Metal Mantra: Public evidence reports for AI agents

Hi Product Hunt! I built Metal Mantra because every "AI agent directory" I found ranked agents by who pitched best, not by what the code actually does. Paste a public GitHub repo and you get a public report: findings with their source, the scan scope (how much we read, and what we could not), and domain evidence. The method is five weighted pillars and fully deterministic. There is no AI in the scanner, so the same input gives the same result. (AI appears in one optional place only: ranking agents against a buyer's description. It never touches a score.) Two rules I will not break: a score is never a security certification, and every report says what it does not establish. The first scan of a repo is free. If you maintain an agent, I would love your feedback on the method, especially rules that feel too harsh or miss something real. Try it: https://metalmantra.io/scan