The Agent Skills Directory

[METADATA_POISONING]: The skill metadata and technical documentation (SKILL.md and AGENTS.md) claim the author is 'vercel' and that the content is maintained by 'Vercel Engineering'. The actual provider is 'liuchiawei'. This discrepancy constitutes a deceptive claim of authorship from a well-known trusted organization in the ecosystem, which may lead to unearned trust in the skill's recommendations and automated refactoring actions.
[INDIRECT_PROMPT_INJECTION]: The skill is intended to be used by an agent to review, write, and refactor React and Next.js codebases, which involves processing potentially untrusted data from the user's project.
Ingestion points: The agent ingests external React components, Next.js pages, and data fetching logic provided in the user's environment during refactoring tasks.
Boundary markers: No specific delimiters or 'ignore embedded instructions' warnings are provided to the agent to distinguish between the performance guidelines and instructions that might be maliciously embedded within the code being processed.
Capability inventory: The agent is granted the capability to modify and write application source code based on the ingested data and the skill's rules.
Sanitization: The skill lacks sanitization or validation mechanisms for the input code before it is used to influence code generation or refactoring outputs.
[DYNAMIC_EXECUTION]: The performance guideline in rules/rendering-hydration-no-flicker.md (and AGENTS.md section 6.5) explicitly recommends the use of dangerouslySetInnerHTML to inject an inline, synchronous JavaScript script into the DOM during server-side rendering. While this is a recognized pattern for initializing client-side state like themes without flickering, it involves the dynamic execution of script content defined within the skill's instructions.

vercel-react-best-practices