Content
60%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 a well-organized, actionable phase-by-phase pipeline with concrete tooling, but it is undermined by a dangling reference structure — all five referenced bundle files are missing — and by redundant summary/principles sections that inflate token cost. Adding the referenced files or inlining their critical content would resolve the biggest weakness.
Suggestions
Ship the referenced bundle files (phase-procedures.md, tool-reference.md, reference-data.md, report-template.md, test_spatial_omics.py) or inline the critical decision logic and tool invocation syntax they were meant to hold — currently every 'See X.md' pointer leads to a nonexistent file.
Delete or drastically shrink the closing 'Summary' section and the 'KEY PRINCIPLES' list, which restate the phase descriptions and reference-file list already given above.
Add explicit per-phase validation checkpoints (e.g., 'if no domains meet the marker threshold, fall back to global enrichment' or 'verify gene ID resolution succeeded before Phase 2') instead of relying on a single end-of-run completeness checklist.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Tables, tool names, and phase listings mostly earn their tokens, but the closing 'Summary' section restates the phase list almost verbatim and 'KEY PRINCIPLES' duplicates phase descriptions, so the body could be meaningfully tightened. Not 4 because an entire redundant section is pure padding; not 2 because there is no over-explanation of concepts Claude already knows. | 3 / 5 |
Actionability | Concrete guidance throughout: named tools per phase ('STRING_functional_enrichment', 'OpenTargets_get_associated_targets_by_disease_efoId'), an input parameter table with examples, FDR cutoffs, HuBMAP parameter details, and a report filename pattern. Falls short of 5 because tool invocation syntax and decision logic are deferred to a reference file rather than shown, leaving minor gaps. | 4 / 5 |
Workflow Clarity | Phases 0-8 are clearly sequenced with Phase 0 marked 'ALWAYS FIRST', conditional phases explicitly gated ('Cancer/Inflammation only', 'if data available'), and a completeness checklist closes the loop. Not 5 because validation checkpoints are implicit (a single end checklist rather than per-phase gates) and the detailed decision logic is delegated elsewhere. | 4 / 5 |
Progressive Disclosure | References are clearly signaled and one level deep ('See phase-procedures.md for detailed workflows'), but the bundle contains no reference files at all — phase-procedures.md, tool-reference.md, reference-data.md, report-template.md, and test_spatial_omics.py are all absent, so every deep-dive pointer dangles and the disclosure structure is non-functional. Not 3 because the actual bundle structure breaks the navigation the body promises. | 2 / 5 |
Total | 13 / 20 Passed |