Content
100%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.
An exemplary workflow skill body: lean imperative prose, exact commands and templates, explicit validation gates and recovery loops, and a deliberate one-level-deep split with references/rules.md. Nothing explains what Claude already knows, and every rule encodes non-obvious repository-specific discipline.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Every line is repo-specific operational guidance Claude could not infer — "The local `main` ref is often stale: squash or diff against `git merge-base HEAD origin/main`", "A queued or in-progress check is not green" — with zero explanation of concepts Claude already knows and no padding. Not 4 because no section could be trimmed without losing operational information. | 5 / 5 |
Actionability | Copy-paste ready throughout: exact commands (`gh pr list --search "<number>"`, `gh run rerun <id> --failed`, `gh api --paginate repos/{owner}/{repo}/pulls/<n>/reviews`, `git patch-id --stable`) and the verbatim Big Bonk trigger comment to post. Not 4 because the common cases are covered with complete, executable instructions including specific thresholds (~1,500 API calls, last ~25 runs, 10 minutes). | 5 / 5 |
Workflow Clarity | Eight numbered phases with explicit validation checkpoints and feedback loops: "once local targeted tests and `vp check` pass", a five-criterion checklist before `gh pr ready`, "Read the logs of each failure and separate code failures from infrastructure flakes", and loop termination "until the latest review on the latest head approves or reports no actionable findings". Not 4 because validation is explicit at every risky stage rather than having minor gaps. | 5 / 5 |
Progressive Disclosure | Clean split of concerns: shared confirm/reproduce/fix/validate/PR rules are delegated to [`references/rules.md`](references/rules.md) with named-section pointers and an explicit exclusion ("Its 'Automated Runs' section doesn't apply here"), while the skill-owned review/CI/Bonk loops stay inline. The single reference is one level deep, well-signaled, and exists in the bundle; not 4 because navigation is section-precise with no organization gaps. | 5 / 5 |
Total | 20 / 20 Passed |