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 thorough, highly actionable body with executable code, clear sequenced workflows, and useful troubleshooting, weakened by monolithic inlining: it duplicates ~2,000 lines of standalone scripts in the bundle without a single pointer to them, and lacks inter-step validation checkpoints in the pipelines.
Suggestions
Reference the existing scripts/ bundle from the body (e.g., "Full CLI implementations: scripts/parse_vcf.py, scripts/calculate_tmb.py, scripts/nmf_metagenes.py, scripts/ddr_network.py") and move the bulk implementations out of SKILL.md, keeping only the Quick Start and one compact example per capability.
Add explicit validation checkpoints between pipeline steps (e.g., verify tumor.cnr/reference.cnn exist and are non-empty after each CNVkit stage before proceeding; check FilterMutectCalls output before annotation), matching the error-recovery pattern already used for Mutect2.
Delete the Overview paragraph that restates the frontmatter description, and make Workflow 4's placeholder constants (PATHWAY_GENES, DRIVER_GENE, GROUP) concrete with a worked example so the snippet is runnable as-is.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly task-specific code that earns its place, but at ~600 lines it is a monolith: the Overview paragraph restates the frontmatter description nearly verbatim, and full implementations (parse_vcf, calculate_tmb, NMF, DDR network) are inlined that duplicate the scripts/ bundle. Mostly efficient but could be tightened, which fits anchor 3 rather than the minor-trimming of 4. | 3 / 5 |
Actionability | Concrete, largely executable code throughout — parse_vcf with quality filters, the Mutect2/CNVkit subprocess wrappers with error handling, the NMF metagene and TMB functions. Minor gaps keep it below fully copy-paste-ready: Workflow 4 uses undefined placeholders (PATHWAY_GENES, DRIVER_GENE, clin.GROUP), estimate_purity_ploidy is self-admittedly a "Rough heuristic", and external tools (GATK, SnpEff, CNVkit) are assumed installed with only a comment. | 4 / 5 |
Workflow Clarity | Workflows 1–4 are clearly numbered and sequenced, subprocess calls surface errors (raise RuntimeError with Mutect2 stderr, check=True), and the Troubleshooting section gives problem/solution recovery guidance. Not 5 because there are no explicit validate-then-proceed checkpoints — e.g., the CNVkit pipeline never verifies intermediate outputs (tumor.cnr, reference.cnn) before the next step — placing it at clear sequence with minor validation gaps. | 4 / 5 |
Progressive Disclosure | Section structure is good (Quick Start, Core Capabilities, Typical Workflows, Best Practices, Troubleshooting, Resources), but ~500 lines of implementation code are inlined in SKILL.md while a scripts/ bundle (calculate_tmb.py, ddr_network.py, nmf_metagenes.py, parse_vcf.py — ~2,000 lines covering the same capabilities) exists and is never referenced anywhere in the body. Scored against the actual bundle: references are absent, not merely unclear, which rules out 4; the genuine in-file structure rules out 2. | 3 / 5 |
Total | 14 / 20 Passed |