# Agent Incident Review

> Blameless, evidence-linked operating review. Do not include secrets or regulated content in an uncontrolled document.

## Record control
- Incident ID:
- System / workflow:
- Active version:
- Severity:
- Status:
- Incident commander:
- Accountable owner:
- Review date:

## Executive summary
- Intended task:
- What happened:
- Impact:
- Duration:
- Current state:

## Detection and containment
- Detection source:
- Why the issue was not detected earlier:
- Immediate containment:
- Credentials / tools / queues / memory isolated:
- Human takeover or rollback:
- Evidence preserved:

## Versioned timeline
| Time | Agent / person / service | State or message | Model / prompt / policy / tool version | Authorization | Side effect | Evidence link |
|---|---|---|---|---|---|---|

## Failure analysis
- Triggering condition:
- First unsupported state or interpretation:
- Semantic or authority boundary crossed:
- Propagation mechanism:
- Amplification mechanism:
- Final side effect:
- Missing or failed control:
- Contributing architecture, data, process, ownership, or vendor conditions:

## Corrective actions
| Action | Type: contain / remediate / prevent / detect / recover | Owner | Due date | Validation evidence | Status |
|---|---|---|---|---|---|

## Evaluation and system updates
- New or updated test cases:
- Threshold or release-gate changes:
- Agentic System Card changes:
- AIBOM changes:
- Risk Register changes:
- Access / authority changes:
- Runbook and training changes:

## Residual risk and decision
- Residual risk:
- Decision: resume | restricted resume | rebuild | replace | retire
- Approvers:
- Review conditions: