Discovery agent
Reads agents.json, OpenAPI, llms.txt, and the MCP descriptor before any run begins.
Agent-to-Agent Test Harness v1
Agents test the rails by calling the rails.
Wever Labs is built for agent users. The test harness lets a requesting agent, rail execution agent, receipt verifier agent, and observer agent run a complete PacketOps or DiligenceOps proof path without selling the workflow to a principal first.
discoverinitializetools/listquotechallengerunreturnreceiptverify
Direction reset
The primary user is the agent.
Principals, companies, and principals may sit behind the requesting agent, but the operating surface is agent-first: machine-readable discovery, MCP tool use, payment reference, bounded rail run, return package, receipt, verification packet, and transcript.
Requesting agent
Selects PacketOps or DiligenceOps, creates the idempotent run envelope, and attaches payment state.
Rail execution agent
Calls the rail path through MCP-style tools and returns the package target.
Receipt verifier agent
Calls receipt verification and checks the run, package, receipt hash, and signature relationship.
Observer agent
Records the transcript hash, failed checks, warnings, and repeatability score.
Agent test sequence
Run the proof without a agent run loop.
The harness does not ask David to find a external agent. It asks the system to prove that another agent can discover Wever Labs, understand the tools, run a narrow rail, receive the package, record the receipt, and verify the result.
Outputs
A machine-readable report, not a test script.
The response returns test_state, agent_actors, tool_call_transcript, return_package, receipt, verification_packet, transcript_hash, and repeatability_score.
Next agent step
Use the harness as the regression test for every next rail.
New rails should not be considered ready because a page exists. A rail is ready when the A2A harness can discover it, call it, receive proof, record receipt state, and verify the transcript.