Content
65%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 exceptionally actionable — every trap comes with an exact command, output format, or decision table — but it is a ~650-line monolith that repeats its key rules multiple times and even contradicts itself on which alignment to use for treeness_over_rcv. Consolidating the duplicated guidance into single canonical sections and moving detail into the existing (partially orphaned) reference files would substantially improve it.
Suggestions
Consolidate the three separate saturation-column discussions (trap 1, 'PhyKIT column conventions', 'column-position cheat sheet') and the duplicated gap-percentage and batch-vs-loop warnings into one canonical section each.
Resolve the direct contradiction on the `treeness_over_rcv` alignment input: 'Which alignment goes with which metric' says untrimmed `.faa.mafft` while 'Treeness/RCV: use the right input file' says trimmed `.faa.mafft.clipkit` — state one verified rule with the numbers that justify it.
Move the long worked examples and per-metric tables into the existing reference files (`sequence_alignment.md`, `tree_building.md`, `parsimony_analysis.md`) with inline 'See X for Y' pointers, and link the currently orphaned `references/troubleshooting.md`.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is ~650 lines with material duplication: the saturation-column convention is explained three times (trap 1, 'PhyKIT column conventions', 'column-position cheat sheet'), the gap-percentage definition twice, and the batch-vs-loop warning four times, alongside concept explanations Claude already knows (what an internal branch is). Most of the trap data is genuinely non-obvious and earns its place, keeping this above the 'noticeably padded' anchor but below 'efficient'. | 3 / 5 |
Actionability | Guidance is fully executable throughout: exact `tu run phykit_batch_analysis '{...}'` payloads, bundled script invocations with flags and expected output blocks (`# SUMMARY group=animals: n=... median=...`), per-metric column tables, and a phrasing-to-statistic mapping table. Copy-paste ready commands cover the common cases. | 5 / 5 |
Workflow Clarity | RULE ZERO gives an explicit check-first sequence and there are validation checkpoints (sanity targets, 'count files first', wrong-column tells), so the batch-operation cap is not triggered. However, the same rules are restated across scattered sections and directly contradict each other — 'Which alignment goes with which metric' mandates the untrimmed `.faa.mafft` for `treeness_over_rcv`/`rcv` while 'Treeness/RCV: use the right input file' mandates the trimmed `.faa.mafft.clipkit` — leaving the reader without a single canonical procedure per metric. | 3 / 5 |
Progressive Disclosure | Reference and script files exist and the paths cited in the body are real, but the References section is a single flat line at the very bottom rather than inline signposts, the bulk of per-metric detail is inlined in SKILL.md instead of split across the available reference files, and `references/troubleshooting.md` and `scripts/format_alignment.py` are never referenced anywhere. | 3 / 5 |
Total | 14 / 20 Passed |