Content
68%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 highly actionable body: the single complete runnable example, explicit contrast/design syntax, and a real, well-partitioned bundle make it easy to execute correctly. Weaknesses are redundancy (Key Features vs. Implementation Details vs. the description), pinned time-sensitive versions, the absence of explicit validation checkpoints for a batch analysis, and hedged, non-linked pointers to the bundle files.
Suggestions
Delete the 'Key Features' section (it duplicates the description and Implementation Details), trim the volcano-plot block to a few lines or move it to references/workflow_guide.md, and relocate the pinned dependency versions (pandas >= 1.4.3, etc.) to references/api_reference.md to improve conciseness.
Add explicit validation checkpoints before fitting in the example script — e.g., assert counts are non-negative integers, confirm all design variables exist in metadata columns, and verify sample alignment — with a brief fix-and-retry note, raising workflow_clarity above the batch-operation cap.
Replace the conditional 'If your repository includes them, use:' phrasing with direct markdown links to references/api_reference.md, references/workflow_guide.md, and scripts/run_deseq2_analysis.py since the files verifiably exist in the bundle.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient (the example script and Implementation Details earn their tokens), but there is measurable padding: the "Key Features" section ("End-to-end DESeq2-like workflow: normalization (size factors), dispersion estimation/shrinkage...") restates the description and Implementation Details; the ~30-line volcano-plot block is generic matplotlib Claude already knows; and pinned version numbers ("pandas >= 1.4.3", "numpy >= 1.23.0", "scipy >= 1.11.0") are time-sensitive without being in a deprecated/old-patterns section, which the guidelines explicitly penalize. Not anchor 2 because the majority of the body is domain-specific and non-redundant; not anchor 4 because the redundancy exceeds minor trimmable instances. | 3 / 5 |
Actionability | The "Example Usage" section is a complete, runnable, copy-paste-ready script covering load, sample alignment ("common = counts_df.index.intersection(metadata.index)"), filtering, fitting ("dds.deseq2()"), Wald testing with contrast ("contrast=[\"condition\", \"treated\", \"control\"]"), shrinkage, export, and plotting. Concrete extras: install command ("uv pip install pydeseq2"), design-formula strings, and contrast format. This matches anchor 5 — fully executable code covering the common treated-vs-control case. | 5 / 5 |
Workflow Clarity | The example script has a clear numbered sequence ("1) Load inputs" through "6) Minimal volcano plot") and some implicit input cleaning (intersection alignment, "metadata.dropna(subset=[\"condition\"])", low-count filtering). But this is a batch operation over a whole count matrix, and there are no explicit validation checkpoints or error-recovery loops: nothing verifies the documented requirements ("non-negative integer counts", design variables present in metadata, samples matched) before fitting, and no verify-then-proceed guidance exists — per the rubric's cap, a batch workflow without validation cannot score above 3. | 3 / 5 |
Progressive Disclosure | The bundle is real and well-partitioned (references/api_reference.md, references/workflow_guide.md, scripts/run_deseq2_analysis.py all exist), and the "Notes on bundled references/scripts" section signals each file's purpose with one-level-deep references. Minor gaps keep it at anchor 4 rather than 5: the pointers use hedged conditional phrasing ("If your repository includes them, use:") instead of direct links, and the ~110-line inline example plus ~50 lines of Implementation Details push the SKILL.md toward monolith territory that a slimmer body plus the existing script could absorb. | 4 / 5 |
Total | 15 / 20 Passed |