Content
82%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-style SKILL.md: maximally token-efficient, unambiguous scenario routing to real per-task reference files, and explicit boundary and fallback rules. The only deductions are second-level reference nesting in the bundle and executable command detail living in references rather than the body.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is a dense, imperative router with zero concept explanation and no padding — every line is a behavioral directive (routing, fallback rules, guardrails), fully assuming Claude's competence. Leaner than the anchor-5 example is hard to achieve for a multi-scenario skill. | 5 / 5 |
Actionability | Gives exact shortcut names (+fetch, +update, +media-insert, +history-revert), concrete commands (lark-cli drive files copy, whiteboard +update, --as user, auth status --verify), and one reference per scenario. Not a 5 because copy-paste command syntax for the core operations is delegated to the reference files rather than shown in the body. | 4 / 5 |
Workflow Clarity | Clear sequencing (先判断场景 → read that scenario's reference once → execute), ordered file-resolution fallback rules, an explicit auth error → read lark-shared → fix → retry feedback loop, and a whiteboard guardrail (必须复用现有 token,禁止新建空白画板). Minor validation gaps remain in the body for edit and history-revert flows, which are left to the references. | 4 / 5 |
Progressive Disclosure | A lean overview routing each scenario to exactly one clearly labeled reference file, all 12 of which exist in the bundle, plus an explicit instruction to read references per-phase rather than upfront. Not a 5 because the bundle has second-level nesting (lark-doc-update.md → lark-doc-xml.md → lark-doc-xml-extended-blocks.md), so some detail sits two-to-three hops from SKILL.md. | 4 / 5 |
Total | 17 / 20 Passed |