Content
92%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: five clearly sequenced phases with backups, scripted validation, bounded feedback loops, and disciplined load-on-demand references, all backed by real bundle files. The single weakness is density — the agent-compatibility/version blockquote and a few duplicated handoff rules could be moved or trimmed for token efficiency.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with operational rules and contains no padding or explanation of concepts Claude already knows; nearly every sentence is a boundary, decision rule, or step. However, the Agent-compatibility blockquote (four hosts' config paths and the multi-branch agents_version notice) is detail that could be trimmed or moved into agent-calls.md, and some rules (e.g. the /story-long-write handoff) appear in both 执行边界 and 流程衔接. This lands at anchor 4 (efficient, minor trimmable instances) rather than 5. | 4 / 5 |
Actionability | Guidance is fully executable: exact commands with flags and exit-code semantics ("node scripts/check-ai-patterns.js --check --fail-on=blocking <正文文件...>", "退出 1 表示有 blocking,退出 2 是命令或文件错误"), a concrete backup path convention (".story/work/去味原稿/{YYYYMMDD}_{原文件名}"), explicit loop limits ("全文扫描最多三轮"), quote-mode flags, and two copy-paste report templates. All cited scripts and reference files exist in the bundle. This matches anchor 5 (copy-paste ready commands covering common cases). | 5 / 5 |
Workflow Clarity | The five phases (Phase 1–5) are clearly sequenced, and the file-editing workflow has explicit validation and feedback loops: backup before edit, Phase 4 回读 against the original and set designs, re-scan with --fail-on=blocking, regenerate damaged segments on degeneration blocking, and bounded retry ("没有新问题即停…最多三轮"). This matches anchor 5 (clear sequence, explicit validation, error-recovery loops). | 5 / 5 |
Progressive Disclosure | SKILL.md is a genuine overview: a table maps the three method files to what each checks, and every reference is load-on-demand with an explicit read condition ("完整词条仅在需要定位某种表达时查 banned-words.md", "只在此时读 agent-calls.md", "短篇另读 short-deslop.md"). All 8 referenced files and 4 scripts exist, and references are one level deep with no pointer chains hiding content — matching anchor 5. | 5 / 5 |
Total | 19 / 20 Passed |