Content
71%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 content is highly actionable with concrete, executable CLI commands and a clear validated workflow sequence, and it appropriately avoids explaining concepts Claude already knows. Its weaknesses are redundancy across three overlapping capability/workflow/methodology sections and a monolithic single-file structure with no progressive disclosure of detailed material into reference files.
Suggestions
Consolidate "Core Capabilities", "Workflow", and "Algorithm / Methodology" into a single sequenced section to eliminate the triple restatement of training, export, and diagnostics content (and fix the duplicated item 5 in Core Capabilities).
Move the Algorithm/Methodology details and per-parameter QC specifics into a reference file (e.g. references/methodology.md) linked one level deep from SKILL.md.
Add a short error-recovery step after input validation (e.g. what message to surface and what to suggest when a processed-like matrix is rejected) to close the workflow feedback loop.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient with well-chosen tables and commands, but the same material (training, latent export, batch diagnostics) is restated across "Core Capabilities", "Workflow", and "Algorithm / Methodology", and "Core Capabilities" contains a duplicated list number (two items numbered 5). It could be meaningfully tightened without losing information, matching the 3 anchor rather than the 4 anchor's "minor instances". | 3 / 5 |
Actionability | The CLI Reference provides fully executable, copy-paste-ready commands covering the common cases (standard, batch-key, scANVI with labels, 10x directory, demo, runner invocation), and the Output Structure and Demo sections give concrete expected artifacts. This matches the 5 anchor's coverage of common cases with complete commands. | 5 / 5 |
Workflow Clarity | The Workflow section gives a clear Validate → Filter → Train → Project → Generate sequence with an explicit validation checkpoint ("reject processed-like matrices") at the start. It falls short of 5 because there is no error-recovery feedback loop describing what to do when validation fails or training inputs are malformed. | 4 / 5 |
Progressive Disclosure | The single SKILL.md is well-sectioned but monolithic: no bundle files exist (references/, scripts/, assets/ are absent), and sizable content such as the Algorithm/Methodology details and CLI reference is inlined where a one-level-deep reference file would be appropriate. This matches the 3 anchor of content that should be separate being inline, rather than the 4 anchor's "most content appropriately placed". | 3 / 5 |
Total | 15 / 20 Passed |