Content
45%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's core analysis guidance is genuinely actionable, with code examples that match the packaged script's API and CLI exactly, and the complete workflow includes real validation checkpoints. However, the body is massively over-inflated with template boilerplate, triplicated troubleshooting, and textbook CRISPR biology, while an existing reference file goes unmentioned — the monolithic structure wastes context and buries the useful content.
Suggestions
Cut the template boilerplate sections ("When to Use", "Key Features", "Dependencies", "Example Usage", "Implementation Details", "Output Requirements", "Input Validation", "Response Template", etc.) that restate the description or generic process rules, keeping the body to the API examples, parameter table, and complete workflow.
Move the "Common Patterns", "Common Pitfalls", "Troubleshooting", and "Quality Checklist" sections into one-level-deep files under references/ and link them clearly, replacing the false "(No reference files currently available for this skill)" claim with a link to the existing references/runtime_checklist.md.
Fix the broken executable guidance: remove the bogus "cd 20260318/scientific-skills/..." path, and correct the "Step 1: Run QC" CLI example so it reflects what scripts/main.py actually outputs when run without --control/--treatment.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~1,180-line body is severely verbose and heavily padded: template boilerplate ("When to Use", "Key Features", "Implementation Details", "Workflow") restates the description verbatim multiple times; troubleshooting content is triplicated ("Common Issues and Solutions" in four capability sections, plus a global "Troubleshooting" and "Common Pitfalls" section); and it extensively explains CRISPR screen biology Claude already knows (MOI, T0/T14 design, synthetic lethality, DepMap). This matches anchor 1; it exceeds anchor 2's "several unnecessary explanations" because the padding is pervasive throughout, not limited to a few sections. | 1 / 5 |
Actionability | The Python examples call the real API in scripts/main.py exactly (qc_metrics(), calculate_lfc(control, treatment), rra_analysis(lfc, fdr_threshold), identify_hits(results, fdr_threshold, lfc_threshold)) and the CLI examples match the script's argparse flags, so guidance is mostly executable and copy-paste ready. Not 5 due to minor gaps: the bogus "cd 20260318/scientific-skills/..." path, and the "Step 1: Run QC" command claims a qc_results output the script never writes when run without --control/--treatment. | 4 / 5 |
Workflow Clarity | The "Complete Workflow Example" provides a concrete QC → LFC → RRA → hit-calling sequence with a real validation checkpoint (qc_pass = all(gini < 0.4 ...) with a warning) plus a pre-run py_compile check and phase-by-phase Quality Checklists, matching anchor 4. Not 5 because the QC gate only warns and proceeds rather than stopping (no validate → fix → re-validate feedback loop), and the generic "Workflow" section is filler rather than a domain sequence. | 4 / 5 |
Progressive Disclosure | Hundreds of lines of patterns, pitfalls, checklists, and troubleshooting that clearly belong in separate reference files are inlined into a monolithic SKILL.md, matching anchor 2. Not 3 because the one existing reference (references/runtime_checklist.md) is not merely poorly signaled but actively denied ("(No reference files currently available for this skill)") and never linked, leaving no functioning reference navigation at all. | 2 / 5 |
Total | 11 / 20 Passed |