Content
63%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 lean, substantive review skill with a concrete output template, explicit severity taxonomy, and useful self-check gates. Its main weaknesses are redundant directive sections (the same rules stated three to four times), an inlined checklist duplicating a reference file, a malformed Output Format block, and incomplete navigation to the bundle — three reference files are orphaned and one cross-skill link is broken.
Suggestions
Consolidate 'Red Flags', 'Rationalization Prevention', and 'Anti-Patterns' into a single section — the ignore-style rule is stated four times and evidence rules three times across the body; state each rule once (conciseness).
Remove the inlined four-item 'Review Checklist (Mandatory)' that duplicates references/checklist.md and keep only the link to it (conciseness).
Add links to the unmentioned bundle files (lenses.md, report.md, request-template.md) in the References section, and fix or remove the broken ../common-tdd/references/quality-contract.md path (progressive_disclosure).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | There is no basic-concept padding and the prose is terse, but the same directives are restated across sections: the ignore-style rule appears in Review Principles ("Substance > Style"), Red Flags ("Stop if you are reviewing style only"), and Anti-Patterns ("No Nitpicking"), and evidence rules appear three times ("Evidence First", "Stop if a claim lacks evidence", "'Probably is not evidence'"). The inlined four-item checklist also duplicates references/checklist.md. Mostly efficient but could be meaningfully tightened — not 4 (redundancy exceeds minor trimming), not 2 (no filler or concept explanation). | 3 / 5 |
Actionability | Mostly concrete and executable: a near copy-paste output template ("[SEVERITY] [File] Issue Description / Why: / Fix:"), defined severity tags ([BLOCKER]/[MAJOR]/[NIT]), and a concrete three-part Test-Review Evidence Gate. Not 5 because "Sync: Enforce active framework P0 rules" is cryptic, "AC" is undefined jargon, the Questions>Commands example is typo'd ("' this handle null?'"), and the prose instruction ("Every substantive finding must include the literal `Why:` field...") is embedded inside the Output Format code fence rather than outside it. Above 3 since the majority of the guidance is directly actionable. | 4 / 5 |
Workflow Clarity | The review flow (apply checklist → evidence-gated findings → severity-tagged output) is inferable from the sections, and explicit self-check checkpoints exist ("Stop if a claim lacks evidence", the Test-Review Evidence Gate before flagging [MAJOR] missing-test findings). Not 5 because there is no explicit end-to-end sequence ordering the sections; above 3 because checkpoints are explicit rather than merely implicit. | 4 / 5 |
Progressive Disclosure | SKILL.md is a proper overview linking references/checklist.md and references/output-format.md one level deep, but three of the five bundle files (lenses.md, report.md, request-template.md) are never referenced from the body, and the body links to ../common-tdd/references/quality-contract.md, which does not exist in the bundle. The inlined checklist also duplicates a reference file. References present but not all clearly signaled and one broken — fits anchor 3; not 4 (orphaned and broken references exceed minor gaps), not 2 (real structure and working links exist). | 3 / 5 |
Total | 14 / 20 Passed |