Content
86%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, actionable, and uses progressive disclosure effectively with verified reference files. The main gap is a missing explicit validation checkpoint before destructive overwrite/patch operations.
Suggestions
Add an explicit pre-overwrite validation step (e.g., re-fetch and confirm intended content before `+overwrite`/`+patch`) to close the destructive-operation validation gap.
De-duplicate the retry/boundary guidance between 快速决策 and 核心边界 to tighten conciseness.
Include one complete end-to-end `+patch` command with all required flags inline so the destructive workflow is fully copy-paste ready without opening the reference.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean and assumes Claude's competence (no definition of Markdown/Feishu), but the 快速决策 and 核心边界 sections overlap slightly on retry/boundary guidance, leaving minor trimmable redundancy. | 4 / 5 |
Actionability | Concrete executable commands with real flags (`+create`, `+patch --file-token ... --regex --pattern ...`) and a copy-paste BAD/GOOD regex-escaping example cover the common cases. | 5 / 5 |
Workflow Clarity | The +patch sequence (download→local replace→overwrite) is explicit with error-class handling (stop vs. limited retry) and pre-flight checks, but lacks an explicit 'verify before overwrite' checkpoint for the destructive +overwrite/+patch path. | 4 / 5 |
Progressive Disclosure | A concise overview with a shortcuts table linking to five real one-level-deep reference files (verified present in references/), plus clearly signaled cross-skill links to lark-shared and lark-drive. | 5 / 5 |
Total | 18 / 20 Passed |