The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes external request data, such as cookies and headers, along with data fetched from remote providers to evaluate feature flags. While this represents an attack surface for data processed at runtime, the risk is managed through standard implementation patterns.
Ingestion points: The identify function accepts headers and cookies from incoming requests; provider adapters ingest data from external feature flag platforms (e.g., Statsig, LaunchDarkly, PostHog).
Boundary markers: The skill documentation does not explicitly show prompt isolation delimiters for the injected data, but it promotes programmatic flag evaluation rather than direct string interpolation into prompts.
Capability inventory: The skill allows the agent to execute vercel flags CLI commands, install npm packages, and modify project files such as flags.ts.
Sanitization: The SDK includes a safeJsonStringify utility designed to escape dangerous characters during JSON serialization, reducing XSS risks.
Sensitive Data Management: The skill provides instructions for generating and storing FLAGS_SECRET using secure methods (e.g., crypto.randomBytes) and utilizing the vercel env pull command to manage project secrets in .env.local. These are standard practices for secure local development.
Vendor-Specific Resource Usage: The skill facilitates the installation and use of various npm packages under the @flags-sdk and @vercel scopes. These resources are official components provided by the skill's author for feature flag management and integration.

flags-sdk