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.
The body is a well-structured behavioral meta-skill: concrete announce and routing templates, a rationalization counter-table that doubles as a self-check, verified one-level-deep platform references, and explicit precedence rules for user instructions versus skills. Its only real weakness is token overhead from the emphatic <EXTREMELY-IMPORTANT> repetition and the repo-provenance commentary, neither of which adds runtime guidance.
Suggestions
Compress the <EXTREMELY-IMPORTANT> block from four sentences to one — keep "如果有 1% 的可能适用，必须调用" and drop the "你没有选择 / 这不可谈判" restatements, since the 红线 table already carries the enforcement.
Move the provenance blockquote (commit `392ff75`, PR #123, and the eval-evidence policy) to the repository's CLAUDE.md or a contributing guide; it addresses future editors of the skill, not the runtime audience.
Trim the editorial rationale in the 中国特色技能路由 section ("上下文里多一份排版参考不会让你写得更好，只会挤掉真正需要的内容") to a single normative sentence about explicit-invocation-only loading.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient — the 红线 table rows and 平台适配 list are tight and assumption-free — but the <EXTREMELY-IMPORTANT> block spends four sentences ("你没有选择。你必须使用它。这不可谈判。你不能通过合理化来逃避。") restating one instruction, and the fork-provenance blockquote (commit `392ff75`, PR #123, eval-evidence policy) is repo history rather than runtime skill guidance. Matches the anchor for mostly efficient content that includes some unnecessary explanation. Not 4 because these are more than minor trims; not 2 because the majority of the body is lean with no re-teaching of known concepts. | 3 / 5 |
Actionability | The guidance is directly executable throughout: an exact announce template ("宣布'使用 [技能] 来 [目的]'"), a concrete todo rule ("为每个条目创建一个待办"), concrete routing pairs ("'让我们构建 X' → 先用 brainstorming"; "'修复这个 bug' → 先用 systematic-debugging"; "/chinese-code-review" → chinese-code-review), and explicit file paths (`references/claude-code-tools.md` etc., all verified to exist). For an instruction-only skill this matches the fully-executable anchor whose specific mappings cover the common cases. Not 4 because there is no material gap — every instruction has a concrete object. | 5 / 5 |
Workflow Clarity | The core loop is clearly sequenced — check skills before any response or action ("包括澄清性问题、探索代码库、或查看文件之前") → invoke → announce → follow strictly with todos — and the 红线 table ("这些想法意味着停下——你在合理化") plus the 技能优先级 and 用户指令 precedence rules act as self-check checkpoints for the primary failure mode. Matches the anchor for a clear sequence with most checkpoints present. Not 5 because how to discover which skills are applicable is never stated (left implicit to the platform references); not 3 because the sequence and all edge cases (unsuitable skill, conflicting skills, user-instruction override) are handled explicitly. | 4 / 5 |
Progressive Disclosure | The body is a clear overview with nine condition-gated, one-level-deep references (Claude Code → `references/claude-code-tools.md` through Muse → `references/muse-tools.md`), every one of which exists in the bundle, and nothing that belongs in a reference file is inlined. Matches the anchor for a clear overview with well-signaled one-level-deep references and easy navigation. Not 4 because navigation is by explicit environment condition and the chinese-* table likewise signals exactly when (and when not) to load those skills. | 5 / 5 |
Total | 17 / 20 Passed |