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, highly actionable reflection skill with a clear sequenced workflow and concrete output templates. Its main defect is that the body leans on three reference files and three scripts that are not present in the bundle, so the progressive-disclosure links are dangling; secondary redundancy between the tone rules and anti-patterns could be trimmed.
Suggestions
Ship the referenced bundle files (references/cognitive_bias_canon.md, references/honest_output_discipline.md, references/conversation_reflection_practice.md and the three scripts/*.py), or remove the links and inline only what the skill truly depends on — currently every referenced path is a dead link.
De-duplicate the Tone and Format Rules and the Anti-Patterns To Reject section, which restate the same 'no vague reassurance' / 'no manufactured problems' / 'no structured-report output' prohibitions.
Add an explicit validation checkpoint before the closing recommendation (e.g., 'confirm the reassessment names the original goal and at least one drift/assumption/bias with cited turns') to lift workflow clarity.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient actionable lists, but there is real redundancy: the Anti-Patterns section restates the Tone and Format Rules, the Stop Directive belabors the pause point, and the full Q1 quoted block is longer than needed. Not below a 4 because the core is tight, not a 5 because of the duplication. | 4 / 5 |
Actionability | Highly concrete for an instruction-only skill: exact output templates ('Continue. {specific reasoning for why}.'), evidence-anchoring prompts ('At turn 3 the goal was X; by turn 12...'), a bias table with recognition cues, and an error-handling table — copy-paste ready guidance covering the common cases. | 5 / 5 |
Workflow Clarity | Clear sequence (trigger → stop → optional Q1 → sequenced 5-dimension checklist → flowing prose → mandatory closing) with an error-handling feedback table; it stays at 4 rather than 5 because there is no explicit self-validation checkpoint confirming the reflection addressed the actual drift. | 4 / 5 |
Progressive Disclosure | The structure is good — overview and framework inline, deeper material signaled as one-level-deep markdown links — but every referenced file is missing: references/cognitive_bias_canon.md, references/honest_output_discipline.md, references/conversation_reflection_practice.md and all three scripts/*.py do not exist in the bundle, so the disclosure promises are broken and navigation fails. | 3 / 5 |
Total | 16 / 20 Passed |