The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it ingests local source code and external Vercel metrics to dynamically construct investigation briefs for sub-agents.
Ingestion points: scripts/scan-codebase.mjs reads the project's source files, and scripts/collect-signals.mjs fetches usage and performance data from Vercel APIs.
Boundary markers: Strict adherence to 'Doctrine Rule 3' (Candidate-bound scope) ensures sub-agents are restricted to reading only the files explicitly identified by metric signals, preventing unauthorized repository-wide scans.
Capability inventory: Scripts within the lib/ and scripts/ directories execute the Vercel CLI and Ripgrep for data collection and verification.
Sanitization: A multi-layered validation system (found in lib/sanitizers/ and lib/verify-claim.mjs) mechanically verifies all recommendations against the filesystem and curated documentation before they are included in the report.
[COMMAND_EXECUTION]: The skill executes trusted CLI utilities to perform its analysis.
Vercel CLI: Used extensively to collect project configuration, billing usage, and route-level metrics from Vercel's platform.
Ripgrep (rg): Utilized by the verification engine to confirm the presence of code patterns and snippets cited in generated recommendations.
[EXTERNAL_DOWNLOADS]: Fetches metrics and configuration data from well-known service endpoints.
Uses the Vercel CLI to download telemetry and billing data from the api.vercel.com domain using the user's authenticated session.

vercel-optimize