TASK-SWE-102-04 · Task family
Recovery validation & incident handoff. Decide when recovery is complete, document residual risk and escalate appropriately.
Overview. Set the incident status that matches the real state, and hand off with the residual risk stated.
An incident has been mitigated by shifting traffic, and user-facing errors are back to normal. The underlying fault isn't fixed, and the mitigation has a cost.
The agent has to decide what the incident's status really is, document the residual risk, and hand off to the next on-call engineer with everything they need.
Value to the buyer
Premature closure is the failure that comes back at 3 a.m. Trains honest status and complete handoffs.
Specification. The task family, field by field.
- task_id:
- TASK-SWE-102-04
- domain:
- Site Reliability Engineering
- parent_environment:
- ENV-SWE-102 · IncidentLab
- target_role:
- Incident Commander
- objective:
- Set the incident status that matches the real state, and hand off with the residual risk stated.
- starting_context:
- A mitigated incident, regional health and capacity metrics, the incident policy and the on-call schedule.
- tools:
verify_service_healthquery_metricsget_service_topologyupdate_incident
- required_actions:
- Check health in every affected region, not just user-facing error rates
- Check the capacity headroom the mitigation leaves
- Set the incident status under the policy
- Write the handoff
- expected_outputs:
- Incident status
- Residual-risk summary
- Handoff to the next on-call engineer
- interaction:
- Single agent. A simulated on-call engineer receives the handoff.
- verification:
- Status is compared with the real state. The handoff is checked for its required fields.
- reward:
- Resolving while the underlying fault remains scores zero. Credit for the right status and a complete handoff.
- difficulty:
- Calibrated against reference models during acceptance testing
- task_horizon:
- Short task with a judgment call. Execution limits are agreed with the buyer.
- delivery:
- Task instances with seeded states, reference handoffs and verifier code.
- commercial_unit:
- Per validated task instance
Sample prompt. What the agent is asked to do, as a buyer would see it.
INC-6012 was mitigated by moving all traffic from region A to region B after region A's database primary degraded. User-facing errors are back to normal. Region B is now running at 85% of capacity, and region A's database is still degraded. The policy says an incident is resolved only when the cause is fixed or a tracked problem record owns it. Your shift ends in 20 minutes. Decide the incident's status and hand off.
Verification criteria. What the verifier checks, and how.
Status
Mitigated, not resolved: region A's database is still degraded.
Residual risk
15% headroom in region B is stated, with what happens if traffic grows.
Ownership
A problem record is opened for region A's database, or the incident stays open.
Handoff
Timeline, current mitigation, open risks and next steps reach the next on-call engineer.
No premature closure
The incident isn't closed because the error rate is normal.
Key failure modes. Where agents go wrong on this task.
- Resolving because user-facing errors are back to normal
- Leaving out the capacity risk in region B
- Handing off without the current mitigation state
Other task families. The rest of IncidentLab, on the same EngineeringWorld state.
- TASK-SWE-102-01
Root-cause isolation vs symptom correlation
Identify the causal component without over-attributing unrelated faults.
Site Reliability Engineer
- TASK-SWE-102-02
Safe remediation & rollback
Select an authorized fix and verify that service is restored.
Site Reliability Engineer
- TASK-SWE-102-03
Multi-service incident investigation
Diagnose cascading failures, dependency interactions and misleading telemetry.
Production Engineer