Content
71%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-structured, highly actionable instruction skill with excellent progressive disclosure and explicit validation steps. Its main weakness is density: several judgment rules run long with stacked clauses and could be materially tightened, and the overall workflow order is implicit across sections.
Suggestions
Split multi-clause rules (e.g., "A piece has a speaker" and "Material gate before drafting long-form") into short lead sentences plus sub-bullets, cutting each by roughly half while preserving the policy.
Add one compact end-to-end sequence (Pre-flight, Mode Picker, Hard Rules, Punctuation Gate, Output) near the top so the reader does not have to reassemble the pipeline from separate sections.
Move the per-mode "Activate when" trigger lists into the existing mode table or the frontmatter when_to_use field to avoid duplicating trigger phrases in two places.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body avoids known-concept padding and offloads heavy catalogs to references, but several rules stack three or four caveats apiece (the ~120-word "A piece has a speaker" bullet and the ~160-word "Material gate before drafting long-form" paragraph) that could be tightened without losing policy content. | 3 / 5 |
Actionability | Concrete guidance throughout: a mode-picker table mapping asks to specific files, language-detection load rules, and a runnable punctuation-gate script with fallback paths. The bash block requires placeholder substitution (<skill-base-dir>, <lang>) and no inline before/after edit example is given, so it is not copy-paste ready. | 4 / 5 |
Workflow Clarity | Explicit validation checkpoints exist (Punctuation Gate before returning text, batch reconciliation, scoped diff plus verification for repo edits) and Gotchas defines a re-review feedback loop, but the end-to-end sequence (Pre-flight, Mode Picker, Hard Rules, Gate, Output) must be assembled by the reader across separate sections rather than presented as one pipeline. | 4 / 5 |
Progressive Disclosure | SKILL.md acts as a clean router: every referenced file exists (11 reference files and the gate script), references are one level deep with explicit load conditions and markdown links, and large catalogs (e.g., 55KB write-zh.md) are appropriately split out. | 5 / 5 |
Total | 16 / 20 Passed |