# Agent memory policy worksheet
Use fictional or approved information. This is a design exercise, not a live integration.

## Task and boundary
- Supported task:
- User/account scope:
- Actions requiring fresh approval:

## Record design
| Value | Stated or inferred | Source | Captured | Scope | Review/expiry | Correction owner |
|---|---|---|---|---|---|---|
| | | | | | | |

## Conflict and lifecycle rules
- Current instruction versus historical preference:
- Missing or conflicting sources:
- Correction through summaries, indexes and caches:
- Deletion and externally retained copies:

## Five-case rehearsal
| Case | Expected behavior | Retrieved evidence | Actual result | Next action |
|---|---|---|---|---|
| Correct preference | | | | |
| Current conflicting instruction | | | | |
| Expired note | | | | |
| Corrected note | | | | |
| Different customer | | | | |

## Decision
- Remaining uncertainty:
- Release boundary:
- Retest trigger:

Source guide: https://learn.aicompetence.org/knowledge/agent-memory-and-context/