Content
85%Weight 40%Scale 1-3Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is a dense, actionable procedural specification with clear sequenced workflows, validation gates, and well-signaled one-level references; its main weakness is mild redundancy across the contract and focus sections that lengthens an already substantial document.
Suggestions
Consolidate the head-SHA / authoritative file-list establishment guidance so it lives in one section (Scope and Evidence) rather than being restated in Core Contracts and the Decision Contract.
Merge the CI-vs-code-correctness rules so the Review Focus table is the single source, with the Decision Contract referencing it instead of restating the conditions.
Consider moving the detailed Output Contract format spec into a reference file, keeping SKILL.md as the overview, which would reduce body length while preserving the copy-ready formatting detail.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is information-dense with no concept-explanation padding, but at ~256 lines it restates some concerns across Core Contracts, Review Focus, Decision Contract, and Review Workflow (e.g. head-SHA/file-list establishment and CI-vs-correctness rules appear in multiple sections), so it could be tightened despite being mostly efficient. | 2 / 3 |
Actionability | Guidance is highly concrete for an instruction-only skill: an ordered Decision Contract, exact status strings ('Local Preflight Result: Pass'), specific script paths ('scripts/build_review_inventory.py'), and a precise output-format specification give copy-ready, executable direction. | 3 / 3 |
Workflow Clarity | A clear 6-step Review Workflow is sequenced with explicit validation checkpoints — the Finding Proof Gate, the coverage-audit completion conditions, and 'Stop only after a full pass finds no new independent actionable finding' — plus an adversarial feedback loop for a batch/audit operation. | 3 / 3 |
Progressive Disclosure | The overview points to four real reference files (evidence-access.md, high-risk-review.md, sql-parser-review.md, review-corrections.md) and real scripts, each one level deep and clearly signaled with 'Read ... when' triggers; sections are well organized for navigation. | 3 / 3 |
Total | 11 / 12 Passed |