Content
78%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-structured operational skill body: a clear 7-step workflow with explicit per-source error handling and a verified, cleanly split bundle. Remaining gaps are small — some duplicated trigger/status text, missing script invocation commands, and no post-write validation or in-run retry for failed sources.
Suggestions
Show how to run the scripts (e.g., a one-line example like `python3 scripts/crawl.py --output /tmp/crawl_result.json`) in the workflow or 资源 section so the orchestration path is executable, not just named.
Add an in-run retry or explicit fallback instruction for sources marked ❌ error before resorting to '建议下次重跑', and a final verification step that the written markdown's mermaid blocks render.
Trim duplication: drop the 触发词 section (it repeats the frontmatter description's triggers) and reference Step 3's status enum from Step 6 instead of redefining it.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense operational content — a source table, a structured per-source result schema, and seven terse workflow steps — with no explanations of concepts Claude already knows. It falls just short of 'Lean and efficient' due to redundancy: the 触发词 section repeats the description's 适用于 triggers, and Step 6's status enum re-lists the statuses already defined in Step 3. | 4 / 5 |
Actionability | Guidance is mostly executable: exact URLs, the dynamic-issue construction rule, a concrete JSON item schema, the explicit date-window rule ("date >= today - 7 days"), the output path with collision handling ("追加 -HHMM 后缀避免覆盖"), and named scripts for each stage. Minor gap: the body never shows how to invoke the listed scripts (e.g., a run command with arguments for scripts/crawl.py), so a couple of steps stop at naming the tool rather than executing it. | 4 / 5 |
Workflow Clarity | Steps 1–7 are clearly sequenced with strong validation checkpoints: per-source status classification (ok/empty/error/human_required), explicit fallback paths for anti-bot sources (search-based committer queries from references/human-verification.md), and a mandated coverage table. It sits below level 5 because error sources get '建议下次重跑' rather than an in-run retry loop, and there is no final verification of the written report (e.g., mermaid syntax/rendering check) before finishing. | 4 / 5 |
Progressive Disclosure | The body is a well-organized overview with clearly signaled, one-level-deep references that all resolve to real, substantive bundle files: the report structure lives in references/output-template.md, anti-bot fallbacks in references/human-verification.md, and crawling/parsing/rendering in the four scripts — all verified present. Nothing that belongs in a separate file is inlined, and no reference is nested beyond one level. | 5 / 5 |
Total | 17 / 20 Passed |