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.
An exemplary router skill body: an intent→file table with exact paths, clear load rules with count limits, fallback escalation to a capability index, and explicit boundary/stop conditions. All referenced bundle files exist and stay one level deep; the only flaw is repeated delegation boilerplate across table rows.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and router-shaped (one-line principles, a compact intent→file table), but the boilerplate "已晋级为独立 Skill…（已安装时优先直接使用；本卡仅作原文与背景补充）" repeats verbatim across six table rows and could be a single footnote — minor trimming possible, not severe padding. | 4 / 5 |
Actionability | "先读本文件，再按路由表加载 1 张能力卡" backed by exact file paths per intent (e.g., references/capabilities/happiness-skill.md), explicit load-count limits ("最多加载 2 张"), and separate routing for non-capability queries — fully executable guidance covering the common cases for an instruction-only skill. | 5 / 5 |
Workflow Clarity | The workflow is a complete decision loop: read entry → route via the intent table → escalate to capability-index.md for misses → "明确告知超出本书范围，不硬套书中框架", with explicit stop conditions for medical/legal/safety topics. This fallback-and-stop path is the error-recovery equivalent for a knowledge router (no destructive/batch ops requiring validation). | 5 / 5 |
Progressive Disclosure | A lean ~60-line entry file pointing to 19 real one-level-deep capability cards (all verified present in references/capabilities/) plus overview/glossary/cheatsheet/capability-index, organized by a navigable intent table; the inlined 核心原则 section is explicitly justified for overview-type questions ("概览类问题读到这里即可回答"). | 5 / 5 |
Total | 19 / 20 Passed |