Content
77%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 rich, highly actionable persona skill with an exemplary checkpointed workflow and concrete worked examples. Its main weaknesses are verbosity from restating the same core principles across multiple sections, and a progressive-disclosure failure: six reference files exist but are never linked from the SKILL.md body.
Suggestions
Link the references/ bundle files from the body (e.g. under a '## 调研来源' section: '完整一手来源调研见 references/费曼著作与系统思考调研-20260404.md 等'), and move the per-model 来源证据 detail into those files so SKILL.md stays a lean overview.
De-duplicate the core principles: the 快速参考 and 反例黑名单 largely restate 核心心智模型/决策启发式 — consolidate or cross-reference instead of repeating, and merge the redundant 中文输出适配 points into the 句式/词汇 subsections.
Trim the expression DNA to concrete rules + one worked example per rule rather than restating each rule across 句式, 词汇, 节奏, 幽默, and 态度光谱.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~490-line body is mostly efficient and assumes Claude's intelligence (it does not explain physics or basic concepts), but the same handful of principles (命名≠理解, 货物崇拜, 演示>论证, 不自欺) are restated across 核心心智模型, 决策启发式, 快速参考, 反例黑名单, and 失败模式树, and the 中文输出适配 section repeats points already made in 句式/词汇, so it could be tightened — fitting the 'mostly efficient but some unnecessary explanation' anchor rather than the lean anchor above. | 3 / 5 |
Actionability | For an instruction-only persona skill it gives highly executable guidance: explicit roleplay rules (用「我」, one-time STOP disclaimer, EXIT trigger), a 3-step protocol with concrete checkpoint questions, two fully worked example dialogues, per-model 检测问题, per-heuristic 应用场景+案例, a failure-mode table with triggers/fixes/fallbacks, and an anti-pattern blacklist with concrete replacements — matching the 'fully executable, specific examples cover common cases' anchor. | 5 / 5 |
Workflow Clarity | The 回答工作流 is a clearly sequenced 3-step protocol (问题分类 → 研究 → 回答) with two explicit gating 🔴 CHECKPOINTS, each listing the exact questions that must be answerable before proceeding, plus a 问题分类 table and a fallback tree for error recovery — matching the 'clear sequence with explicit validation steps and feedback loops' anchor. No destructive/batch operation cap applies. | 5 / 5 |
Progressive Disclosure | Internal sectioning is reasonable (clear headers for 角色扮演规则, 回答工作流, 核心心智模型, 决策启发式, 表达DNA, etc.), but the 6 reference bundle files in references/ (the ~100KB of 调研 material) are never linked or signaled from the body, and content that could live in separate files (the full mental-model/heuristic/DNA details) is inlined into a ~490-line monolith — fitting the 'some structure, references not signaled, content that should be separate is inline' anchor rather than the well-signaled one-level-deep anchor above. | 3 / 5 |
Total | 16 / 20 Passed |