The Agent Skills Directory

[METADATA_POISONING]: The skill metadata inaccurately identifies the author as 'vercel' in the SKILL.md frontmatter and 'Vercel Engineering' within AGENTS.md. This directly contradicts the provided skill author context which identifies 'ivantsxx' as the author. Such deceptive metadata can lead to misplaced trust in the provenance and safety of the instructions.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to provide high-level instructions that override or supplement the agent's default behavior when handling React-related tasks. The inclusion of text explicitly stating the document is 'optimized for automation and consistency by AI-assisted workflows' indicates an intent to deeply integrate with and influence the agent's logic.
Ingestion points: SKILL.md, AGENTS.md, and various rule files in the rules/ directory act as the primary injection vector for these instructions.
Boundary markers: The skill lacks explicit boundary markers or warnings to the agent that these guidelines should not override its core safety or operational constraints.
Capability inventory: The skill seeks to influence the agent's capabilities in writing components, implementing data fetching, performing code reviews, and executing refactors.
Sanitization: There is no evidence of sanitization or filtering applied to these instructions before they are incorporated into the agent's context.

vercel-react-best-practices