Content
81%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 tightly engineered 8-phase workflow skill with outstanding sequencing, hard checkpoints, per-node QA protocol, and phase-indexed progressive disclosure. Its two real weaknesses are the missing theme-profiles/ bundle directory (which breaks both the Checkpoint 1 theme step and the scaffold.sh entry command as shipped) and systematic restatement of the decision-collection/QA rules across four sections.
Suggestions
Ship the missing theme-profiles/ bundle (index.json + per-theme <id>.md): it is referenced in the Checkpoint 1 decision table ('读 theme-profiles/index.json'), the Phase 2 reading guide, the resources table, and throughout references/, and scripts/scaffold.sh hard-fails without it (theme_exists greps $SKILL_DIR/theme-profiles/index.json), so the default 'scaffold.sh ./my-article --theme=<id>' command aborts as bundled.
De-duplicate the decision-collection iron rule: it is stated in full in 硬性质检协议 (rule 5), restated in Phase 3, re-referenced in Checkpoint 2, and recapped in 默认策略 — keep one canonical statement and cross-reference it from the other locations.
Trim the cover / PDF / language-policy recaps that repeat nearly verbatim between the phase sections (Phase 1, Phase 4, Phase 8) and the 默认策略 section.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Per-line density is high — aggressive use of tables, no explanation of concepts Claude already knows — but core rules are systematically restated: the decision-collection iron rule ('禁止静默替用户选择…每项独立列出 + 等用户答复') appears near-verbatim in 硬性质检协议 rule 5, again in full in Phase 3, referenced again in Checkpoint 2, and recapped in 默认策略; the QA methods and cover/PDF/language policies are likewise duplicated between the 质检协议 table, phase sections, and 默认策略. Efficient overall with more than minor trimming opportunity — anchor 4, not 5; clearly above 3 since the redundancy is policy reiteration, not concept padding. | 4 / 5 |
Actionability | Highly concrete throughout: executable commands ('bash <path-to-beautiful-article>/scripts/scaffold.sh ./my-article --theme=<id>', 'npm run typecheck / build', 'bash <skill>/scripts/html-to-pdf.sh'), a copy-paste Plan Checkpoint opening-message template, an exact workspace layout, section file-naming convention ('article/sections/NN-*.tsx'), and fully enumerated option tables with retention ratios. The gap: the primary Phase 4 entry command fails as bundled — scaffold.sh's theme validation greps '$SKILL_DIR/theme-profiles/index.json', which does not exist in the bundle, so even the default theme aborts with '未知主题'. Mostly executable with one real execution gap — anchor 4, not 5; well above 3 since nothing is vague or pseudocode. | 4 / 5 |
Workflow Clarity | Exemplary sequencing: a Phase 0→8 overview diagram, three hard-stop checkpoints ('★Checkpoint 1 必须停…逐项确认 5 件事'), a per-node QA table mapping node→method→artifact with rationale, explicit feedback loops ('先按 fail 项把产出改完,再汇报…直接拿原始结论汇报但不修复 = 违规'), inline checklists, and a minimal-slice repair policy ('禁止只反馈一处就重写整篇'). This is exactly the anchor-5 pattern: clear sequence, explicit validation steps, error-recovery loops, and checklists for a complex process. | 5 / 5 |
Progressive Disclosure | Navigation is excellent — a per-phase reading table ('各阶段文件读取指南…必读 | 按需查'), a resources table annotated by '何时读', and one-level-deep references, all of which resolve to real bundle files except one: the body instructs '读 theme-profiles/index.json' (Checkpoint 1 table), lists 'theme-profiles/*.md' in the Phase 2 reading guide, and 'theme-profiles/index.json + *.md' in the resources table, but no theme-profiles/ directory exists in the bundle (also required by scaffold.sh and 10+ cross-references in references/*.md). Good structure with one significant broken reference — anchor 4, not 5; clearly above 3 since signaling and content placement are otherwise exemplary. | 4 / 5 |
Total | 17 / 20 Passed |