Content
92%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 well-structured, highly actionable skill body: concrete git commands with explicit defaults, a hard gating checkpoint on loading references, a complete output template, edge-case handling for large diffs and trivial changes, and correctly externalized one-level-deep reference files. The only weakness is mild duplication between the phase instructions, the What-NOT-to-Do table, and the Conversation Style section.
Suggestions
Deduplicate repeated guidance: the commit-messages-contain-intent point appears in both Phase 2 and the What-NOT table, and 'be specific / avoid abstract advice' appears in both Conversation Style and the What-NOT table — keep each in one place to trim tokens.
Drop the stylistic framing ('Not a lecture -- a mirror. Show the user what their code already demonstrates') and fold its intent into the Phase 3 or Conversation Style bullet, since the body already demonstrates reflective tone through its instructions.
Consider merging the 'Conversation Style' bullets that restate What-NOT table rows (e.g., specificity and leading with positives) into a single guidance section to reduce redundancy.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is efficient — tables for scope and anti-patterns, a copy-paste output template, no explanation of git or SE basics Claude already knows. It stays at 4 rather than 5 because a few points are stated twice: commit messages 'contain intent that raw diffs miss' appears in both Phase 2 and the What-NOT table, and 'Be specific / abstract advice is worthless' is duplicated in Conversation Style and the What-NOT table; the framing line 'Not a lecture -- a mirror' is disposable flavor. It is clearly above the 'mostly efficient but some unnecessary explanation' midpoint because the padding is duplication, not concept explanation. | 4 / 5 |
Actionability | Concrete, copy-paste-ready git commands cover the common cases ('git log main..HEAD --oneline', 'git diff main...HEAD', 'git show <sha>', 'git diff --cached'), with explicit defaults (N=5, feature-branch vs main behavior) and a fill-in output template with per-field sentence counts. This matches the 'fully executable; specific examples cover the common cases' anchor — an instruction-only skill whose guidance is entirely actionable. | 5 / 5 |
Workflow Clarity | A clear four-phase sequence (load references → determine scope → gather → analyze → present) with explicit checkpoints: 'Do not proceed until you've loaded at least se-principles.md' is a hard gate, Phase 1 resolves scope ambiguity with a decision table plus defaults, Phase 2 has a conditional for large diffs (>500 lines → --stat, read top 3-5 files), and the What-NOT-to-Do table plus the trivial-changes escape hatch ('no deep lesson here, just good housekeeping') act as checklists and error-avoidance guidance. No destructive or batch operations exist, so no feedback loop is required; the sequence is unambiguous. | 5 / 5 |
Progressive Disclosure | The SKILL.md body is a workflow overview, and the catalog content is correctly externalized: 'Read references/se-principles.md to have the principle catalog available' and 'Optionally read references/anti-patterns.md' — both verified to exist, one level deep, no nested references inside them, and the optional file is clearly signaled as conditional. This matches the 'clear overview with well-signaled one-level-deep references; content appropriately split' anchor. | 5 / 5 |
Total | 19 / 20 Passed |