Content
100%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 exemplary operational skill body: exact runnable commands, an ordered audit with authorization gates and error-recovery paths, a priority system tied to measurable check ids, non-destructive fix mode, and honest-reporting rules that guard against over-claims. Dated vendor facts are properly quarantined in a refreshable reference matrix, and all six bundle references are real, well-signaled, and one level deep.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes Claude's competence (never explains what robots.txt, WAF, or CLS are) and every section earns its place: commands, a sequenced audit, a priority-to-check-id table, report structure, fix mode, error handling. Fast-moving dated facts are delegated to references ("For any vendor fact, use references/vendor-matrix.md... If a row is older than 60 days, say so or run /seo agentic refresh"), and the lone version mention (Lighthouse 13.2) functions as an error-handling diagnostic, not padding. This matches the "lean and efficient; every token earns its place" anchor rather than score 4, which requires over-explanation to trim. | 5 / 5 |
Actionability | Guidance is fully executable: exact commands like "${CLAUDE_PLUGIN_ROOT}/scripts/claude-seo" run lighthouse_agentic.py <url> --strategy both --json", copy-paste fix commands with example flags (--signal "search=yes, ai-input=yes, ai-train=no"), every priority row mapped to a tool check id, and an executable fallback (npx lighthouse@latest <url> --only-categories=agentic-browsing --output=json) in the error table. This matches the top anchor (copy-paste ready, common cases covered) rather than score 4, which allows minor gaps. | 5 / 5 |
Workflow Clarity | The audit is an explicitly ordered 6-step sequence ("Run the steps in this order and keep every tool's JSON for the report") with authorization gating for the WAF step, an Error Handling table mapping failure modes to concrete actions, and feedback loops ("how to confirm it worked (rerun the named check)", rerun with --from-json after a local Lighthouse run, the refresh procedure). Fix mode is explicitly non-destructive ("Drafts go to stdout for review; nothing is deployed and no file is overwritten"), so the missing-validation cap for destructive/batch workflows does not apply. Matches the top anchor with explicit validation steps, error-recovery loops, and a checklist-style priorities table. | 5 / 5 |
Progressive Disclosure | The SKILL.md body is an operational overview and all detail lives in six real, one-level-deep reference files (verified present in references/: lighthouse-agentic-category.md, agent-friendly-pages.md, access-policy.md, discovery-and-markdown.md, webmcp.md, vendor-matrix.md, each 75-133 lines). References are clearly signaled in context ("Read references/lighthouse-agentic-category.md before explaining it", "Read references/webmcp.md first") and indexed with one-line descriptions in a "Reference files" section; no nested references exist. Matches the top anchor (clear overview, well-signaled one-level-deep references, easy navigation) rather than score 4, which allows organization gaps. | 5 / 5 |
Total | 20 / 20 Passed |