An agent says 'tests all green'. Another agent says 'I reviewed it — it's fine.' Two independent executions, two keys, one verdict — or no decision at all. OpenWorkProof v0.5's dual-verifier cross-validation, built in 7 review rounds against adversarial probes.
Two Verifiers, One Verdict: Closing the "Self-Reported Exit Code" Hole
Repo: github.com/dengyier/OpenWorkProof
This update: 04e62d5..ec0e72e (29 commits across three phases; candidate inventory 837dbb10a3fdc4…)
The problem: "two sub-agents reviewed it" is not independence
In public code reviews of agent tooling, two objections keep coming back:
OpenWorkProof v1.0: verifiable execution for multi-agent systems — Ed25519 signatures, offline verification, 2,283 tests
A sub-agent's success report is generator output, not ground truth. Verify the work yourself, and...
How do you verify an AI agent actually did the work? OpenWorkProof adds Ed25519-signed work contracts, causal evidence chains,…
One night, two audits, one identical score The evening of 17 May, I finish version 0.4.1...
A reader turned my sycophancy gate into an observable test, and the gate failed it — its escape condition was a keyword in the…
The argument, in one line: a reviewer is a mechanism for drawing a line. Every fix moves the line —...
