Content
88%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 process skill: the multi-phase workflow is unambiguous with hard checkpoints, validation feedback loops, and executable commands throughout, and detail is appropriately pushed into one-level-deep references. The two real weaknesses are internal redundancy (the CHAPTER-CRAFT.md single-entry message and the two overlapping when-to-read tables) and bundle paths (themes/, templates/scripts/tts-providers/) that are referenced in the body and consumed by scaffold.sh but missing from the bundle.
Suggestions
Consolidate the「各阶段文件读取指南」table and the「相关资源」table into a single canonical when-to-read mapping — they cover the same files and nearly the same reading conditions, and the CHAPTER-CRAFT.md '单一必读入口' reminder is repeated in four places that could point to one location.
Fix the dangling bundle references: the body points to templates/scripts/tts-providers/README.md and themes/*/theme.json, and scaffold.sh copies from $SKILL_DIR/templates and $SKILL_DIR/themes, but neither directory ships in the bundle — either include them or update the paths so agents don't hit missing files at Checkpoint Plan and Phase 2.1.
Reduce the repeated restatements of the 硬性自检协议 and 双源原则 (each appears in its own section plus again in §1.2 and §2.4) to a single definition with brief cross-references.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes Claude's competence (no teaching of Vite/React/TTS basics), but there is noticeable redundancy: the fact that CHAPTER-CRAFT.md is the "单一必读入口" is repeated in at least four places (reading-guide table, §2.3, §2.4, 十条原则 intro), and the「各阶段文件读取指南」table and the「相关资源」table substantially duplicate each other's when-to-read mapping. This fits 'efficient with minor instances of over-explanation that could be trimmed'; not score 3 because the repetition is deliberate long-session reinforcement rather than padding of concepts Claude already knows, and not score 5 because the duplicated tables and repeated reminders could be consolidated into one canonical location. | 4 / 5 |
Actionability | Guidance is fully executable: concrete commands ("bash <path>/scripts/scaffold.sh ./presentation --theme=<id>", "rm -rf presentation/src/chapters/01-example", "npx tsc --noEmit", "PRESENTATION_TTS=openai npm run synthesize-audio"), a fully specified working-directory tree, verbatim checkpoint message templates for user alignment, and a machine-checkable invariant ("if (step === N) 出现的最大 N + 1 必须等于 narrations.length"). It is not score 4 because the common cases are copy-paste ready with specific, quantified constraints. | 5 / 5 |
Workflow Clarity | The four-phase sequence is explicit with two mandatory checkpoints ("[Checkpoint Plan] 必须停", "[Checkpoint Audio] 必须停"), a forced first-chapter user acceptance gate, three named development modes with a stated default, and validation feedback loops: the self-check escalation ladder (Agent Teams → subagent → 自检) with the rule "先按 fail 项把产出改完，再向用户汇报...直接拿原始结论汇报但不修复 = 违规", plus tsc verification and the STORAGE_KEY bump rule after restructuring. This matches 'clear sequence with explicit validation steps; feedback loops for error recovery; checklists for complex processes'. | 5 / 5 |
Progressive Disclosure | Good structure with well-signaled, one-level-deep references organized by when-to-read (SCRIPT-STYLE, OUTLINE-FORMAT, CHAPTER-CRAFT, THEMES, AUDIO, RECORDING, EXAMPLES/ all exist in references/), and the body explicitly avoids one-shot reading ("按'何时读'标注，避免一次性全读"). However, the body references "templates/scripts/tts-providers/README.md" and "themes/*/theme.json" / "themes/"（每个含 theme.json + tokens.css）, which are absent from the actual bundle — only references/ and scripts/scaffold.sh exist, and scaffold.sh itself copies from the missing $SKILL_DIR/templates and $SKILL_DIR/themes, so an agent following the file would hit missing paths. This fits 'good structure; references mostly clear; minor organization gaps'; not score 5 because of the dangling referenced paths, and not score 3 because all other navigation is clear and one level deep. | 4 / 5 |
Total | 18 / 20 Passed |