Content
85%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is a well-engineered, highly actionable skill with a clear phased workflow, strong validation/feedback loops, and clean progressive disclosure into three real reference files. Its only weakness is conciseness: it runs long and occasionally restates guidance, sitting at the edge of the token budget rather than fully lean.
Suggestions
Trim or move the subagent-prompt bulleted list (Phase 2) and the economics paragraph into references/ to reduce SKILL.md length — they are valuable but not needed on first load.
Consolidate the duplicated vetting/reconciliation guidance that recurs across Phase 2, Phase 3, and the Invocation variants into a single statement plus a pointer.
Consider moving the full quick/standard/deep effort-level table into a reference and keeping only the default and the keyword mapping inline.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is high-signal and assumes Claude's competence, but at ~120 lines it is long for a SKILL.md and restates guidance in places (the economics paragraph, the verbose subagent-prompt bullet list, and the effort-level table), which is 'mostly efficient but could be tightened' rather than lean where every token earns its place. | 2 / 3 |
Actionability | Provides concrete, executable guidance throughout — exact commands ('git rev-parse --short HEAD', 'git merge-base', 'tsc --noEmit', 'npm audit', 'gh repo view --json visibility'), concrete directory layouts ('plans/001-<slug>.md'), explicit finding-table columns, and a precise subagent-prompt specification. | 3 / 3 |
Workflow Clarity | Four phases (Recon → Audit → Vet → Write plans) are clearly sequenced with explicit validation checkpoints ('Vet before presenting — subagents over-report', 'open the cited code yourself and confirm it'), drift detection via a stamped commit, and feedback loops (vet→downgrade/reject, executor diff review→verdict). | 3 / 3 |
Progressive Disclosure | The body is an overview that signals three one-level-deep references — audit-playbook.md, plan-template.md, closing-the-loop.md — each a real file under references/ and clearly flagged ('read it now', 'read it before writing the first plan', 'Read before the first dispatch'), with detail appropriately split out of SKILL.md. | 3 / 3 |
Total | 11 / 12 Passed |