Content
78%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, appropriately lean skill body with excellent progressive disclosure — every referenced bundle file exists and is clearly signposted. The main gaps are the absence of a consolidated numbered workflow and of a single concrete example command in SKILL.md itself.
Suggestions
Consolidate the operating sequence (uv setup, license notification, ask species, map identifiers, run command, post-process output) into a single numbered workflow so the order of operations is explicit rather than implied across sections.
Include one copy-paste-ready example invocation with a real subcommand (e.g., a minimal 'map' call) in SKILL.md so the CLI's usage pattern is concrete before the reader jumps to a reference file.
Trim the repetitive MUST-emphasis in Core Rule 1 to a single unambiguous statement (e.g., 'Never guess the species — if unstated, ask for the NCBI Taxon ID first') to save tokens without losing the behavioral constraint.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean with no explanations of concepts Claude already knows, but Rule 1's layered emphasis ("MANDATORY", "You MUST NOT guess", "you MUST stop and ask", "you MUST still ask — do not default to Human") repeats the same instruction several times and could be tightened. It is above level 3 (only one padded passage) but not level 5's every-token-earns-its-place. | 4 / 5 |
Actionability | The body provides a concrete executable template ("uv run scripts/string_cli.py <command> [options] --output /tmp/out.tsv"), exact file paths, and pointed navigation to reference files that contain full command examples. It falls short of level 5 because no fully concrete worked invocation (with a real subcommand and arguments) appears in SKILL.md itself — the command remains a placeholder. | 4 / 5 |
Workflow Clarity | A clear sequence is discernible (prerequisites, license check, mandatory species checkpoint, map identifiers first, run with --output, post-process via jq/grep) with one hard validation checkpoint (the species question). It is not level 5 because the steps are distributed across sections rather than consolidated into an ordered workflow, and there is no explicit output-verification step; the destructive/batch cap does not apply since these are non-destructive read-only queries. | 4 / 5 |
Progressive Disclosure | The body is a clear overview with four well-signaled, one-level-deep references (mapping.md, interactions.md, enrichment.md, valuesranks.md), each annotated with when to read it, plus explicit guidance to start with the most appropriate file. All referenced files exist with real content, and scripts/string_cli.py is present, so navigation is complete and easy. | 5 / 5 |
Total | 17 / 20 Passed |