Content
86%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 highly actionable with executable commands and examples, well-structured progressive disclosure into real bundle files, and a clear workflow with an embedded verification loop. The main improvement area is conciseness, where a few redundant phrases could be trimmed.
Suggestions
Tighten redundant phrasing such as 'The script automatically detects the file format and loads it appropriately' to a terser statement.
Promote the before/after visualization review into an explicit numbered 'Validate' step in the workflow so the feedback loop is a first-class checkpoint.
Example 3 ends after computing per-subset masks; add the combine-and-write step so all three examples are complete and copy-paste runnable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient with domain-specific advice (mt- vs MT- prefixes, neuron/cardiomyocyte MT content) that genuinely earns its place, but contains minor padded phrases like 'automatically detects the file format and loads it appropriately'. Not a 5 because a few sentences could be trimmed without losing clarity. | 4 / 5 |
Actionability | Provides copy-paste bash commands, a full parameter list with a --help pointer, three executable Python examples, and documented function signatures covering the common cases. Not a 4 because the guidance is fully executable rather than having notable gaps (Example 3's subset logic is inherently illustrative). | 5 / 5 |
Workflow Clarity | Four clearly sequenced workflow steps with a before/after-plot verification and an 'Iterate if needed' feedback loop for the destructive/batch filtering operation, so it is not capped at 3. Not a 5 because the validation checkpoint lives in Best Practices rather than as an explicit step in the numbered workflow. | 4 / 5 |
Progressive Disclosure | SKILL.md is a clear overview split into Approach 1/2, pointing one level deep to scripts (qc_analysis.py, qc_core.py, qc_plotting.py) and references/scverse_qc_guidelines.md, all of which exist and are well-signaled with explicit 'Load this reference when...' guidance. Not a 4 because structure and navigation are clean with no nesting or burial. | 5 / 5 |
Total | 18 / 20 Passed |