Proof Layer

Cryptographic audit trail for AI agent actions, issued before execution rather than after. Exposes eight MCP tools (prefixed `pl_*`) that let Claude request governance verdicts (EXECUTE, BLOCK, REVIEW, SHADOW) from a remote evaluation service, then write Ed25519-signed, hash-chained receipts to a local audit log. Ships with an offline verifier that checks signatures and chain integrity using only a public key, no API calls required. Reach for this when you need tamper-evident proof that an agent requested permission before acting, especially in regulated environments or multi-agent systems where you can't trust post-hoc logs. The MCP server and verifier are MIT-licensed, the backend governance engine is closed-source SaaS.