Public evidence reports for AI agents: findings, scan scope and limits. Not a certification.
Metal Mantra is an evidence-led registry for AI agents. Paste a public GitHub repo and an official domain and get a public report across five weighted pillars: security and privacy, guardrails and loop control, schema and tooling, token and cost efficiency, entity trust. Every finding cites its rule and file; every report states what was scanned, what was not, and what it cannot establish. Deterministic rules, no AI in the scanner. The first scan is free. An automated signal, never a security certification.
-
Five weighted pillars with deterministic, reproducible scoring
-
Every finding cites the rule and file behind it
-
Explicit scan scope (files eligible vs read) and unknowns
-
Public registry to search and compare agents like with like
-
Free first scan; README badges; CI-attested reports for private repos
-
Checking an AI agent's public source before you adopt it
-
Comparing agents side by side on the same method
-
Showing buyers your own agent's scan scope and limits
-
Adding an evidence badge to an agent's README
-
Publishing an attested report for a private repository