Content
57%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 body is extremely actionable — exact commands, expected output formats, worked examples, and sanity-check targets — but it is undermined by heavy redundancy (the same PhyKIT column and batch-tool guidance repeated three times) and two outright contradictions about which alignment file treeness_over_rcv takes and whether the batch tool supports it. The scogs playbook should live in a reference file rather than inflating SKILL.md to ~650 lines.
Suggestions
Resolve the two contradictions: pick one alignment input for treeness_over_rcv/rcv (the trap-3 section with measured medians 0.2683 vs 0.3050 supports the untrimmed .faa.mafft) and delete the contradicting 'use the right input file' section; likewise reconcile whether phykit_batch_analysis supports treeness_over_rcv.
Merge the three saturation-column sections and the three batch-tool advocacy sections into one canonical 'PhyKIT column conventions' section, cutting roughly a third of the body.
Move the detailed BUSCO scogs question-answering playbook into a reference file (e.g., references/scogs_playbook.md) and reference it with a one-line description; add references/troubleshooting.md to the References list or remove it from the bundle.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~650-line body repeats the same guidance in multiple sections: the saturation two-column convention is explained three times (trap 1, 'PhyKIT column conventions', the column-position cheat sheet), and batch-tool-over-manual-loop advocacy appears in three separate sections. Much of the trap documentation is genuinely non-obvious and earns its place, but whole duplicated sections plus filler like the contentless 'LOOK UP, DON'T GUESS' stub make this noticeably padded rather than merely in need of tightening. | 2 / 5 |
Actionability | Guidance is fully executable and copy-paste ready: exact invocations with all arguments ('tu run phykit_batch_analysis {"operation":"batch",...}'), documented output blocks with line-by-line parse guidance ('# MWU animals_vs_fungi: U=... p=...'), worked examples contrasting wrong vs. right code, and phrasing-to-answer mapping tables. Common cases are concretely covered. | 5 / 5 |
Workflow Clarity | RULE ZERO gives a clear entry decision (check pre-computed results first) and sanity-target checkpoints exist ('median saturation ~0.4-0.7... values an order of magnitude off these mean wrong column'), which is good. But two decision points are internally contradictory: one section (and trap 3, with measured numbers) says treeness_over_rcv/rcv take the UNTRIMMED .faa.mafft while a later section says use the TRIMMED .faa.mafft.clipkit; likewise the batch tool's supported functions include treeness_over_rcv in one section and are said to have 'no matching function' for it in another. A reader following the document sequentially receives incoherent instructions at exactly the fragile decision points. | 3 / 5 |
Progressive Disclosure | Bundle structure is real and referenced paths check out (scripts/scogs_paired_compare.py, scogs_phykit_pipeline.py, busco_target_orthologs.py, phykit_batch.py, tree_statistics.py all exist, as do references/sequence_alignment.md, tree_building.md, parsimony_analysis.md). But the four reference files are only listed as a bare path dump at the bottom with no per-topic signaling, references/troubleshooting.md is never mentioned anywhere, and ~400 lines of scogs-specific question-answering guidance that belongs in a reference file are inlined directly in SKILL.md. | 3 / 5 |
Total | 13 / 20 Passed |