Content
92%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 exceptionally operational skill body: exact executable commands, a strictly ordered audit with priority-gated fixing, an honest-reporting ruleset, and a complete error-recovery table, all kept lean with detail correctly offloaded to six real reference files. The only weakness is a couple of fast-moving vendor/version facts stated inline in the body instead of being confined to the dated vendor-matrix reference.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and lean — tables for commands, priorities, and error handling, no explanations of concepts Claude already knows, and every section carries operational rules (e.g. "Never convert it to a percentage and never assume N"). It loses a point only for time-sensitive facts inlined in the body rather than confined to the dated references: "WebKit opposes it, Mozilla is neutral, and only ChatGPT desktop calls tools by default" and "The report predates Lighthouse 13.2" — minor instances of the version/vendor-date pattern the rubric penalizes, keeping it just below the every-token-earns-its-place anchor 5. | 4 / 5 |
Actionability | Fully executable, copy-paste-ready commands with exact flags and example values throughout: `"${CLAUDE_PLUGIN_ROOT}/scripts/claude-seo" run lighthouse_agentic.py <url> --strategy both --json`, `agentic_fix.py ai-catalog --publisher example.com --entry "Name|media type|url"`, and a concrete fallback `npx lighthouse@latest <url> --only-categories=agentic-browsing --output=json`. Check ids, evidence fields (`inconclusive: true`, `html_only_fallback`), and tool annotations are specified to the exact field name, matching anchor 5. | 5 / 5 |
Workflow Clarity | The audit process is an explicitly ordered 6-step sequence ("Run the steps in this order"), fixes are priority-sequenced ("Fix in order P0, then P1. Do not recommend lower priorities while a measured P0 fails"), recommendations must state "how to confirm it worked (rerun the named check)", and the Error Handling table gives explicit recovery loops (quota exceeded → local run then `--from-json`; 429 → report not-tested and rerun later). Clear sequence, explicit validation, and feedback loops for error recovery match anchor 5. | 5 / 5 |
Progressive Disclosure | The body is a genuine overview: it keeps only the process, priorities, and rules inline, and pushes all detail to six one-level-deep reference files, all of which exist in the bundle and are each summarized in the "Reference files" section (e.g. "references/vendor-matrix.md: dated, source-graded vendor facts and the refresh procedure"). Inline pointers are topic-matched ("Read `references/webmcp.md` first"), matching the anchor-5 pattern of well-signaled, one-level-deep references. | 5 / 5 |
Total | 19 / 20 Passed |