Content
82%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 a strong, execution-focused reference: every workflow comes with a runnable command and realistic arguments, prerequisites include a conditional license-notification procedure, and anti-patterns add genuine expert judgment. Weaker spots are minor: some trimmed-able asides, no explicit output-error validation step, and an inlined track-version table plus an orphaned citation.bib that slightly blur the file structure.
Suggestions
Add a brief validation step after each query, e.g. 'Check the output JSON for an errorMessage field before parsing; UCSC returns 400 errors inline for invalid track names', which would close the workflow_clarity gap.
Move the hg38/hg19 collection/track tables (roughly lines 87-105) into a reference file (e.g. references/collections.md) and link to it, keeping the SKILL.md body as a lean overview.
Either reference references/citation.bib from the body (e.g., a one-line 'Cite as' note) or remove it from the bundle so every bundled file is discoverable from SKILL.md.
Trim explanatory asides like 'which is a strong signal for pathogenicity' that restate knowledge Claude already has.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is operationally dense with little concept tutoring, but contains minor padding that could be trimmed, e.g. "which is a strong signal for pathogenicity" and "matching the UCSC Genome Browser's default comparative genomics tracks", and inlines version-bearing track lists (jaspar2024/jaspar2026) outside any deprecated/old-patterns section. Not 3 because there is no unnecessary explanation of basics; not 5 because every token does not quite earn its place. | 4 / 5 |
Actionability | Every task section provides copy-paste-ready, fully executable commands with real coordinates, flags, and --output paths (e.g. "uv run scripts/get_conservation.py --coordinates \"chr1:215867804\" ... --output /tmp/cons_output.json"), covering the common cases: single bases, regions, collections, acceleration analysis, TFBS filtering, and track discovery. This matches the 5 anchor; a 4 would have minor gaps in coverage. | 5 / 5 |
Workflow Clarity | Prerequisites are sequenced with explicit conditional logic (check for the license file, notify, then create it), each task maps to a single command, and the CAUTION note preempts a known 400 failure mode. Not 5 because there is no explicit post-query validation step (e.g., checking the output JSON for API errors before parsing); not 3 because sequence and most checkpoints are present and operations are read-only, so the destructive/batch cap does not apply. | 4 / 5 |
Progressive Disclosure | The three utility scripts are clearly signaled with one-line purposes and are one level deep, and sections are well organized. Not 5 because the ~30-line hg38/hg19 collection tables are inlined rather than split into a reference file, and references/citation.bib exists in the bundle but is never referenced or navigable from the body — minor organization gaps matching the 4 anchor. | 4 / 5 |
Total | 17 / 20 Passed |