Content
61%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 exceptionally actionable — complete commands, output contracts, and question-phrasing-to-answer-line mappings — and its bundle structure is real and well-signaled. However, it is severely overlong due to repeated restatement of the same rules and basic statistics explanations, and it contains a direct contradiction (R DESeq2-first vs a PyDESeq2 workflow) that muddies the intended execution order.
Suggestions
Deduplicate: consolidate the Venn union-denominator rule, the exclusive-vs-inclusive set semantics, the 'also DE in strain X' intersection rule, and the R-vs-pydeseq2 guidance into single canonical sections — each currently appears 2-4 times across 'CRITICAL', 'LOOK UP DON'T GUESS', and 'Analysis conventions'.
Remove or relocate material Claude already knows: the Interpretation Framework threshold table ('padj < 0.05 = significant', 'LFC > 1 = 2x up'), 'Core Principles', and 'When to Use' (which duplicates the frontmatter description) add tokens without adding skill-specific knowledge.
Reconcile the execution-order contradiction: make the 7-step 'Analysis Workflow' explicitly subordinate to the PRIMARY SCRIPTS path (notebook/authoritative script first, R DESeq2 wrappers second) so Step 3's 'Run PyDESeq2' and the 'Python (PyDESeq2): ALL DESeq2 analysis' boundary no longer conflict with CRITICAL #2.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~500-line body duplicates its own guidance repeatedly: the Venn union-denominator rule appears three times (CRITICAL #6, the 'LOOK UP DON'T GUESS' bullet, and 'Set-operation percentages'), exclusive-vs-inclusive set semantics twice ('Uniquely DE in A/B not C' vs CRITICAL #4/#5), 'also DE in strain X' twice, and R-vs-pydeseq2 guidance in at least four places. It also explains concepts Claude already knows ('log2FoldChange > 1 ... 2x up or down', 'padj < 0.05 | Statistically significant', the 'Core Principles' and 'When to Use' filler sections). This matches the level-2 anchor (noticeably verbose; several unnecessary explanations or padded sections); it is not level 3 because the duplication and basic-concept padding are extensive rather than incidental. | 2 / 5 |
Actionability | Every script comes with a copy-paste-ready invocation with full flags and placeholder paths (e.g. the r_deseq2_wrapper.py multi-contrast example), exact parseable output formats ('# SIG_sex_M_vs_F_unshrunk_strict ...'), concrete `tu run` JSON commands, and a decision table mapping literal question phrasings to specific output lines. This matches the level-5 anchor (fully executable, copy-paste ready, common cases covered). | 5 / 5 |
Workflow Clarity | Two competing execution paths coexist without reconciliation: the top-of-file 'PRIMARY SCRIPTS — use these FIRST' / CRITICAL #2 ('Use R DESeq2, not pydeseq2') route versus the later 7-step 'Analysis Workflow' whose Step 3 says 'Run PyDESeq2' with 'Python (PyDESeq2): ALL DESeq2 analysis'. Validation checkpoints do exist (read the executed notebook first, `ls` for authoritative scripts, scan every sheet, match filters exactly, error-quick-reference table), but the contradiction plus a duplicated list item ('3.' appears twice) makes the authoritative sequence genuinely ambiguous — fitting the level-3 anchor rather than level 4, whose 'minor validation gaps' standard the contradiction exceeds. | 3 / 5 |
Progressive Disclosure | All 10 reference files and all 9 scripts cited in the body exist, are one level deep, and are indexed twice (inline links in workflow steps plus the 'Reference Files'/'Utility Scripts' lists with one-line descriptions) — clearly signaled and easy to navigate. Not level 5: the body is not a lean overview — several hundred lines of detailed pitfall guidance are inlined that could live in references (e.g. the pydeseq2 workflow guidance duplicated in both the body and references/pydeseq2_workflow.md), matching the level-4 anchor (good structure, most content appropriately placed, minor organization gaps). | 4 / 5 |
Total | 14 / 20 Passed |