Testing agent-to-agent systems combines API testing with AI evaluation.
Protocol Tests
- Validate Agent Cards against the specification.
- Send tasks and check state transitions, streaming, cancellation and errors.
- Test authentication failures and permission boundaries.
Mock Agents
Build simple mock remote agents that return predictable results, failures or slow responses. They let you test client behaviour reliably.
Delegation Tests
With real models, check that the client agent:
- Chooses the right remote agent for each request.
- Sends the necessary information.
- Handles input-required states correctly.
- Uses results properly.
End-to-End Scenarios
Run realistic workflows across agents, with success criteria such as correct bookings or accurate reports.
Failure Scenarios
Simulate unavailable agents, wrong results and injected instructions in remote outputs.
Contract Testing
When teams own different agents, agree on skill contracts and test both sides against them so changes don't break integrations.
Repeat Runs
Agent behaviour varies; run scenarios several times and track success rates.