Content
92%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A highly actionable, well-sequenced triage skill with executable commands and clear validation checkpoints; its only weakness is progressive disclosure, as all content lives inline in a single 174-line file with no referenced detail files.
Suggestions
Extract the Classification table and/or PostHog CI notes into a referenced reference file (e.g. references/classification.md) and link to it one level deep, so SKILL.md stays an overview and progressive disclosure can reach a 3.
Tighten the repeated 'hogli ci:insights is fresher than the gh API because it can lag or rate-limit' rationale, which appears ~3 times (intro, 'Always start with' paragraph, and section 1) — state it once.
Consider a short explicit validation/verification checkpoint before the Report step (e.g. confirm the classified class is consistent with the extracted failing command) to make the workflow's feedback loop fully explicit.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean, table-and-command dense body that assumes Claude's competence (no explaining of what CI or GitHub Actions is); every section earns its place, though the 'insights is fresher than the gh API / can lag or rate-limit' rationale is restated ~3 times and could be tightened. Still closest to the 'lean and efficient' anchor. | 3 / 3 |
Actionability | Fully executable, copy-paste-ready commands throughout (hogli ci:insights variants, gh pr checks, gh run view --log-failed) plus concrete classification and local-reproduction tables giving exact first actions per class. | 3 / 3 |
Workflow Clarity | Clear numbered sequence (start with insights, then find the failing run) with explicit checkpoints ('Stop at the first failing step'), an error-recovery fallback (insights not installed -> fall back to gh), and a defined handoff (flaky -> fixing-flaky-tests); the skill is read-only triage so the destructive-ops feedback-loop cap does not apply. | 3 / 3 |
Progressive Disclosure | Well-sectioned (Safety rules, Workflow, Classification, Local reproduction, PostHog CI notes, Report shape) but it is a ~174-line monolith with an empty bundle and no one-level-deep references to split material into; over the 50-line simple-skill threshold, so it does not meet the 'well-signaled one-level-deep references; content appropriately split' bar for a 3. | 2 / 3 |
Total | 11 / 12 Passed |