React Native Performance Playbook

The material is primarily read-only React Native coding guidance and requests no credentials, external writes, or elevated permissions. No malware, covert exfiltration, destructive default, or fabricated provenance red line is evident, so baseline trust is relatively strong; however, user confirmation, rollback, sensitive-data guidance, dependency-security boundaries, and verifiable per-rule attribution are incomplete, preventing a near-full score.

The documentation is organized and gives rules, incorrect/correct examples, and some failure explanations, making the happy path understandable. Static evidence also shows reliability concerns: for example, a useCallback example is placed inside renderItem in a way that violates Hooks rules, and an animation example appears to misuse withTiming inside interpolate. There are no committed tests, compatibility matrix, or systematic abnormal-input handling, so substantial points are deducted and the static cap applies.

The audience and scenarios—React Native/Expo development, performance, animation, native modules, and monorepos—are explicit, with reasonably searchable trigger terms. However, non-fit boundaries, input/output expectations, platform-version requirements, and Chinese-language support are not defined; exceptions for short lists, alternative component libraries, and project constraints are also limited.

SKILL.md provides frontmatter, version, license, priority ordering, routing, and progressive quick reference; the README adds repository structure and maintenance context. The selected skill lacks a changelog, clearly assigned maintenance responsibility, dependency-installation notes, FAQ, explicit limitations, and a stable compatibility policy. The rule counts and category descriptions are not fully consistent across SKILL.md, AGENTS.md, and README.md, so points are deducted.

Many rules include directly usable TypeScript/TSX comparisons covering common list, animation, navigation, and UI optimization tasks, so the core use case is plausible. Nevertheless, some examples have execution or semantic doubts, recommendations are sometimes overly absolute, and there is little measurement, verification, or alternatives guidance. Considerable human review remains necessary, and the static cap limits this score to 7 or below.

The skill includes rule files, concrete code comparisons, license and version metadata, and a scheduled vendored-update workflow, providing some auditability. It lacks a skill-specific test suite, CI validation, performance benchmarks, and strong cross-source corroboration; citations are sparse and include an example.com placeholder. Static verifiability is therefore limited.