Conciseness

The high-value core (QC gating rationale, ToolUniverse catalog with live-verified parameter gotchas, RULE ZERO, analysis conventions) is dense and lean, but several sections re-teach material Claude already knows: the scipy 'Statistical Tests (Quick Reference)' (ttest_ind, f_oneway, pearsonr, multipletests), the 'Scanpy vs Seurat Equivalents' table, and the boilerplate 'Complete Pipeline' block. This matches 'mostly efficient but includes some unnecessary explanation' rather than score 2, since the padded material is a minority of the ~435 lines and the rest is reference-grade rather than prose-padded.

Actionability

Code is copy-paste ready and covers the common cases: data loading with orientation check, QC metric computation, the full scanpy pipeline, Harmony batch correction, and a concrete `tu run run_deseq2_analysis '{...}'` command. Referenced script invocations are real and verified — `scripts/scrna_qc.py` exists and exposes both `--doublets` and the `--install-plan` flag, with the honest-execution fallback ('do NOT fabricate metrics — print the install plan and stop') making the failure path executable too.

Workflow Clarity

A clear sequence exists (RULE ZERO pre-flight check → workflow decision tree → 'QC ... do this BEFORE downstream analysis' → DE decision tree), with real checkpoints: 'Always visualize distributions first', 'never report cutoffs you did not actually run', per-step removal reporting, the QC interpretation table, and a troubleshooting error-recovery table. It falls short of 5 because the 'Complete Pipeline (Quick Reference)' reuses hardcoded cutoffs (`pct_counts_mt < 20`, `min_genes=200`) that the QC section explicitly warns are 'a starting point only' — an internal tension that could lead Claude to skip the MAD-based gating the skill mandates.

Progressive Disclosure

The 8 reference files are well documented with one-line descriptions in 'Reference Documentation' and are one level deep, but `analysis_patterns.md` is referenced three times — including as the guide for 'Pattern 1' (per-cell-type DE, described as the most common pattern) and 'Pattern 2' (correlation) — and does not exist anywhere in the bundle. A primary navigation path pointing to a missing file is a structural failure beyond the 'minor organization gaps' of score 4, landing at 'some structure but could be better organized'.

Specificity

The description enumerates concrete actions at metric-level detail: "h5ad data loading", "QC gating (n_genes_by_counts, total_counts, mitochondrial percent / pct_counts_mt, pct_counts_ribo, doublet detection with Scrublet/scDblFinder, ambient RNA / SoupX awareness, empty-droplet filtering, MAD-based thresholds)", plus normalization, "dimensionality reduction (PCA, UMAP, t-SNE)", "clustering (Leiden, Louvain)", marker identification, cell-type annotation, and "pseudotime/trajectory analysis". Coverage is comprehensive with no noticeable gaps, matching the anchor 'Lists multiple specific concrete actions; comprehensive coverage' rather than the score-4 anchor (minor gaps).

Completeness

The 'what' is explicit and detailed (loading, QC gating, normalization, reduction, clustering, annotation, pseudotime) and the 'when' is explicit with concrete triggers: "Use for any scRNA-seq workflow, including deciding which cells to filter, flag, or investigate before downstream analysis". This matches the anchor 'Clearly and explicitly answers both what AND when with concrete trigger phrases'; the score-4 anchor ('when' could be more explicit) does not apply since the trigger clause is specific and actionable.

Trigger Term Quality

Natural user vocabulary is comprehensive with synonyms and file formats: "Single-cell RNA-seq" and "scRNA-seq", "h5ad", "quality control", "doublet", "clustering", "marker gene", "cell-type annotation", "PCA, UMAP", "trajectory", tool names (scanpy, anndata, Scrublet, SoupX), and action phrases like "deciding which cells to filter, flag, or investigate". Matches the anchor 'Comprehensive coverage of natural terms including synonyms and file extensions'; only trivially misses terms like '10X' or 'count matrix'.

Distinctiveness Conflict Risk

The niche is unambiguous — single-cell RNA-seq — with distinct triggers (scRNA-seq, h5ad, doublets, cell-type annotation, pseudotime) that do not overlap with adjacent bulk-RNA-seq, variant-analysis, or enrichment skills. It consistently self-scopes as "Single-cell"/"scRNA-seq" throughout, so a generic 'RNA-seq' request would not naturally match this description over the bulk DESeq2 sibling.