Content
56%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 delivers strong domain-specific actionability — complete CLI examples, a thorough parameter table, explicit input/output contracts, and executable preprocessing code — wrapped in a heavily padded shell of duplicated governance boilerplate. Progressive disclosure is underused: a real reference file exists but is never named, while reference-worthy material (methods detail, security checklists) is inlined.
Suggestions
Cut the redundant meta-sections (Key Features, Implementation Details, Output Requirements, Output Contract, Inputs to Collect, Input Validation, Validation and Safety Rules, Error Handling, Risk Assessment, Security Checklist, Evaluation Criteria, Lifecycle Status) down to one consolidated workflow/safety section — they restate the same rules ~5 times across ~150 lines.
Replace the generic 'references/ contains supporting rules' pointer with a named link, e.g. '**Runtime rules**: See [references/runtime_checklist.md](references/runtime_checklist.md)', and move the Methods detail and Security/Risk boilerplate there.
Fix the inconsistent audit example: 'python scripts/main.py --input "Audit validation sample..." --format json' passes a prose string where the documented contract requires a .h5ad path — replace it with a real minimal .h5ad example or the --help smoke check only.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Roughly 150 lines are redundant boilerplate: 'Key Features' restates the description, 'See `## Usage` above for related details' and 'See `## Workflow` above for related details' are filler, and 'Output Requirements', 'Output Contract', 'Inputs to Collect', 'Input Validation', 'Error Handling', 'Validation and Safety Rules', 'Risk Assessment', 'Security Checklist', 'Evaluation Criteria', and 'Lifecycle Status' repeatedly restate the same scope/validation/output-discipline rules. This is several padded sections beyond the 'some unnecessary explanation' of anchor 3; it is not 1 because the domain sections (Usage, Parameters, Methods, Input/Output formats) are dense and non-padding. | 2 / 5 |
Actionability | Mostly executable guidance: complete CLI examples ('python scripts/main.py --input data.h5ad --start-cell stem_cell_cluster --method diffusion --output ./results'), a full 17-row parameter table with defaults, a concrete output file tree, and a runnable scanpy preprocessing snippet. Minor gap: the audit command passes a prose sentence as --input where the documented input contract requires a .h5ad AnnData path, so it would not run as written; this keeps it below anchor 5. | 4 / 5 |
Workflow Clarity | A clear sequenced workflow exists ('1. Confirm the user objective... 2. Validate that the request matches the documented scope and stop early... 5. If execution fails... switch to the fallback path') with an explicit pre-execution smoke check ('python -m py_compile scripts/main.py') and a stated recovery path on failure. Not anchor 5 because validation guidance is scattered across many redundant sections rather than one explicit validate→fix→retry loop with checkpoints, leaving minor validation gaps. | 4 / 5 |
Progressive Disclosure | The bundle has one reference file (references/runtime_checklist.md), but the body only says 'references/ contains supporting rules, prompts, or checklists' — never naming or linking the actual file, so the reference is present but not clearly signaled. Meanwhile content that belongs in references (four Methods subsections, Risk Assessment, Security Checklist) is inlined in a ~400-line body. This matches anchor 3; it is not 4 because key content is not appropriately split out and the reference is buried. | 3 / 5 |
Total | 13 / 20 Passed |