Verify ageval

Drive uv run ageval. A missing credential or AGEVAL_SKIP_REAL_ACP=1 is not run, never passed. Report format: references/report.md.

Work in a copy of the dataset and a temporary AGEVAL_HOME. Leave the run evidence. Stop only the gateway and viewer processes this procedure started.

Doctor

uv run ageval --version
uv run ageval run examples/datasets/minimal-demo --task tau2-dialog-min --probe

--probe is a flag on ageval run: lock plus preflight, and it does not start an Attempt. The shipped minimal-demo profiles use docker and ACP. Without a docker daemon or an ACP entry, record that probe not run (or failed, if the tool was present and the probe itself failed). Do not treat it as a pass of the loopback path.

Lock rejects a bad profile

Copy examples/datasets/minimal-demo aside. Point a profile at executor: acp with options.entry: pi and options.idle_timeout_secnds: 30. uv run ageval lock must exit non-zero and name idle_timeout_secnds.