Conciseness

The body is dense with decision tables, a sizing formula, and checklists; it never explains background concepts Claude already knows (e.g., no "what is a deep link" prose). Every section carries non-generic specifics like "one prompt at install gets ~2% adoption; 3+ contextual prompts get 15–25%" that Claude could not invent. Anti-drift: anchor 4 ("minor instances of over-explanation") doesn't fit — there is nothing to trim.

Actionability

For an instruction-only skill the guidance is fully executable: a hard sizing bound ("Max referral reward (per side) ≤ (LTV × target margin) - other CAC"), per-vertical default reward values, a fraud table mapping vector → specific mitigation (device fingerprint, qualifying actions, reward caps), named tools (Branch, OneLink, Adjust), named analytics events, and a fill-in output template. No vague directives remain.

Workflow Clarity

The process is clearly sequenced — numbered initial assessment → fit gate → reward design → mechanics → fraud → plan output → measurement — with explicit decision checkpoints ("If LTV is unclear, route to `asc-metrics` first"; "If 'weak fit,' steer the user toward `creator-ugc-marketing` or `retention-optimization`"). The mechanics checklist, launch checklist ("Deep links tested cross-platform", "Reward issuance tested end-to-end"), and weekly K-factor measurement loop provide the checklists and feedback loop the top anchor requires.

Progressive Disclosure

No bundle files exist (references/, scripts/, assets/ are absent), so the skill is a single well-sectioned file with clear headers and cross-skill handoffs — good structure with appropriately placed content, matching anchor 4. It does not reach anchor 5, which requires well-signaled one-level-deep reference files; the ~165-line body inlines the full fraud table, tooling guide, and output template that could live in separate reference files.

Specificity

The description lists multiple concrete capabilities — "design, launch, or optimize an in-app referral / invite / share-to-earn program — including reward structure, mechanics, fraud prevention, deep link setup, and viral coefficient measurement" — giving comprehensive, non-vague coverage of the domain. It names the domain plus five specific action areas, matching the top anchor; nothing is abstract padding.

Completeness

Both questions are answered explicitly: the "what" is the design/launch/optimize capability list, and the "when" is a literal "Use when the user mentions..." clause with concrete trigger phrases. This mirrors the anchor-5 good example structure exactly; anchor 4 ("'when' could be more explicit") is clearly too low.

Trigger Term Quality

Trigger coverage is exhaustive and natural: "referral program", "invite a friend", "refer and earn", "share to earn", "viral loop", "viral coefficient", "K-factor", "double-sided rewards", "give X get X", "invite link", "share sheet", "Branch referrals", and "how to make my app go viral" cover synonyms, industry jargon, and lay phrasings users would actually say. Anti-drift check against anchor 4 ("a few natural terms missing") fails — no common variation is missing.

Distinctiveness Conflict Risk

The referral/invite niche is distinct with dedicated triggers, and the description actively de-conflicts against neighbors: "For deep link infrastructure that referrals depend on, see attribution-setup. For organic content-driven virality (UGC, creator), see creator-ugc-marketing." Minimal overlap risk with sibling growth skills.