Conciseness

The core content is dense and skill-specific (the representation/inference contracts section, the Harmony 2 transpose-bug note, '.raw' vs 'layers["counts"]' semantics), but several sections pad or duplicate: 'Understanding AnnData Structure' explains basics Claude already knows and which the description routes to the anndata skill; 'Bundled Resources' re-lists the scripts already tabulated in 'Script Toolkit'; and 'Tips for Effective Analysis' repeats pitfalls (e.g., 'Save checkpoints' ≈ 'Save intermediate results'). This matches the 'mostly efficient but includes some unnecessary explanation or could be tightened' anchor, not 2, because the essential material is non-inferable and tightly written.

Actionability

Fully executable throughout: a complete script table with exact typical invocations, copy-paste one-shot and step-by-step command chains, a '--config params.json' reproducibility pattern, and real bundled assets to copy ('cp assets/analysis_template.py my_analysis.py'). Common cases are covered by ready-to-run CLI scripts, matching the copy-paste-ready 5 anchor.

Workflow Clarity

The seven-step workflow is clearly sequenced, and validation/feedback checkpoints are written explicitly rather than implied: 'inspect mitochondrial fraction and counts before choosing thresholds rather than copying defaults', 'inspect results/markers/*.csv, decide labels, write a mapping JSON, then: annotate.py', 'Check PCA variance ratio', and 'Try multiple clustering resolutions'. Error-recovery guidance is also present ('Long workflows can fail partway through' → save intermediate results), matching the explicit-validation-and-feedback-loops 5 anchor.

Progressive Disclosure

All referenced files exist and SKILL.md links are one level deep and clearly signaled ('Read references/r_interop.md for agent-run installation and conversion instructions'). Minor organization gaps remain: the script toolkit is documented twice (table in 'Script Toolkit' plus a file listing in 'Bundled Resources'), two overlapping workflow references (analysis_workflow.md vs standard_workflow.md) are not differentiated, and reference files cross-link to each other (e.g., to upstream-review.md), matching the 'minor organization gaps' 4 anchor rather than 5.

Specificity

Lists ten-plus concrete actions ('QC, normalization, HVG selection, PCA/UMAP/t-SNE, clustering, exploratory marker ranking, pseudobulk preparation, visualization, and Seurat or SingleCellExperiment RDS conversion to h5ad') with comprehensive coverage of the Scanpy workflow. No meaningful action is left unnamed, so this matches the comprehensive-coverage anchor rather than the 'minor gaps' anchor at 4.

Completeness

The 'what' is explicit and concrete, and a 'when' clause exists ('Applies to established exploratory scRNA-seq workflows with explicit count and expression provenance'), so the missing-'Use when' cap at 3 does not apply. However the 'when' is phrased as a scope constraint rather than concrete user-side trigger phrases (e.g., 'Use when analyzing .h5ad files or 10X output'), matching the 4 anchor ('when' could be more explicit or specific) rather than 5.

Trigger Term Quality

Good natural keyword coverage including 'single-cell RNA-seq', 'scRNA-seq', 'Scanpy', 'clustering', 'pseudobulk', 'Seurat', 'SingleCellExperiment', 'RDS', and 'h5ad'. A few natural terms users would say are missing — 'batch correction'/'integration', 'cell type annotation', and 'trajectory'/'pseudotime' — all of which the skill body actually covers, so it fits the 4 anchor ('a few natural terms missing') rather than the comprehensive 5.

Distinctiveness Conflict Risk

Clear niche (Scanpy-based single-cell RNA-seq analysis in Python) with distinct triggers, and the routing clause 'complementary skills cover scvi-tools models and AnnData format details' explicitly deflects the closest sibling skills. Overlap risk is actively managed, matching the minimal-conflict 5 anchor rather than the minor-overlap 4 anchor.