Content
78%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, actionable instruction-only skill body with concrete templates, tables, and properly disclosed one-level-deep references that all resolve to real files. Its main weakness is conciseness: the tone/format rules, closing recommendation, and anti-patterns sections overlap, creating redundancy that could be trimmed.
Suggestions
Collapse the duplication between Tone and Format Rules, the Closing Recommendation section, and Anti-Patterns To Reject — state each rule once and reference it, rather than restating 'no manufactured problems', 'no vague reassurance', and 'no structured-report output' across three sections.
Trim the 'This matters because' bullet list under the Stop Directive; a single line ('Reflection is a pause, not a side-quest — continue only after reassessing') conveys the same point without explaining reasons Claude can infer.
Consider condensing the bias-check table cues if they duplicate content already in references/cognitive_bias_canon.md, keeping only what is needed to act inline and deferring the fuller canon to the reference.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient instructional content, but it includes notable duplication — the closing recommendation is stated in the intro, restated in its own section, and re-listed under Anti-Patterns, and Tone/Format Rules overlaps heavily with Anti-Patterns — plus explanatory padding like the 'This matters because' bullet list that Claude could infer, so it could be tightened. | 3 / 5 |
Actionability | Highly actionable for an instruction-only skill: the 5-dimension framework gives specific bullet questions, the bias table pairs recognition cues with corrective moves, and the closing recommendation table supplies exact output templates ('Continue. {specific reasoning for why}.') plus an error-handling table with concrete behaviors. | 5 / 5 |
Workflow Clarity | A clear sequence is conveyed — detect trigger, halt the thread, optionally ask Q1, re-read the full conversation, run the 5 numbered dimensions, end with a mandatory directional recommendation — with explicit checkpoints (Stop Directive, 'Stop condition: One question max', mandatory closing) and a validator script; the flow is distributed across sections rather than one explicit ordered checklist, a minor gap. | 4 / 5 |
Progressive Disclosure | SKILL.md acts as a clear overview with well-signaled, one-level-deep references to three real reference files (cognitive_bias_canon, honest_output_discipline, conversation_reflection_practice) linked inline and gathered in a References section, plus a Tooling table listing three real scripts — easy to navigate with no nested references. | 5 / 5 |
Total | 17 / 20 Passed |