The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill facilitates the ingestion of data from external providers (e.g., Vercel Flags, Statsig, LaunchDarkly) and user-provided request context (cookies, headers) to determine feature flag states. If this data originates from an untrusted source, it could theoretically influence the agent's behavior or output if it is instructed to act based on those flag values.
Ingestion points: Flag values are fetched from external providers; evaluation context is extracted from headers and cookies in identify functions.
Boundary markers: No specific boundary markers are required for flag values, although the SDK provides safeJsonStringify for XSS protection in rendered outputs.
Capability inventory: The skill involves file system writes (creating flags.ts and discovery routes), and network communication via the Vercel CLI.
Sanitization: The safeJsonStringify utility is provided in the core library to escape dangerous characters in flag values.
[Official Package Dependencies]: The skill instructs the installation of several Node.js packages such as flags, @flags-sdk/vercel, and @vercel/toolbar. These are official resources provided by the skill author for feature flag management.
Evidence: Multiple pnpm i and npm install commands are listed throughout SKILL.md and the reference documentation.
[Sensitive Environment Configuration]: The skill guides the user in managing sensitive credentials like VERCEL_OIDC_TOKEN and FLAGS_SECRET using official CLI tools (vercel env pull). Proper handling of these variables is emphasized, such as marking production secrets as sensitive.
Evidence: Detailed instructions for pulling environment variables and generating secrets are provided in the Pull environment variables and FLAGS_SECRET sections.

flags-sdk