Content
85%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 instruction-style skill body: tight tables replace prose, execution detail is correctly pushed to a single one-level reference with real scripts, and the fetch/save workflow carries explicit failure inspection, retry, and non-overwrite validation. The only costs are slight redundancy between the Outcome Contract and later sections, and the absence of inline copy-paste commands in SKILL.md itself.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and rule-driven (routing table, fetch tiers, gotcha table) and assumes Claude's competence throughout, e.g. 'fetch.sh URL ... falls back to a stdlib HTML stripper'. Minor over-explanation could be trimmed — the four-part 'Outcome Contract' (Outcome/Done when/Evidence/Output) partially duplicates the later Routing/Saving/Output sections, and the 🥷 prefix line is decorative. Above the 'mostly efficient but some unnecessary explanation' anchor, short of the lean 5 anchor where every token earns its place. | 4 / 5 |
Actionability | Concrete guidance dominates: a routing table mapping exact domains to methods, 'load references/read-methods.md and run the commands', 'pip install --user readability-lxml html2text', 'fetch.sh --use-proxy URL', the structured stderr contract '[fetch] tier=<name> status=<ok|fail> reason="..."', and exact save/output templates. Mostly executable with minor gaps — the body itself carries few copy-paste code blocks (the actual commands live in the reference file), matching the 'mostly executable guidance' anchor rather than the fully copy-paste-ready 5. | 4 / 5 |
Workflow Clarity | The sequence is explicit (route → load read-methods.md → run command → inspect stderr tier → save or display → report) with validation checkpoints and feedback loops: 'Read the stderr if a fetch fails; it names the specific tier and reason', 'Network failures → prepend local proxy env vars ... and retry once', 'Built-in fetcher first; WeChat browser script if extraction fails', and a Gotchas table of failure modes with recovery rules. Save operations are guarded ('append -1, -2 ... Never overwrite without confirmation'), so the destructive/batch cap does not apply. Matches the 5 anchor's explicit validation, error-recovery loops, and failure checklist. | 5 / 5 |
Progressive Disclosure | SKILL.md is a true overview: routing, policy, and output contracts inline, with the command detail delegated by a clearly signaled single reference ('After routing, load references/read-methods.md and run the commands for the chosen method'). The reference exists, is one level deep, and contains the executable commands per method; the referenced scripts (fetch.sh, fetch_feishu.py, fetch_weixin.py, fetch_local.py) all exist in scripts/. Matches the 5 anchor's clear overview with well-signaled one-level-deep references. | 5 / 5 |
Total | 18 / 20 Passed |