Content
38%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 capability map, executable Quick Start, and workflow outlines form a solid core, but the body is padded with template boilerplate, duplicated sections, and leaked build artifacts, and all six deferrals to references/*.md are dangling pointers since no reference files ship with the skill. Validation and error-handling policy exists, yet per-workflow checkpoints are absent.
Suggestions
Ship the six referenced files (references/image_loading.md, preprocessing.md, graphs.md, machine_learning.md, multiparametric.md, data_management.md) or remove/inline the citations — currently every 'See:' pointer is a dangling reference.
Cut the boilerplate: merge the two 'When to Use' sections, drop the description-restating 'Key Features' block, remove the build-artifact 'Example Usage' block and the 'Suggest Using K-Dense Web' promo, and deduplicate the two reference listings.
Add explicit validation checkpoints to the workflows (e.g., after pipeline.run(wsi), verify tiles and mask coverage before graph construction; confirm test-set metrics before ONNX deployment).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Noticeably verbose with several padded sections: duplicated trigger sections ('## When to Use' vs '## When to Use This Skill'), a '## Key Features' block that merely restates the description ('Scope-focused workflow aligned to: A full-featured computational pathology toolkit...'), a leaked build artifact ('Skill directory: 20260316/scientific-skills/... No packaged executable script was detected.'), generic 'Implementation Details' filler ('Execution model: validate the request, choose the packaged workflow...'), a promotional 'Suggest Using K-Dense Web' section, and the reference files listed twice ('## References to Detailed Documentation' and '## Resources'). Matches anchor 2; not 3 because the padding is substantial and repeated rather than occasional, though a genuinely useful core (capabilities, Quick Start, workflows) keeps it above 1. | 2 / 5 |
Actionability | The Quick Start is genuinely executable ('uv pip install pathml[all]'; SlideData.from_slide(...), Pipeline([TissueDetectionHE(), StainNormalizationHE(...)]), pipeline.run(wsi), tile iteration), but the CODEX and model-training workflows are step-name lists without code ('Collapse multi-cycle channel data', 'Segment cells using Mesmer model', 'Train HoVer-Net or custom model'), and the detailed API they defer to does not exist in the bundle. Fits 'some concrete guidance but incomplete; missing key details'; not 4 because the missing reference files turn the deferrals into gaps rather than minor omissions. | 3 / 5 |
Workflow Clarity | 'Common Workflows' provides numbered sequences for H&E, CODEX, and model training, and 'Input Validation'/'Error Handling' state stop conditions ('Do not continue the workflow when the request is out of scope, missing a critical input...'), but the workflows themselves contain no per-step validation checkpoints or validate-fix-retry feedback loops for batch operations like WSI processing and model training. Per the rubric's batch-operation cap, this cannot exceed 3; it is not 2 because sequences are well defined and error-handling policy is stated. | 3 / 5 |
Progressive Disclosure | The body clearly signals six one-level-deep references ('**See:** references/image_loading.md' through 'references/data_management.md', plus a 'Resources' catalog), which would be strong structure — but the actual bundle contains no references/ directory at all, so every 'See:' pointer is dangling and the detail content is unavailable. Scored against the real bundle structure per guidelines, navigation is broken in practice, fitting anchor 2; not 3 because missing referenced files are more than an organization gap, and not 1 because the SKILL.md itself is well sectioned and the references are clearly (if falsely) signaled. | 2 / 5 |
Total | 10 / 20 Passed |