The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill ingests data from external sources, specifically flag configurations and evaluation metrics via the vercel flags CLI.
Ingestion Points: Configuration data enters the context through vercel flags inspect, versions, and evaluations commands as described in SKILL.md and references/cleanup.md.
Boundary Markers: While the skill provides structured guidance on how to interpret these values (e.g., choosing values to inline), it does not explicitly instruct the agent to ignore potentially malicious text embedded within flag descriptions or variant names.
Capability Inventory: The skill has the capability to execute CLI commands (vercel flags create/set/archive) and modify local source code (inlining values during cleanup).
Sanitization: The skill relies on the agent to correctly identify and inline flag values based on the gathered evidence, but does not specify automated sanitization for the flag metadata itself.
[External Resource Downloads]: The skill references and provides instructions for installing the Vercel CLI and specific Vercel-authored SDKs.
Evidence: Mentions of pnpm i -g vercel and npx skills add vercel/flags@flags-sdk in SKILL.md.
Context: These resources originate from a well-known service provider and are documented as standard prerequisites for using the feature flagging platform.
[Credential Management]: The skill provides instructions for handling sensitive credentials like FLAGS_SECRET and OIDC tokens.
Evidence: SKILL.md and references/cli.md discuss the use of vercel env pull for local development and the generation of override cookies.
Context: The skill explicitly instructs the agent: "Do not print secrets to the terminal, logs, or chat," which aligns with security best practices for handling sensitive environment data.

vercel-flags