Conciseness

The body is instruction-dense rather than padded with background concepts Claude already knows, but the 22-item Hard Rules section heavily restates content already stated in the body — rules 11–12 repeat Step 4.5's citation rules, rule 16 restates Section G's subset requirement, and rules 19–22 restate the C.5/I/J 'mandatory in every output' declarations. This matches the score-3 anchor ('mostly efficient but includes some unnecessary explanation or could be tightened') — above score 2, since there is no concept-teaching padding, but below score 4, since the rule/section duplication is systematic rather than minor.

Actionability

As an instruction-only skill, the guidance is highly concrete: exact output sections A–J with required contents, exact redirect and Dataset Disclaimer text to reproduce verbatim, input examples, minimum retrieval counts ('2–4 outcome/exposure background references... 1 explicit evidence-gap note'), and a worked dependency-formula list. It stays at score 4 rather than 5 because a few instructions remain process-level without a worked example inline (e.g., 'provide the full workflow using the required stepwise format' delegates the format entirely to a reference file, and no sample workflow step or sample reference entry is shown in the body).

Workflow Clarity

The seven steps are explicitly ordered ('always run in order'), and Step 5 is a genuine validation gate before output: a dependency consistency checklist ('Do causal claims appear without explicit instrument-selection and harmonization logic?'), forbidden-claim rules, and explicit failure handling ('If dependency fails, remove or downgrade the downstream claim rather than silently keeping it'). Insufficient-detail handling ('infer a reasonable default and state assumptions explicitly') and the six-element self-critical risk review add error-recovery feedback. This matches the score-5 anchor (clear sequence with explicit validation steps and feedback loops).

Progressive Disclosure

The body is a well-organized overview with clearly signaled one-level-deep references ('→ Detailed pattern logic: [references/study-patterns.md]'), and all five linked reference files exist with no nested second-level references. However, three of the eight bundle files — analysis-modules.md, method-library.md, and workflow-step-template.md — are linked from neither SKILL.md nor any other reference file, making them undiscoverable. Good structure with a real navigation gap places this at score 4 rather than 5; it is above score 3 because the linked references are clear and the overview/detail split is appropriate.

Specificity

The description lists multiple concrete actions — 'instrument selection, harmonization, IVW-primary estimation, complementary estimators, sensitivity analyses, optional multivariable upgrades' — plus concrete deliverables ('recommended primary plan, stepwise workflow, figure plan, validation hierarchy, minimal executable version, publication upgrade path'), giving comprehensive coverage with no padding. This matches the score-5 anchor ('lists multiple specific concrete actions; comprehensive coverage') and not score 4, since there are no meaningful coverage gaps.

Completeness

Both questions are explicitly answered: the 'what' ('Generates complete two-sample Mendelian randomization research designs from a user-provided outcome, exposure or exposure family...') and an explicit 'Use when' clause ('Use when a study centers on summary-statistics causal inference with instrument selection...'). This exactly matches the score-5 anchor with concrete trigger phrases, and clearly exceeds score 4 where the 'when' is less specific.

Trigger Term Quality

Strong natural domain terms users would actually say: 'two-sample Mendelian randomization', 'exposure', 'outcome', 'IVW', 'sensitivity analyses'. However, phrases like 'complementary estimators', 'optional multivariable upgrades', and 'conservative evidence interpretation' lean toward methodology jargon, and common user variations like 'MR study', 'GWAS summary statistics', or 'causal inference' as standalone triggers are thin. This sits above the score-3 anchor (some relevant keywords but missing variations) but below score 5 (comprehensive coverage including synonyms).

Distinctiveness Conflict Risk

The two-sample MR exposure-screening niche is highly specific with distinct triggers ('IVW-primary estimation', 'MVMR', 'pleiotropy'), so it would not be confused with general research-planning, statistics, or literature-review skills. Clear niche with minimal conflict risk, matching the score-5 anchor rather than score 4, which presumes overlap with closely related skills.