The body is dense but token-efficient: tables, checklists, and formulas carry the load with no concept explanations Claude already knows (no 'what is a referral program' padding). Every section adds domain-specific tactical knowledge rather than filler, matching the lean-and-efficient anchor. Not level 2 because there is no unnecessary explanation to tighten.

Highly actionable for an instruction-only skill: a reward-sizing formula ('Max referral reward ≤ (LTV × target margin) - other CAC'), concrete default reward ranges by app type, a K-factor formula with interpretation bands, a fraud-vector/mitigation table, and a copy-ready Output Template. Per the scoring notes, absence of code is not penalized when guidance is this concrete; not level 2 because guidance is complete and specific rather than pseudocode-like.

A clear sequenced flow runs from Initial Assessment → fit decision gate ('Is a Referral Program Right for You?') → reward structure/sizing → mechanics → fraud → Output Template, with conditional routing ('If LTV is unclear, route to asc-metrics first') and verification checklists (Mechanics Checklist, Launch Checklist with test items). Not level 2 because decision gates and checklists supply the checkpoints the anchor requires, and the task is planning rather than destructive execution so heavy feedback loops are not expected.

No bundle files exist (references/scripts/assets absent) and all content lives inline in a single ~160-line SKILL.md that is well-sectioned but monolithic; catalogs like the reward-pattern table and fraud-mitigation table could plausibly offload to reference files. Not level 3 because there are no one-level-deep reference files to navigate to, and not level 1 because section organization is clear rather than a wall of text.

Lists multiple concrete actions — 'design, launch, or optimize', 'reward structure, mechanics, fraud prevention, deep link setup, and viral coefficient measurement' — matching the multiple-specific-actions anchor. Voice stays third person ('When the user wants...', 'Use when the user mentions...'), so no specificity penalty applies.

Explicitly answers both what (design/launch/optimize a referral program with named sub-tasks) and when ('Use when the user mentions ...'), satisfying the both-what-and-when anchor. Not level 2 because the 'Use when' trigger clause is present and explicit.

Broad coverage of natural user phrasings — 'invite a friend', 'refer and earn', 'share to earn', 'viral loop', 'K-factor', 'give X get X', 'how to make my app go viral' — well beyond common variations. Not level 2 because it goes far beyond a handful of relevant keywords into the full vernacular users would actually say.

Occupies a clear niche (in-app referral/viral growth) with distinct triggers and actively routes adjacent needs away ('For deep link infrastructure ... see attribution-setup', 'For organic content-driven virality ... see creator-ugc-marketing'), minimizing wrong-skill activation. Not level 2 because the disambiguation handoffs make overlap unlikely.