pr-comment-sweep
Skill1 starsChanged 4 months agoApache-2.0View on GitHub
Fetch GitHub pull request comments with gh, assess actionable feedback, then apply, validate, commit, push, and resolve scoped fixes without approval pauses. Use when asked to sweep, address, or fix PR comments end to end.
In this repository19 other files it gives its agents.
- AGENTS.mdAGENTS.md
- agent-aeoSkill
- agent-memorySkill
- beanquerySkill
- ci-repairbaySkill
- delegated-developmentSkill
- ghSkill
- issue-blueprintSkill
- issue-workbenchSkill
- minifluxSkill
- og-imageSkill
- opSkill
- pencil-to-figma-migrationSkill
- pr-launchpadSkill
- repo-surveyorSkill
- review-checkpointSkill
- review-repairbaySkill
- shipyardSkill
- skill-optimizerSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 9
Run standalone; never invoke, defer to, or modify Shipyard.
- Very specific: line 33
Follow Thread triage and ledger every item as actionable, non-actionable, or blocked, with evidence, smallest fix, and regression.
- Very specific: line 36
Verify feedback against code, callers, tests, and PR intent.
- Very specific: line 47
Commit accepted fixes with scoped Conventional Commit message(s).
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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.
Files
PR Comment Sweep
Address actionable feedback on current-branch PR or supplied PR number/URL. Run standalone; never invoke, defer to, or modify Shipyard.
Resolve
<skill>, set optional$PR, and require a clean tracked tree. On resume, dirt, or head mismatch, read Sweep recovery and stop unless its checkpoint or adoption rules pass. For a fresh sweep:SNAPSHOT=$(mktemp) if [ -n "${PR:-}" ]; then "<skill>/scripts/verify-pr-target.sh" "$PR" node "<skill>/scripts/pr-feedback.mjs" fetch --pr "$PR" --out "$SNAPSHOT" else "<skill>/scripts/verify-pr-target.sh" node "<skill>/scripts/pr-feedback.mjs" fetch --out "$SNAPSHOT" fi PR=$(jq -r '.pullRequest.url' "$SNAPSHOT")Verifier requires authenticated
gh, open PR, exact local/PR head, and one matching push remote. Inspect base, full diff, and raw snapshot. Maintain recovery reference's one-line checkpoint after each phase.- Initial fetch prints all conversation comments, review bodies, unresolved
current/outdated threads, replies, and status. Later fetches into a copied
snapshot print additions, edits, and state changes only; raw JSON remains complete.
Follow Thread triage and ledger every item as
actionable,non-actionable, orblocked, with evidence, smallest fix, and regression. Never reconstruct GraphQL pagination. - Verify feedback against code, callers, tests, and PR intent. For SDK/framework
claims, inspect installed types and runtime caller; missing public capability
is
blocked, never a private-API or cast workaround. Invocation authorizes scoped edits, commits, push, and addressed-thread resolution. Stop only for unclear behavior, conflicting feedback, material product choice, or scope expansion. - Inspect resulting diff and run smallest relevant non-destructive validation. Stage only inspected paths; stop on failed or unavailable required validation unless user accepts risk. Never change secrets or unrelated work. Never hand-edit generated files: change source, run canonical generator for required tracked artifacts, then inspect both. Never wait or poll checks.
- Commit accepted fixes with scoped Conventional Commit message(s). With no
fixes, do not commit or push. After clean-tree validation, run
node "<skill>/scripts/pr-feedback.mjs" push --snapshot "$SNAPSHOT"; helper pushes once and accepts only exact local-head confirmation. Copy baseline before final fetch:
FINAL_SNAPSHOT=$(mktemp) cp "$SNAPSHOT" "$FINAL_SNAPSHOT" node "<skill>/scripts/pr-feedback.mjs" fetch --pr "$PR" --out "$FINAL_SNAPSHOT"Assess delta, then resolve all addressed IDs in one command, repeating flag:
node "<skill>/scripts/pr-feedback.mjs" resolve --pr "$PR" --expected-head "$(git rev-parse HEAD)" --thread "$ID1" --thread "$ID2". Never resolve non-actionable or blocked threads. Re-fetch once intoFINAL_SNAPSHOT, assess late delta, batch any newly addressed IDs, then runnode "<skill>/scripts/pr-feedback.mjs" checks --pr "$PR"once. Read-only transient retries are allowed; mutation retries and polling are not. Do not reply unless explicitly requested.- Report
PR | actioned | resolved IDs | skipped IDs | pending IDs | checks | commits | push | blockers.
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
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.

