Verify readiness
Installed is not the same as working. A harness with no provider key — or a key with an empty balance — will happily accept a run and produce no changes. AM gives you two ways to know before you run.
Background probes ("Say HI")
Each worker periodically sends every installed harness a tiny "Say HI" and records whether it answered. The Workers tab shows a chip per harness:
✓ claude— answered; a working provider key is configured.⚠ opencode (out of balance)— installed, but the probe failed, with the reason.
A worker is ready when at least one harness answers.
The Test button (on‑demand, exact model)
Background probes test each harness's default model. Your team might use a specific model that fails for a different reason. In the team builder, click Test next to a role's cli/model to probe that exact combination:
✓ reachable (2598ms)— a real run with this cli/model will launch.✗ out of balance/✗ no/invalid API key/✗ unknown model/✗ not installed/✗ rate limited— the precise reason it won't.
This is the fastest way to answer "why did my run do nothing?" — usually it's a balance or key problem, surfaced here in one click.
Reasons you may see
| Reason | What to do |
|---|---|
| out of balance | Top up the provider, or switch model/provider |
| no/invalid API key | Set/rotate the key on the worker, then restart it |
| unknown model | Fix the model id in the team |
| not installed | Install the harness on the worker |
| rate limited | Wait, or use a different provider |
| timeout / unreachable | Check the worker's network/provider status |