Content
52%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 skill has a clearly sequenced eight-step workflow with explicit validation checkpoints and an unusually well-integrated set of one-level-deep reference modules mapped to specific steps. However, the body is heavily redundant — the resource-bucket rule and hard rules are each restated across four or five sections — and the guidance remains abstract, with no worked example of a constraint-driven redesign or a sample output section.
Suggestions
Consolidate 'Hard Rules', 'What This Skill Should Not Do', and 'Quality Standard' into one rule list, and state the available/obtainable/unavailable resource buckets once (in Step 2) instead of four times across the document.
Remove the inlined constraint and route-family lists from Steps 3–4 since they duplicate references/constraint-taxonomy.md and references/study-route-family-library.md, keeping SKILL.md as a lean overview that points to those modules.
Add one short worked example (e.g., a two-line sample of a deferred component and a minimum executable version for a public-data-only constraint profile) to make the abstract Step 5/6 guidance concrete.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is noticeably verbose through redundancy, not explanation of known concepts: the available/obtainable/unavailable resource buckets are stated at least four times ("If the user has not clearly stated the resource situation...", the Input Validation follow-ups, Step 2, Hard Rule 3, and Output Section C), and "Hard Rules", "What This Skill Should Not Do", and "Quality Standard" are three sections largely restating the same rules. Not a 3 because this goes beyond "could be tightened" — roughly a third of the ~320-line body is repeated restatement; not a 1 because it never explains concepts Claude already knows (no primer on cohorts or budgets). | 2 / 5 |
Actionability | The guidance is structured and partly concrete — 8 sequenced steps, 11 mandatory output sections A–K, and concrete category lists (constraint types like "sample size", route families like "retrospective clinical cohort") — but key direction stays abstract, e.g., "Narrow the study until it becomes realistically buildable" and "Every major change should be tied to a concrete constraint" give no worked example or sample output. Not a 4 because the gaps exceed "minor" (no example of a constraint-driven redesign anywhere); not a 2 because the step structure and output template are concrete, executable directions rather than high-level hints. | 3 / 5 |
Workflow Clarity | Steps 1–8 are clearly sequenced with explicit checkpoints: an input-validation gate ("If critical feasibility information is missing, do not assume ideal access. Instead, ask targeted follow-up questions"), a fallback path ("the final output must be explicitly labeled as provisional and assumption-dependent"), and a closing assumption review (Section J). Not a 5 because there is no explicit feedback loop back from the Step 7 dependency audit to revise the recommendation when a critical dependency is found broken; not a 3 because checkpoints are explicit, not merely implied. | 4 / 5 |
Progressive Disclosure | All 11 referenced files exist in references/ and each is mapped to a specific workflow step or output section with one-level-deep, clearly signaled navigation (e.g., "references/constraint-taxonomy.md → use when classifying the dominant feasibility constraints in Section B"). Not a 5 because SKILL.md is ~320 lines and inlines constraint lists and route families that are duplicated in references/constraint-taxonomy.md and references/study-route-family-library.md, so the body/overview split is not clean; not a 3 because structure and signaling are genuinely good. | 4 / 5 |
Total | 13 / 20 Passed |