Content
67%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.
The body is a well-structured operational overview: imperative bullets, a numbered package-safety gate, an explicit six-item validation checklist, and a failure/recovery loop, with bundle references that exist and are one level deep. Its weaknesses are the inlined generic constraint boilerplate that dilutes the token budget and the duplication of script-scope detail already carried by references.
Suggestions
Move the 公共硬约束 boilerplate block into a separate reference file (e.g., references/COMMON_CONSTRAINTS.md) and keep a one-line pointer in SKILL.md, reclaiming roughly a third of the body for task-specific content.
Replace the bare script-name list under 官方入口验证 with the full build commands (as already documented in references/PRODUCT_LINE_RULES.md), or point explicitly to that table instead of restating the names.
Move the bug-recording rules out of 执行步骤 into a dedicated short section or reference, and number the top-level 流程 steps so the overall sequence is explicit rather than implied by section order.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The core workflow sections are lean imperative bullets ("只改与当前任务直接相关的文件", "绕过官方构建入口只跑裸 xelatex" under 避免), but roughly a third of the body is generic boilerplate that is not specific to this task — the eight 公共硬约束 bullets (secret-handling, SSRF/path-traversal, CHANGELOG policy) and the three bug-recording bullets in 执行步骤. That matches anchor 3's 'mostly efficient but includes some unnecessary explanation or could be tightened'; it is above 2 because nothing explains concepts Claude already knows, and below 4 because the shared-constraint block could be moved out of the main body. | 3 / 5 |
Actionability | Concrete executable guidance is present: "运行 python3 skills/make-latex-model/scripts/plan_package_regression.py <packages/bensz-*>", named official entries (nsfc_project_tool.py, paper_project_tool.py, thesis_project_tool.py, cv_project_tool.py), and named analysis scripts (analyze_pdf.py, compare_headings.py, compare_pdf_pixels.py, optimize_heading_linebreaks.py) that all exist in scripts/. This fits anchor 4's 'concrete code or commands with minor gaps'; it misses 5 because the official-entry commands are given as bare script names without invocation arguments in the body (the full commands live in references/PRODUCT_LINE_RULES.md), and several steps like 判断修改层级 defer their actual criteria to reference files. | 4 / 5 |
Workflow Clarity | The 流程 sections form a clear sequence (输入 → 判断验收口径 → 判断修改层级 → 最小范围实现 → 包层安全门禁 → 官方入口验证 → 输出 → 校验 → 失败与恢复), the package safety gate is a numbered 4-step procedure, 校验 is an explicit 6-item checklist, and 失败与恢复 provides a retry-from-last-failed-step feedback loop — matching anchor 4. It misses 5 because the overall sequence is implied by section order rather than explicitly numbered, and the 执行步骤 section mixes bug-recording policy into the flow rather than actual execution steps, slightly blurring the sequence. | 4 / 5 |
Progressive Disclosure | The body is an ~120-line overview with clearly labeled one-level-deep pointers to real bundle files — "产品线识别：references/PRODUCT_LINE_RULES.md", "脚本职责：references/SCRIPT_SCOPE.md", "工具说明：scripts/README.md" — all of which exist and contain the promised detail, matching guideline scoring against the actual bundle. This fits anchor 4; it misses 5 because the script list under 官方入口验证 duplicates content that SCRIPT_SCOPE.md already owns, and the docs/ paths (docs/WORKFLOW.md, docs/for-developers/*-template-standard.md, docs/BASELINE_GUIDE.md) point outside the bundle and cannot be navigated from it. | 4 / 5 |
Total | 15 / 20 Passed |