Content
48%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 skill delivers concrete, well-sequenced decision flows and specific expert-vs-novice phrasings for clinical cognitive-bias correction, but nearly half the body is educational padding (bias-development timelines, metaphors, shibboleths) that costs tokens without adding instructions. Splitting the anti-pattern catalog into a reference file and cutting the padding would materially improve it.
Suggestions
Remove or drastically cut the 'Timeline:' paragraphs, 'Navigation Metaphor', 'Trap Metaphor', and 'Core Shibboleths' sections — they describe how human expertise develops, which does not guide execution.
Move the six detailed anti-pattern entries (novice/expert/timeline each) into a references/ file (e.g. ANTI-PATTERNS.md), keeping one-line summaries and the expert questions inline in SKILL.md.
Add a brief error-recovery loop to the diagnostic flow: what to do when the bias check flags a problem (e.g. regenerate the differential from scratch and re-examine disconfirming evidence before proceeding).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Several padded sections explain material Claude either already knows or does not need to perform the skill: a 'Timeline:' paragraph for each of the six biases describing how expertise develops over years, a 'Navigation Metaphor', a 'Trap Metaphor', 'Core Shibboleths', and a References section restating the book's thesis. This matches 'noticeably verbose; several unnecessary explanations or padded sections' (2) rather than the mostly-efficient level of 3. | 2 / 5 |
Actionability | The three ASCII decision flows are genuinely concrete — specific questions to ask ('What else could this be?', 'Does this explain ALL findings?', 'How is this patient NOT like that case?') and explicit branches — and the novice/expert contrasts give usable phrasings. However, roughly 40% of the body (timelines, metaphors, shibboleths, references) is descriptive rather than instructive, so guidance is incomplete in coverage, fitting score 3 rather than the 'mostly executable with minor gaps' of 4. | 3 / 5 |
Workflow Clarity | The diagnostic flow is a clearly sequenced decision tree with a mandatory checkpoint ('[BIAS CHECK - Always perform]'), the error-analysis flow separates knowledge gaps from thinking errors before branching, and the communication flow ends with a comprehension-verification step ('What is your understanding of the risks/benefits?'). Checkpoints are present and explicit; it falls short of 5 only because there are no error-recovery/feedback loops (e.g. what to do when the bias check flags a problem mid-workup) beyond the branch instructions. | 4 / 5 |
Progressive Disclosure | The body is well-sectioned with clear headers, but it is a single ~200-line monolith with no bundle files (no references/, scripts/, or assets/ exist) — the six detailed anti-pattern entries and mental-models material are inline content that would fit a one-level-deep references file. This matches 'some structure but could be better organized; content that should be separate is inline' (3); the under-50-line simple-skill exception does not apply. | 3 / 5 |
Total | 12 / 20 Passed |