Content
88%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A tight, executable maintainer-triage playbook: real `gh` commands, explicit scoring rubrics, and hard-fail validation gates that prevent premature merge recommendations. Structure is clean and self-contained; the only gap is that nothing is offloaded to reference files, so progressive disclosure is good rather than exemplary.
Suggestions
Tighten the prose in Phase 2/3 bullets (e.g. the 'Stale-but-high-severity...' and contributor-audit narrative sentences) into signal-only clauses to lift conciseness toward a 5.
Consider extracting the per-PR scoring rubric (severity/mergeability/scope/locale/tests/contributor definitions) into a references/scoring-rubric.md so SKILL.md stays an overview and progressive disclosure reaches the 5 anchor.
Add an explicit validation checkpoint in Phase 1 (e.g. confirm `gh` auth/scope before the batch scan, or abort with a clear message) so the batch gather step has a verifiable pre-condition like the PR-scoring phases do.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly lean and actionable with no padding about what gh, PRs, or scraping are, but a few prose clauses ('Stale-but-high-severity ranks higher than fresh-but-cosmetic', the contributor-audit narrative) could be tightened further. Not a 5 because a handful of explanatory sentences don't fully earn their tokens. | 4 / 5 |
Actionability | Copy-paste-ready `gh` commands with exact `--json` field lists, a runnable `grep -E` locale-safety audit, numeric 1–5 scoring scales with per-level definitions, and concrete output table templates — fully executable with no pseudocode. | 5 / 5 |
Workflow Clarity | A clearly sequenced 4-phase pipeline (Gather → Score issues → Score PRs → Rank) with explicit validation checkpoints: locale/DOM-selector hard fails, test-coverage requirement, and a 'Never recommend merge as-is until checks pass' gate before the final ranking. Evidence-citation is enforced as a non-negotiable feedback loop. | 5 / 5 |
Progressive Disclosure | Well-organized into labelled sections (Inputs, Phase 1–4, Non-negotiables) with cross-links to sibling skills (/2-repro-issue, /3-verify-pr-fix) at one level deep, and no nested reference chains. Not a 5 because all guidance is inline in a single ~90-line file rather than split across reference files, and the under-50-lines simple-skill exception does not apply. | 4 / 5 |
Total | 18 / 20 Passed |