The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Vercel CLI to collect project metrics and configuration. All shell-outs are performed using execFile with arguments passed as arrays, which prevents shell injection vulnerabilities. It correctly identifies and handles CLI execution across POSIX and Windows environments.
[DATA_EXPOSURE]: The skill reads project configuration files (.vercel/project.json, .vercel/repo.json) to identify the Vercel environment. It includes a robust redaction mechanism (redactSensitiveText in lib/vercel.mjs) to ensure that authentication tokens and private identifiers are not exposed in logs or report output. It specifically warns agents against putting auth tokens in shell commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from the project codebase and Vercel CLI output. It mitigates injection risks by using deterministic JavaScript "gates" to control the investigation scope, providing the AI agent with highly constrained "briefs" instead of allowing full repository access. It uses clear boundary markers and specific headers to isolate processed content.
[EXTERNAL_DOWNLOADS]: The skill references Vercel and framework documentation. These references are strictly validated against a curated allow-list in references/docs-library.json. Sanitizers automatically strip any unauthorized or version-mismatched URLs before recommendations are presented to the user.
[DYNAMIC_EXECUTION]: The skill executes local Node.js scripts to perform complex analysis and report rendering. This execution is limited to the skill's own code and is used to provide deterministic thresholds and verification that the AI agent cannot bypass.

vercel-optimize