Content
81%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 dense, well-structured instruction-only audit methodology with concrete templates, explicit evidence-sufficiency gates, and worked pressure tests. The main gaps are the absence of a fully filled worked example and some length that could be offloaded to reference files.
Suggestions
Add one fully worked example: a completed claim ledger row plus the resulting three-axis verdict for a concrete claim (e.g. the MemU 65% case), so the templates are demonstrated end-to-end.
Move the detailed minimal execution format and/or Pressure Tests into a separate reference file referenced one level deep from SKILL.md to tighten the core overview.
Trim the Common Mistakes table to the highest-impact rows or collapse low-frequency entries to reduce length while preserving the failure-mode coverage.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Densely packed operational methodology with no padding of concepts Claude already knows, though at ~208 lines some sections (Common Mistakes table, Pressure Tests) could be tightened without losing substance. | 4 / 5 |
Actionability | Provides concrete copy-paste templates (claim ledger table with column specs, minimal execution format field blocks, provenance line format) and explicit verdict criteria, but lacks a fully worked filled example of a completed ledger row or verdict for a real claim. | 4 / 5 |
Workflow Clarity | Clear sequence (Trigger → Claim Ledger → 八问 checklist → three-axis Verdict → Provenance) with explicit validation gates (最低证据面 caps that downgrade verdicts) and Pressure Test '合格输出必须' criteria serving as feedback checkpoints. | 5 / 5 |
Progressive Disclosure | Well-organized into clearly headed sections with no nested references and no bundle files to navigate; over the 50-line simple-skill threshold, and some detailed material (execution format, Pressure Tests) could live in reference files, so it stops short of a 5. | 4 / 5 |
Total | 17 / 20 Passed |