How vercel-labs/agent-skills works
This repository packages agent skills; its most substantial executable workflow is the Vercel Optimize audit. The README describes collecting Vercel metrics, investigating only implicated routes and files, and producing a ranked report. The graph follows that workflow and its gate, scanner, and report layers. Other documented skills are represented by instructions and content not present in the supplied tree. Several workflow scripts are unsampled, so their internal calls are not asserted.
Main entry points: gates/index.mjs Candidate gates · collect-signals.mjs Collect signals · render-report.mjs Render report · scanners/index.mjs Code scanners
Components 15
Audit workflow
- Collect signals
skills/vercel-optimize/scripts/collect-signals.mjs
Signal analysis
- Scan codebase
skills/vercel-optimize/scripts/scan-codebase.mjs - Candidate gates
skills/vercel-optimize/lib/gates/index.mjs - Code scanners
skills/vercel-optimize/lib/scanners/index.mjs - Route normalization
skills/vercel-optimize/lib/route-normalize.mjs
Findings and investigation
- Reconcile candidates
skills/vercel-optimize/scripts/reconcile-candidates.mjs - Investigation brief
skills/vercel-optimize/lib/investigation-brief.mjs - Deep dive
skills/vercel-optimize/scripts/deep-dive.mjs - Verify claims
skills/vercel-optimize/lib/verify-claim.mjs
Report and trust
- Sanitize recommendations
skills/vercel-optimize/lib/sanitizers/index.mjs - Deduplicate recommendations
skills/vercel-optimize/lib/dedup-recs.mjs - Render report
skills/vercel-optimize/lib/render-report.mjs - Cost coverage
skills/vercel-optimize/lib/cost-coverage.mjs
Other components
- Agent or user
- Vercel project
Connections 15
- Agent or user → Collect signals: requests audit
- Collect signals → Vercel project: collects metrics
- Collect signals → Candidate gates: provides signals
- Scan codebase → Code scanners: runs scanners
- Code scanners → Candidate gates: feeds findings
skills/vercel-optimize/lib/gates/index.mjs - Candidate gates → Route normalization: normalizes routes
- Candidate gates → Reconcile candidates: produces candidates
- Reconcile candidates → Investigation brief: selects investigations
- Investigation brief → Deep dive: guides investigation
- Deep dive → Verify claims: checks claims
- Verify claims → Sanitize recommendations: supports findings
- Sanitize recommendations → Deduplicate recommendations: filters recommendations
- Deduplicate recommendations → Render report: provides recommendations
- Render report → Cost coverage: computes cost coverage
skills/vercel-optimize/lib/render-report.mjs - Render report → Agent or user: returns ranked report