Open protocol for AI agent work contracts and verifiable execution. Ed25519-signed authorization, causal evidence chains, offline verification. 2,281 tests. Apache-2.0. MCP connects agents to tools. A2A connects agents. But when your agent says "I ran the tests," nobody can verify that claim at the protocol level — until now. OpenWorkProof is a middleware layer that records every authorization decision, execution receipt, and evidence publication in a cryptographically signed chain.

I built this after running into the same problem repeatedly: agent A says "I ran the tests," and agent B has no protocol-level way to verify that claim. MCP and A2A solve connectivity, but not accountability. Every tool call in OpenWorkProof carries a signed authorization decision, execution receipt, and evidence publication. The entire chain can be replayed offline by a third party with just the evidence bundle and public keys. The protocol is framework-agnostic. It works as a middleware layer that sits between any agents and their tools. 2,281 tests, 2 real-world bugs demonstrated end-to-end (Rich #4196, Dify #33013), Apache-2.0. Would love to hear what verification challenges you've hit in your own agent setups.

OpenWorkProof

Cryptographic proof that your AI agent actually did the work

Cryptographic proof that your AI agent actually did the work