Content
82%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 well-structured regulatory skill: every workflow is numbered with a validation step, quick-reference tables carry the MDR facts Claude needs, and detail is correctly pushed to three real one-level-deep reference files with a usable script. Weaknesses are modest: some generic advisory padding, no error-recovery loops around the validation checkpoints, and a TOC/navigation gap for the final section.
Suggestions
Trim the generic 'Notified Body Interface' selection-criteria table and the 'Risk' row of the classification matrix — both are padding relative to the MDR-specific content surrounding them; move any needed Notified Body detail into technical-documentation-templates.md.
Add error-recovery guidance to the validation checkpoints, e.g., 'if the gap analyzer reports critical gaps, remediate documentation before NB submission' or 'if rules conflict, apply the higher-classification rule (Annex VIII, last indent) and document the rationale'.
Add 'Notified Body Interface' to the Table of Contents so every section is navigable, and consider relocating 'Qualified Evaluator Requirements' to clinical-evidence-requirements.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly dense, MDR-specific facts (PSUR frequencies, vigilance timelines, conformity routes) that Claude cannot be assumed to know, but there are trimmable padded passages: the generic 'Notified Body Interface' advisory table ('Designation scope', 'Capacity', 'Fee structure') and the simplistic 'Risk: Lowest/Low-medium/Medium-high/Highest' row of the classification matrix add little regulatory value. Anchor 4 ('minor instances of over-explanation that could be trimmed') fits better than 5, and it is well above the 3 anchor's 'unnecessary explanation'. | 4 / 5 |
Actionability | Guidance is fully executable: copy-paste-ready script invocations ('python scripts/mdr_gap_analyzer.py --device "Device Name" --class IIa --output json'), worked classification examples with definite outcomes (suture → Rule 8 → Class IIb), and concrete lookup tables (PSUR schedule, incident reporting timelines of 2/10/15 days) covering the common cases. This matches the top anchor rather than level 4's 'minor gaps'. | 5 / 5 |
Workflow Clarity | All five workflows are clearly numbered with an explicit bolded '**Validation:**' checkpoint (e.g., 'Classification confirmed with Notified Body', 'PMS system audited annually') plus checklists (GSPR matrix, UDI label, pre-submission). However, validations are single checkpoints with no error-recovery feedback loops (what to do when the gap analyzer flags critical gaps or the NB challenges the rationale), which is exactly the gap between anchor 4 and anchor 5; the sequence quality keeps it clearly above 3. | 4 / 5 |
Progressive Disclosure | Good structure overall: three real, one-level-deep reference files each summarized with a content bullet list, a working script, and a TOC. Minor organization gaps keep it at anchor 4 rather than 5: the 'Notified Body Interface' section is missing from the Table of Contents, and its advisory detail (plus 'Qualified Evaluator Requirements') arguably belongs in a reference file rather than the overview body. | 4 / 5 |
Total | 17 / 20 Passed |