agentleFS
WeaverseWeaverse/.agents/skills/pr-merge-gates/SKILL.md

pr-merge-gates

Skill0 starsChanged 6 days agoMITView on GitHub

Check GitHub signals that can falsely clear a Weaverse PR: skipped CI, remapped review comments, incomplete automated cycles and moving heads. Use when preparing, approving or merging a PR, not as a generic code-review checklist.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 3 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read on, branch/path filters and job conditions.

    line 14
  2. Very specific:

    Read inline threads, formal review bodies and PR discussion.

    line 26
  3. Distinctive:

    Wait for the repository's automated-review completion signal on this exact trigger and head.

    line 28
  4. Distinctive:

    If a finding disproves the PR description, correct the claim as well as the code.

    line 36
  5. Distinctive:

    Check supersession against the current base and serving history.

    line 43
  6. Very specific:

    Read workflow triggers before pushing: on: [push] can deploy feature branches.

    line 50
  7. Very specific:

    Use only supported gh pr view --json fields.

    line 58

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

PR merge gates

Apparent green versus executed checks

  • gh pr checks can exit 0 when material jobs were skipped. Require completed/success results for each material job on the current head, and explain any intentionally absent job. "No checks reported" is absence, not green.
  • Marking a draft ready runs CI only if the workflow subscribes to ready_for_review or another matching event. Read on, branch/path filters and job conditions. Do not wait for a run that the workflow cannot create.
  • gh pr checks --required can report no required checks when branch protection enforces none. Read all checks and the actual protection/ruleset separately; policy may require more than GitHub enforces.
  • Re-runs and matrices can share names. Bind conclusions to the head, workflow/run ID, attempt and material job, not the first matching name or an old green row. Auto-merge waits only for enforced checks.
  • A setup/download failure did not run tests. Report it as unexecuted, then rerun the same failed run only within authority. A test count in the PR body needs execution evidence behind it.
  • For exported API changes, run any checked-in API/declaration report gate before review. A typecheck alone may trust an ambient declare module that only the production bundler will disprove.

Review identity

  • Read inline threads, formal review bodies and PR discussion. An empty body or COMMENTED state is not a pass. The fix-pr helper filters GraphQL threads by isResolved=false and isOutdated=false; flat comments cannot.
  • Wait for the repository's automated-review completion signal on this exact trigger and head. Multiple findings can arrive minutes apart. A reaction is sufficient only where the repository explicitly uses it as a verdict; require the configured reviewer's identity and creation after the latest trigger. Quiet time is not completion.
  • Match the configured reviewer's account on each API surface; app and bot identities can differ. Do not hardcode a person's handle or accept any user's reaction as the automated verdict.
  • Bind a remapped finding to original_commit_id and creation time, not only commit_id. Classify it against current source before rewriting a working fix. Severity labels inform, but do not replace, impact assessment.
  • A clean automated review does not waive required human approval or reviewDecision: REVIEW_REQUIRED.
  • If a finding disproves the PR description, correct the claim as well as the code. A failed or partial review is not a review pass; findings from a previous SHA remain historical until rechecked.

Placebo changes and ancestry

  • Rebuild with and without a suspected configuration fix at the real command boundary. Framework CLIs can override Vite config inline; a changed source value may produce an identical deployment artifact.
  • Check supersession against the current base and serving history. Older PRs can restore older file content without conflicts. Inspect a disposable integration result; patch equivalence alone does not establish ancestry.
  • Resolve generated-artifact conflicts by regenerating from the combined sources, then run the generator's check. Do not hand-merge lockfiles, API reports or spec indices just to remove conflict markers.

Authorized publication and merge

  • Read workflow triggers before pushing: on: [push] can deploy feature branches. Inherited push configuration can publish extra refs or tags; use an explicit refspec with tag-following disabled:

    git -c push.default=nothing -c push.followTags=false push origin "$SHA:refs/heads/$BRANCH"
  • Pass PR prose with --body-file to avoid shell expansion of backticks and dollar signs.
  • Use only supported gh pr view --json fields; one invalid field aborts the whole read. The REST PR resource gives both exact repository identities when abbreviated CLI fields are insufficient.
  • Within merge authority, use gh pr merge --match-head-commit "$SHA", then read merged status and the resulting base commit back. If the base moved or the merge strategy changes the tree, recheck the integration result. Even on an unchanged base, require the merged tree to match the reviewed tree; do not assume it.
  • Attribute a newly red base gate by executing the same gate on candidate and base, not by guessing from history. Development-branch merge and production delivery are separate outcomes.

Report READY, BLOCKED or NOT REVIEWABLE, with head SHA, executed/absent checks, review-cycle evidence and live blockers bound to original commits. This playbook never grants publication or merge permission.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.