on a repo whose push trigger already runs full coverage, scoped too broadly: "this section" also holds the `--require-event workflow_dispatch` wait's exit-code handling and the tag-creation step itself. Following the instruction literally either tags with no CI wait on the release commit at all (reproducing the v0.27.0 regression on exactly the repo shape this fix names as its own motivating example), or strands the tag-creation steps with nothing left to trigger them. Fixed: the skip is now scoped to only the dispatch step and its --require-event wait -- the paragraph above still applies unconditionally, and a repo without the wider shape now waits on the plain push-triggered run (`release_ci_wait.py --commit "$COMMIT_SHA" --wait`, no --require-event) before proceeding to the same tag-creation step. 2. tests/test_release_gate1_repo_specific_1375.py's second test asserted two substrings that already existed, verbatim, in the *other* paragraph (the post-push section) before this fix -- so it passed on the pre-fix file too and exercised nothing about the actual change. Fixed: the test now isolates gate 1's own paragraph (anchored on the phrase "does not satisfy this gate", asserted unique in the file) and checks the derivation instruction within it specifically. Verified this version is red against the pre-fix file (the "never something to assume here" phrasing is absent there) and green against the current file. Both reviewers classified with scripts/review_return.py: the Explore spawn as states-findings (2 findings, both handled above), oss:auditor as no-findings (0 findings; checked absence/guard/injection/platform classes, none applicable to a prose-only change). Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QYm4rkixzFWu3MXm2r9hhG
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #1375.
Gate 1 in commands/release.md stated, as settled fact, that this repository is one instance of exactly that shape -- describing claude-oss's own CI setup (a reduced push matrix, full coverage reserved for a workflow_dispatch input, #1246) as though every managed repo shared it, inside a document every managed repo's release reads identically.
Read against a different real repo (claude-supertool at commit c5e3044), whose push trigger already runs the full matrix and defines no such workflow_dispatch input, that assertion is simply false there. A releaser following the gate literally, plus the post-push CI-wait section's (#1266/#1324) own unconditional dispatch instruction, waits forever (PENDING) on a dispatched run that will never appear. This is exactly the class CLAUDE.md's own governing rule names: a fact about one repository must never live in shared code.
What changed
Self-review
Two rounds. Round 1 (Explore + oss:auditor against commit de3fa2b) found two real problems in that commit: the new skip-this-section wording was scoped too broadly (it could be read as skipping the CI wait and tag-creation step themselves, reproducing the v0.27.0 tag-before-CI regression on exactly the repo shape this fix names as its own example), and the second test asserted substrings that already existed, verbatim, in the file's unchanged part -- so it proved nothing about the actual diff. Both fixed in commit d9fd6fe. Since that fix touched 3 files, scripts/fix_commit_scope.py returned needs-second-pass, so a lighter round 2 ran over the fix commit alone (Explore + oss:auditor again): NO FINDINGS from both, verified by hand against both the pre-fix and post-fix file content.
Full reviewer exchange and by-hand verification transcripts are in the developer report's note file.
Pushback / judgment call
The issue named the post-push section as presumably needing the same treatment and left it to my judgment how much surrounding text to adjust. It needed more than a wording tweak: the original unconditional instruction had no defined behavior for a repo that genuinely lacks the wider workflow_dispatch shape (the exact claude-supertool case the issue raises), so I added a parallel plain-wait path rather than only softening the prose. This is a slightly larger change than delete one sentence but the round-1 self-review confirmed it was necessary -- the narrower fix left a real regression risk on the shape the issue is about.
🤖 Generated with Claude Code
https://claude.ai/code/session_01QYm4rkixzFWu3MXm2r9hhG
[AI-generated]