contributor-pr
Skill99 starsChanged 8 days agoMITView on GitHub
Review, merge and release an external contributor's PR in web-ui, then refresh the contributors surfaces (README ×17 + cvstart.org). Trigger when a fork PR appears, the user says "проверь pr и замерж", "merge the contributor PR", or asks to update the contributors list.
In this repository7 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- swarmSkill
- hermes-bridgeSkill
- parent-syncSkill
- swarmSkill
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
- Very specific: line 24
Check the contributor's tests honor CI isolation (dynamic imports in before(), CAREER_OPS_ROOT mktemp — the paths-once rule).
- Very specific: line 31
Run the full suite on EACH pr-branch: npm test, capture the exit code directly (never pipe to grep).
- Very specific: line 33
Before merging, re-check gh pr view <n> --json headRefOid still equals the SHA you reviewed.
- Distinctive: line 49
Add a drift-gate test pinning any literal the PR changed across every surface it lives on (server fallback ↔ client dropdown ↔ dicts ×17 ↔ …
≈ “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.
Files
contributor-pr — external PR intake pipeline
Proven on v1.125.2 (PRs #144/#145 by @Alien10140 — Gemini defaults + the headless deep-prompt 502 fix). External code gets a stricter path than our own branches: fork CI does not auto-run, review claims are verified empirically, and merges need explicit user authorization.
Phase 1 — Review (local runs ARE the CI gate)
- Fork PRs from first-time contributors show "no checks reported" —
GitHub Actions waits for maintainer approval. Don't chase it; the local
full-suite run below substitutes, and push-CI on main re-validates after
merge. (Approving runs via
gh api …/actions/runs/<id>/approveandgh pr mergeare both classifier-blocked for external PRs — when blocked, STOP and hand the user the exact commands; never work around.) git fetch origin pull/<n>/head:pr-<n>and read the FULL diff. Extra scrutiny on LLM surfaces (prompts.mjs,routes/llm.mjs), sanitizers, any new ingress, and i18n snapshot regeneration (dup keys / glosses). Check the contributor's tests honor CI isolation (dynamic imports inbefore(),CAREER_OPS_ROOTmktemp — the paths-once rule).- Verify factual claims against live sources, never training data.
v1.125.2 lesson: an AI review called
gemini-3.6-flash"fabricated" as a BLOCKER; fetching https://ai.google.dev/gemini-api/docs/models proved the contributor right (3.6-flash Stable, 2.0-flash shut down). Model ids, API shapes, deprecations → WebFetch the official docs. - Run the full suite on EACH pr-branch:
npm test, capture the exit code directly (never pipe to grep). Record per-branch counts. - Before merging, re-check
gh pr view <n> --json headRefOidstill equals the SHA you reviewed.
Phase 2 — Merge
- Needs the user's explicit go for THESE PRs (standing "merge all branches" does not auto-extend to external code; the classifier enforces this).
gh pr merge <n> --squash --delete-branch, oldest first; thengit checkout mainas its OWN command, thengit pull(a checkout inside a blocked compound command silently never runs; v1.125.2: a README commit landed on the local pr-144 branch that way and had to be cherry-picked to main).- Full suite on merged main.
Phase 3 — Release rollup (patch bump, one release for the pack)
- Add a drift-gate test pinning any literal the PR changed across every
surface it lives on (server fallback ↔ client dropdown ↔ dicts ×17 ↔
help ×17) — text-extraction, not imports, to dodge paths-once. Regex
traps:
match()returns the FIRSThintFallback:-style hit in the file — slice from the owning field first. - Sweep stale literals the PR missed (help ×17 carried
gemini-2.0-flashtwice per bundle). - Then the standard ship: CHANGELOG ×17 crediting the @handle, banner
four touch-points ×17 (see parent-sync Phase 3), QA prompt, counts in
CONVENTIONS/PROJECT-CONTEXT/CLAUDE.md,
npm version, PR → CI → merge → tag → Release →gh workflow run publish-package.yml --ref vX.Y.Z→ wiki → site changelog sync → local redeploy.
Phase 4 — Contributors surfaces
README ×17 — hand-maintained borderless row of circular avatars (contrib.rocks was dropped: its ~24h server cache lags new people), in the site's contribution order, bots excluded, then a bold localized "All contributors →" link to the graph. Template (id via
gh api users/<login> --jq .id;&output=pngis MANDATORY — JPEG avatars have no alpha and the circle corners render white without it):<a href="https://github.com/<login>" title="<login>"><img src="https://wsrv.nl/?url=avatars.githubusercontent.com/u/<id>%3Fv%3D4&w=160&h=160&fit=cover&mask=circle&output=png" width="80" height="80" alt="<login>"/></a>- cvstart.org — the landing block bakes GitHub's
/contributorsintofacts.jsonat Pages build. GitHub recomputes that list asynchronously (hours after merge): if a fresh deploy still misses the person, re-dispatchdeploy-pages.ymllater or let the weekly cron catch it — nothing to fix in code. - A one-line localized credit also goes in the CHANGELOG entry and the README banner lead (Phase 3), not as a standalone line under the block.
Traps
- A sweep regex that removes "the old avatar line" will also eat a fresh
table/row cell containing the same
<login>.pngsubstring — scope deletions to the exact line shape, then re-verify cell count ×17. - Country flags go on language LISTS only; a broad
[label](FILE.md)flag-adder will pollute code-literal links in historical changelog entries (438 sites cleaned in v1.125.2). - contrib.rocks / GitHub stats caches cannot be busted from our side — don't burn cache-key variants trying.
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.

