A clearer view of agent evidence.
Metal Mantra is a public registry of automated signals for AI agents. It helps readers inspect a source snapshot, understand notable findings and see the limits of a scan before trusting a score.
Evidence before assurance
Every report starts with a specific public GitHub repository and a scan method. The result links findings to source locations when available, records its scope and timestamp, and separates code checks from domain-control evidence.
What we do not claim
A score is not a security certification, a proof of runtime behavior, a guarantee that an agent is safe, or an independent audit. DNS verification indicates control of the named domain, not ownership of the repository or security of the code.
A registry you can question
Find a public report, inspect its method and evidence, then use the support form if you believe a finding, repository claim or contact detail needs review. Archive v1 and Agent Signal v0.2 use different methods; their scores are not comparable.
Who builds it
Metal Mantra is built and operated by Danhoangda, an independent AI lab in Hanoi, Vietnam, which also builds HowAI Suite. Questions about a report go through the support form; to learn about the lab itself, visit its site.