Content
81%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-engineered advisor skill body: lean, opinionated, and driven by concrete commands, tables, and hard rules, with an explicitly validated multi-phase workflow. The only real weaknesses are mild persona padding, a subagent-prompt spec given as requirements rather than a template, and bundle references (AUDIT.md, PLAN-TEMPLATE.md) that do not resolve within the provided files.
Suggestions
Replace the 'Operating Posture' persona prose with a compact list of the leverage heuristics it gestures at, and consolidate the read-only rule into Hard Rules instead of restating it in the intro and Phase 1.
Provide the Phase 2 subagent prompt as a verbatim template block (as Hard Rule 4 already is), so the executor does not have to assemble it from a requirements list.
Ensure AUDIT.md and PLAN-TEMPLATE.md ship in the skill bundle next to SKILL.md so the two load instructions resolve for executors that only receive the bundle.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes React competence — no space is wasted explaining what hooks or memoization are, and every section carries operational weight. It stops short of a 5 because of atmospheric padding ('with a brutal eye for what ships to users', the 'Operating Posture' persona prose) and repeated assertions of read-only-ness across the intro, Hard Rules 1–2, and Phase 1. | 4 / 5 |
Actionability | Concrete copy-paste commands appear throughout: 'npx react-doctor@latest --json --json-out react-doctor-report.json', 'npx react-doctor@latest rules explain <rule>', 'git rev-parse --short HEAD', 'npx react-doctor@latest --scope changed', plus the canonical fix URL pattern and explicit output table formats. Minor gaps keep it at 4: the subagent prompt is specified as a list of requirements rather than a verbatim template, and the JSON report cleanup ('delete it when done') has no command. | 4 / 5 |
Workflow Clarity | Four clearly sequenced phases (recon → parallel audit → vet/prioritize → plans) with explicit validation checkpoints: Phase 3 mandates re-reading cited code ('Never present a finding you haven't confirmed at its file:line'), rejection criteria for by-design/duplicate/over-reported findings, a stop-and-wait user checkpoint, and mechanical plus behavioral verification steps baked into every plan. The effort-level table and invocation-variant table make scaling unambiguous. | 5 / 5 |
Progressive Disclosure | The body is a genuine overview that delegates the rule catalog to [AUDIT.md](AUDIT.md) and the plan format to [PLAN-TEMPLATE.md](PLAN-TEMPLATE.md), both one level deep and clearly signaled at point of use ('Load them when you audit and when you write plans'). Not a 5 because the two referenced files are not present in the bundle alongside SKILL.md, so the load instruction cannot be verified against actual content. | 4 / 5 |
Total | 17 / 20 Passed |