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.
An unusually actionable instruction-only skill: concrete persona selection rules, routing, severity semantics, an example output block, an explicit integrity checklist, and exit conditions with feedback loops. Its two weaknesses are redundancy (the anti-fabrication rule and several Rules/Anti-Patterns entries repeat) and the missing references/persona_principles.md file the entire workflow depends on.
Suggestions
Ship references/persona_principles.md with the bundle (or inline a minimal sourced-principles table): the body cites it 6+ times as mandatory grounding for Steps 1-2, but the file is absent, so the workflow's load-bearing step cannot be executed as packaged.
Consolidate the duplicated honesty rules: the anti-fabrication/no-quote-invention guidance appears in Attribution discipline, Rules, the Anti-Patterns table, and the Integrity check — state it once authoritatively and cross-reference it to save tokens.
Trim the 'Theory' cross-reference (de Bono, Kahneman) and fold the Anti-Patterns rows that restate Rules verbatim ('Fabricating a finding to hit ≥1 issue', 'Product person skipped') into the Rules section only.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and organized with no explanation of concepts Claude already knows, and the example output, routing, and severity tables each earn their tokens. However, the anti-fabrication rule is stated four times (Attribution discipline items 1 and 3, Rules, the Anti-Patterns table, and the Integrity check), several Anti-Patterns rows duplicate Rules verbatim, and the 'Theory' citation (de Bono, Kahneman) is padding — more than minor trimming available, so anchor 4 fits better than 3. | 4 / 5 |
Actionability | Fully concrete instruction-only guidance: exact persona counts ('pick 2 per round' engineers plus 1 mandatory product persona), a routing table by problem type, severity definitions each with a required action, a copy-ready example output block, a quantified zero-finding burden ('3+ principles the code demonstrably satisfies'), and a default output destination ('.claude/review-[timestamp].md'). Covers the common cases completely — anchor 5. | 5 / 5 |
Workflow Clarity | Clear sequence (Step 0 read twice → Step 1 ground principles → Step 2 review → Step 3 synthesize and post) with explicit validation checkpoints: the Integrity Check is a 4-item checklist run after each round, and the Exit Conditions define feedback loops (BLOCKER/CRITICAL → fix → 1 re-review round; CLEAN on 2 consecutive rounds → done). This is a review process, not a destructive/batch operation, so no cap applies — anchor 5. | 5 / 5 |
Progressive Disclosure | References are well-signaled and one level deep ([references/persona_principles.md] linked from multiple sections, cross-references to sibling skills), and persona detail is appropriately split out of SKILL.md rather than inlined. However, that reference file — cited 6+ times as the load-bearing grounding resource required by Step 1 and Rule 1 — is not present in the bundle, leaving the skill's core dependency unshipped. Good structure with a real gap: anchor 4, not 5. | 4 / 5 |
Total | 18 / 20 Passed |