Feature
Give an agent one job, the sources and tools you approve and an output format, then decide which of its steps have to stop for a person to confirm.
Updated 21 Aug 2026
Definition
Agent, skill, template and provider configuration surfaces are implemented, with availability controlled by the deployment.
What a deployment actually offers depends on the release you are buying and how it is configured.
Use representative documents and record the expected result before running the test.
Record enough for someone else to run it again: how it was set up, what kind of documents went in, who ran it, where the data went, what happened when it failed and who reviewed the result.
Pick a real piece of work, agree what a good answer looks like, then go through the results together.