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.
The body is well-structured with excellent progressive disclosure — a verified, one-level-deep reference bundle and clear navigation. Its weaknesses are moderate redundancy in the lint instructions, maintainer-oriented sections that inflate token cost, and a missing error-recovery loop after lint failures.
Suggestions
State the lint run instructions once (in Quick start) and reference them from the "How it works" table to remove the near-verbatim duplication.
Trim or relocate the "Rule boundary" and "Rule format" sections into RULES_INDEX.md, keeping only a one-line pointer, since they primarily serve skill maintainers rather than execution.
Add an explicit feedback step after lint runs (e.g. "If lint exits 1, read the flagged rule's 修复指引, fix the code, and re-run") to close the validation loop.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly tight tables and lists with no concept-teaching, but the lint copy-to-temp instructions are repeated near-verbatim in both the "How it works" table and "Quick start", and "Rule boundary"/"Rule format" lean toward maintainer meta-documentation. Not 4 because the duplication and meta sections go beyond "minor" over-explanation; not 2 because there is no padded explanatory content. | 3 / 5 |
Actionability | Provides concrete file paths ("references/rules/cross-cutting/AUTH001.md"), a copy-paste command ("node cloudbase-lint.mjs --project-dir ."), and exact sections to read ("LLM 检查"). Not 5 because it omits what to do after lint reports issues and how to report findings; not 3 because the guidance is directly executable as written. | 4 / 5 |
Workflow Clarity | Quick start gives a clear two-step sequence with the lint layer as a deterministic validation checkpoint and "inspect your code before claiming done" as a completion gate. Not 5 because there is no explicit fix-and-re-run feedback loop in the body for lint failures; not 3 because sequence and checkpoints are mostly explicit. | 4 / 5 |
Progressive Disclosure | SKILL.md is a lean overview with all 10 referenced bundle files verified present, organized one hop deep by module (cross-cutting / postgresql / storage), plus a full link index and a RULES_INDEX.md matrix for navigation. Bulk rule content is appropriately deferred to reference files with no deep nesting. | 5 / 5 |
Total | 16 / 20 Passed |