Content
63%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 well-organized, table-dense, and rich in the regulation-specific facts (schedules, timelines, routes, checklists) that make a compliance skill valuable, with clearly sequenced workflows ending in validation steps. Its two real weaknesses are that every referenced bundle file and the gap-analyzer script is missing from the bundle — so the only copy-paste commands would fail — and the validation steps name outcomes rather than executable checks.
Suggestions
Ship the cited bundle files (references/mdr-classification-guide.md, references/clinical-evidence-requirements.md, references/technical-documentation-templates.md, scripts/mdr_gap_analyzer.py) or remove the citations — as delivered, every reference and the only executable commands in the skill are dead paths.
Make the **Validation:** steps operational: replace outcome statements like 'Technical file reviewed for completeness' with concrete checks (e.g., 'verify every GSPR matrix row has evidence attached; if any row is empty, resolve before submission') and add a fix-and-retry loop after the gap analyzer runs.
Trim sections that restate general knowledge without regulatory specificity — the notified body Selection Criteria table and Qualified Evaluator Requirements — or move them into the technical-documentation reference file.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, table-driven, and factual — PSUR cadences per Art. 86, 2/10/15-day incident timelines, conformity routes per class — exactly the precision-dependent details worth encoding in a skill. It is not level 5 because a few sections restate knowledge Claude already has without adding regulatory specificity (the notified body "Selection Criteria" table, "Qualified Evaluator Requirements"), and the general Classification Matrix is simplistic relative to the referenced Annex VIII rules; this is more than the lean level-5 anchor but stays above level 3's 'several unnecessary explanations'. | 4 / 5 |
Actionability | There is substantial concrete guidance — worked classification examples with outcomes, a GSPR compliance checklist, a PSUR schedule table, and UDI label checklists — but it falls short of level 4 because the only executable commands in the skill (`python scripts/mdr_gap_analyzer.py --device "Device Name" --class IIa`, etc.) point to a script that does not exist in the bundle, and the bolded validation steps are non-operational ("Technical file reviewed for completeness", "Classification confirmed with Notified Body" name outcomes without criteria or commands). This matches the level-3 anchor ('concrete guidance but incomplete... missing key details') better than level 4's 'minor gaps'. | 3 / 5 |
Workflow Clarity | Each of the five major processes has a clearly numbered 8-step sequence terminating in an explicit bolded **Validation:** step, plus a pre-submission checklist — matching the level-4 anchor ('clear sequence with most checkpoints present'). Not level 5: the validation steps state outcomes rather than executable checks, and there are no fix-and-retry feedback loops (e.g., what to do when the GSPR matrix or gap analysis fails), though the destructive/batch cap does not apply since this is documentation/planning work. | 4 / 5 |
Progressive Disclosure | Per the guideline to score against the actual bundle structure: the body clearly signals one-level-deep references with a dedicated 'Reference Documentation' section describing each file's contents, but none of the cited files (`references/mdr-classification-guide.md`, `references/clinical-evidence-requirements.md`, `references/technical-documentation-templates.md`) or `scripts/mdr_gap_analyzer.py` exist in the bundle — every pointer is dead. Meanwhile, detailed content that the skill's own structure says belongs in those files (classification rules, CER structure, GSPR matrix) is inlined in SKILL.md, duplicating the described reference contents. This lands between the level-3 anchor ('references present but not clearly signaled; content that should be separate is inline') and level 2 ('references that lead nowhere'); the strong section headers and TOC keep it at 3 rather than 2. | 3 / 5 |
Total | 14 / 20 Passed |