The Agent Skills Directory

[METADATA_POISONING]: The skill's YAML frontmatter identifies 'vercel' as the author, despite the skill being provided by Galyarder Labs as indicated by the footer and context. This impersonation of a trusted organization can lead to a misjudgment of the skill's safety and official status.
[PROMPT_INJECTION]: The instructions include 'Global Protocols' that attempt to override default agent behavior. Specifically, the 'INCIDENT Mode' directive instructs the agent to 'Bypass planning for hotfixes', which uses a bypass marker to circumvent standard safety and planning procedures.
[EXTERNAL_DOWNLOADS]: The skill recommends the installation and use of several external third-party libraries, including 'better-all' from GitHub (shuding/better-all), 'lru-cache' (isaacs/node-lru-cache), and 'svgo' via npx. These are well-known resources used for performance optimization and are documented neutrally as best practice suggestions.
[DYNAMIC_EXECUTION]: The documentation suggests a pattern for preventing hydration flicker by injecting and executing synchronous JavaScript directly into the DOM using 'dangerouslySetInnerHTML'. While a common React pattern, it involves the runtime execution of generated script strings.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect injection as it is designed to ingest and analyze external React and Next.js codebases.
Ingestion points: The skill triggers on tasks involving React components, Next.js pages, and data fetching logic (SKILL.md).
Boundary markers: The 'Security & Multi-Agent Hygiene' section explicitly instructs the agent to treat web content and external data as hostile (SKILL.md).
Capability inventory: The agent is expected to perform automated refactoring and code generation based on the ruleset.
Sanitization: The instructions recommend redacting secrets and PII before sharing context with subagents.

vercel-react-best-practices