Decide whether work needs intervention | Awaken Workforce documentation

Read /api/issues/{id}/diagnosis. It is the single Issue-level view of scheduling, active Attention, the active WorkUnit, and public recovery actions.
Do nothing when scheduling is ready or running. A queued WorkUnit is polled, and a live local execution is bounded by the reaper. Do not edit a lease or dispatch the same Issue again.
For blocked_by_dependency, inspect the named blocker. Closing it causes Workforce to re-evaluate dependants through the resident pump; there is no manual wake command. A pending approval is also an explicit decision state, not a stalled run.
Intervene only when the response names an external correction:
- for attention, read /api/issues/{id}/attention-signals and follow its exact reason_code and remedy;
- for waiting_on_resource, wait until retry_at when it is present; when it is absent, restore the named Resource and its reported reason_code;
- for an Agent placement failure, read /api/agent-fleet/readiness and restore the missing capability or Worker connection;
- for execution_gated, complete the missing Agent, model, provider, credential, or Resource binding named in the error detail.
If a WorkUnit is already failed, read its /events and /state, then follow the Attention signal Workforce raised for that terminal outcome. Do not revive the lease or rewrite WorkUnit state.
Use /message, /pause, /resume, /interrupt, /redirect, or /cancel only for an intentional control decision on the exact WorkUnit, not as a generic retry mechanism.

There is no current doctor command. If an explicit condition still cannot be corrected, capture the diagnosis response, relevant events, reason code, Worker identity, and correlation IDs. Remove credentials, message content, and secret-bearing URLs before sharing them.