The description names the domain (in-app paywalls, upgrade screens) and some actions ('create or optimize'), but doesn't list multiple specific concrete actions like 'design layout,' 'write copy,' 'A/B test variants,' etc. The actions remain at a high level.

Clearly answers both 'what' (create/optimize in-app paywalls, upgrade screens, upsell modals, feature gates) and 'when' with an explicit and extensive list of trigger phrases. Also includes helpful boundary-setting by distinguishing from related skills (page-cro, pricing-strategy).

Excellent coverage of natural trigger terms including 'paywall,' 'upgrade screen,' 'upsell,' 'feature gate,' 'convert free to paid,' 'freemium conversion,' 'trial expiration screen,' 'limit reached screen,' 'how do I get users to pay,' and many more variations that users would naturally say.

Explicitly distinguishes itself from 'page-cro' (public pricing pages) and 'pricing-strategy,' clearly carving out its niche as in-product upgrade moments. The boundary 'where the user has already experienced value' is a sharp differentiator.

The skill is reasonably organized but includes some unnecessary framing ('You are an expert...') and sections that state obvious UX principles Claude already knows (e.g., 'Don't trap or pressure,' 'Maintain trust'). Several bullet points could be tightened, and the core principles section largely restates common sense.

The wireframe-style paywall templates are helpful and somewhat concrete, but they're pseudocode/mockups rather than executable code. The guidance is mostly directional ('show what the feature does,' 'quick path to unlock') rather than providing specific copy formulas, exact component structures, or implementation-ready examples with real copy variations.

There's a logical flow from assessment → principles → trigger points → components → optimization → testing, but the skill lacks explicit sequencing for the actual task of creating a paywall. There are no validation checkpoints (e.g., 'verify your paywall includes these 7 components before finalizing') or feedback loops for iterating on designs.

There's one reference to 'references/experiments.md' and cross-references to related skills, which is good. However, the main file is quite long (~180 lines) with sections like anti-patterns, timing rules, and A/B testing metrics that could be split into reference files. The content is somewhat monolithic for its length.