Content
63%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 TCGA workflow guide with a clear phase pipeline, validation guidance, and executable examples. Its weaknesses are token efficiency (duplicated tool tables and textbook interpretation guidance) and the absence of progressive disclosure — reference-grade material is inlined rather than split into bundle files — plus an internal contradiction about whether GDC_get_survival returns a p-value.
Suggestions
Resolve the p-value contradiction: remove "+ pValue" from the GDC_get_survival row in the Tool Quick Reference and the "+ p-value" claim in Workflow 1, since Phase 5 correctly states the tool returns no log-rank p-value.
Trim general-knowledge interpretation guidance (log-rank p<0.05 meaning, HR>1 semantics, driver/passenger thresholds) and keep only TCGA-specific caveats (not treatment-stratified, n<20 unreliable, DUP counts are sample frequencies), or move this reasoning material to a references/ file.
Move the Tool Quick Reference table and the Programmatic Access (GDC API) section into a references/ file linked from the relevant phases, since they duplicate or extend the per-phase details and inflate the SKILL.md body.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient — tool params, return shapes, and caveats are dense and non-padded — but two sections explain knowledge Claude already has: the Interpretation Guidance block ("A log-rank p-value < 0.05 suggests...", "Hazard ratio (HR) > 1 indicates worse prognosis") is textbook statistics, and the Tool Quick Reference table largely restates the per-phase tool details. Matches the level-3 anchor (some unnecessary explanation, could be tightened) rather than 4, where over-explanation would be only minor. | 3 / 5 |
Actionability | Concrete, copy-paste-ready code per phase with real values (chr7 EGFR coordinates, NCIT:C4017, verified survival example), exact parameter lists and return schemas, plus an executable GDC API pagination loop. Falls short of 5 due to an internal contradiction: the Quick Reference table and Workflow 1 claim GDC_get_survival returns a "pValue" / "p-value" while Phase 5 explicitly states it returns no log-rank p-value — a gap that could mislead execution. | 4 / 5 |
Workflow Clarity | Clear six-phase pipeline diagram, "Build your cohort FIRST" ordering rule, and explicit validation steps ("LOOK UP DON'T GUESS: never assume TCGA project IDs... use GDC_list_projects to confirm"; Workflow 2 confirms the project exists before deeper queries) plus a Synthesis Questions checklist. Not 5 because there are no error-recovery feedback loops (e.g., what to do on empty results or failed calls) during batch pagination; above 3 because checkpoints are explicit rather than implicit. | 4 / 5 |
Progressive Disclosure | Good section structure with clear headers, but the skill is a ~325-line monolith with no bundle files: the Tool Quick Reference, Reasoning Framework, and Programmatic Access sections are exactly the bulk material the level-3 anchor describes as "content that should be separate is inline". Not 2 because navigation and sectioning are genuinely clear; not 4 because substantial reference material remains inlined in SKILL.md. | 3 / 5 |
Total | 14 / 20 Passed |