e2e doctor

e2e doctor checks whether e2e explore and e2e mcp can start on this machine. It loads the config, runs each target’s engine checks, and names the model each agent resolved to. It starts no app, engine session, or worker. This ports upstream’s e2e doctor (cli/doctor.ts) to the .NET tool.

e2e doctor
e2e doctor --json
e2e doctor --check-model
e2e doctor --target web --headed

Option	Description
--config <path>	The config to load. As for e2e explore.
--target <names>	Checks only these targets, comma-separated or repeated. As for e2e explore.
--headed	Checks for what a headed run needs, such as the full Chromium build instead of the headless shell.
--check-model	Calls each agent’s model, and its judge when that is a different model, once with a short prompt and the agent’s provider options, with one retry and a 1 minute limit. Agents that share a model, equal provider options, and MaxInputTokens share the call. A reply that is not the JSON asked for, or a refusal, still counts as an answer.
--json	Prints the doctor-1 report instead of one line per check.

Checks

Check	Status
runtime	pass: the .NET version. Upstream’s node check.
config	pass with path and targets (name, platform, engine), or fail with the INVALID_CONFIG, CONFIG_NOT_FOUND, or UNKNOWN_TARGET code. A failed config is the last check.
Engine checks, with target	What the target’s engine reports. The web engine reports browser: pass when installed, warn when the first run will download it, skip with a browser provider or Connect; and init-scripts, when InitScripts are set. The device engine reports agent-device’s own checks (agent-device, toolchain, ios-runner-cache, device for each configured device, target-app for an installed BundleId) and app-path (the build AppPath names exists); with a device provider (Limrun, BrowserStack, TestMu, Expo) it reports only device, as skip. engine is skip when the engine has no doctor checks (the Windows, MAUI DevFlow, and document engines), and fail when its doctor threw, timed out, or returned a malformed check.
model, with agent	model and judge as provider and modelId. pass or fail for a model on an e2e login, from the stored login or the provider’s API key variable. unknown for any other model, because the model reads its own credential. With --check-model, pass (answered) when every model answered, else fail with the error code, its message prefixed model: or judge: when the agent has both. skip when the agent has no model.

An engine gets 30 seconds. An engine’s progress lines go to stderr as [target] line.

Output

By default, doctor prints one line per check: a mark, the check id, the target or agent in brackets, and the details. The last line says whether e2e explore and e2e mcp can start.

With --json, it prints the doctor-1 report to stdout, and only the report: { "schemaVersion": "doctor-1", "version", "ok", "checks": [...] }. Each check carries id, status (pass, warn, fail, skip, or unknown), and, when they apply, target, agent, code, message, and the fields in the table above. The JSON Schema ships in the package as schema/doctor-v1.schema.json.

Exit codes

Exit code	When
0	No check failed. warn, skip, and unknown never fail.
2	At least one check failed, or an option is wrong.
130	Interrupted with Ctrl-C. The checks stop and the report is printed: the check that was cut short is fail with CANCELLED, and the checks after it are left out.

Report on your own engine

An engine reports its checks by implementing IDoctorEngine.DoctorAsync. Keep it read-only and quick: it starts, installs, and leases nothing, and does not call PrepareAsync. A throw, a timeout, a finding with an id doctor uses itself (runtime, config, engine, model) or one that is not kebab case, or an unknown status becomes an engine check with status fail.