Start from the exact API error or Attention remedy; do not diagnose from a summary. Queued work, a pending approval, a live lease, or a temporarily blocked dependency is not by itself a fault. Workforce polls queued work, bounds eligible local execution, and re-evaluates dependency recovery without a manual repair command.
If the actions below do not resolve the failure, record the Workforce version, topology, route, HTTP status and error code, Project/Issue/WorkUnit/Run IDs, and correlation ID. Remove tokens, credentials, message content, and secret-bearing URLs before sharing the evidence.
| Symptom | Inspect | Likely action |
|---|---|---|
Process exits before /healthz becomes available | stderr and the exact configuration field named in the startup error | Correct the named seal-key, database, or schema condition; use the deployment topology guide rather than changing stores directly. |
Dispatch returns execution_gated and its detail names a missing Agent revision, model route, provider, credential, or Resource | The returned detail and the Project’s execution-readiness response | Complete the missing governed join, then retry the same public command. Do not bypass admission with a database write. |
Issue shows an open Attention signal with a concrete reason_code and remedy | The signal and the current authoritative resource or configuration state | Follow Attention recovery; resolve the signal only after the named condition is repaired. |
A transition request returns no_transition | Current Issue state, submitted event, and the pinned Workflow revision | Submit an event declared from that state, or save a reviewed Workflow revision for future Issues. Do not infer or patch an edge. |
Use /healthz for liveness, /metrics for Prometheus counters, and
/api/openapi.json for the exact route contract of the running binary.