Use Case

Scanner evidence.
Not a certificate.

This page explains the registry, not the agent directory. gitleaks, Semgrep, and Trivy examine pinned source. A failed or incomplete scan never becomes a pass. A CLEAN result is not permission to deploy.

The Problem
  • ✗ NPM packages with known-malicious transitive dependencies
  • ✗ Hardcoded API keys and secrets in agent code
  • ✗ Over-permissioned tool bindings that exfiltrate data
  • ✗ Deprecated registries pointing to compromised packages
  • ✗ No CVE history on modules from GitHub repos
What the scanners record
  • ✓ gitleaks scans mirrored source for exposed secrets and credentials
  • ✓ Semgrep checks agent-specific unsafe code and tool-call patterns
  • ✓ Trivy checks dependencies and source for known vulnerabilities
  • ✓ Provenance review records source, revision, content-hash, license, and evidence gaps
  • ✓ Verdicts distinguish CLEAN, FLAGGED, ERROR, QUARANTINE, and incomplete scans

Related reading, not products

TIER 3Dependency-Auditor

Scans lockfiles and transitive dependencies against our CVE database. Identifies known-malicious packages, deprecated registries, and supply-chain anomalies before they reach production.

CVE lookup + transitive risk scoring
TIER 4Security-Auditor

Static analysis using gitleaks, Semgrep, and Trivy. Surfaces exposed credentials, unsafe code patterns, and vulnerable dependencies; human review adds context for the paid assessment.

gitleaks + Semgrep + Trivy + human review
TIER 4MCP-Audit

Audit framework specifically for Model Context Protocol servers. Maps tool permissions, auth flows, and data exfiltration paths in MCP server implementations.

MCP-specific permission analysis

What the active audit actually runs

Tier-2 verdicts combine the scanner evidence available for each source. Entries without sufficient evidence remain visibly pending or incomplete.

gitleaks
Hardcoded credentials, API keys, tokens, private keys. Custom ruleset tuned for agent code paths.
semgrep
Static analysis against custom agent-rules — prompt injection sinks, exfil channels, unsafe tool calls.
trivy
Dependency and source vulnerability scan using Trivy's current advisory database, with recorded severity and fixed-version fields where available.
provenance review
Source path, captured commit, exact content hash, license evidence, and scanner-evidence gaps are reviewed explicitly.
verdict engine
Combines available scanner output into CLEAN, FLAGGED, ERROR, QUARANTINE, or INCOMPLETE status. Each record shows the evidence and timestamp currently stored.

Tier-2 audit pipeline active across the public registry.

Free registry. Optional scoped review.

Free — no signup
Browse the full registry, inspect recorded verdicts and evidence gaps, and follow upstream source before evaluating anything in your own environment.
Scoped Repository Review — Request-Based
Human interpretation of available scanner and provenance evidence under a written scope agreed before work begins.

Working with an authorized repository and need help interpreting the available evidence? Send the source and the decision you need help making. Do not send credentials by email.

Request a scoped review →