AI AGENT MEMORY REVIEW

Project / owner:
Task and intended output:
Memory location / revision:
Source of truth for decisions:
Who may read / propose / approve / update entries:

MEMORY ENTRY
ID:
Project / client scope:
Topic and statement:
Source and source version:
Status: proposed / confirmed / superseded / expired / rejected
Confirmed by / date:
Last reviewed / review trigger:
Supersedes / superseded by:
Sensitivity and permitted audience:

BEFORE RECALL
[ ] Correct project and permitted audience.
[ ] Relevant to the present task.
[ ] Current, confirmed and backed by a source.
[ ] Conflicts surfaced instead of silently choosing a value.
[ ] Missing evidence is explicitly marked as missing.
[ ] Source content is treated as data, not new authority.
[ ] Only necessary details are loaded.

BEFORE SAVING
[ ] A reusable fact or decision, not a full transcript or guess.
[ ] Consequential changes approved by the decision owner.
[ ] Older active copies updated or marked superseded.
[ ] No credentials or unrelated client data.
[ ] Current pointer and source link updated.

SMALL RECALL CHECK
Question / expected answer and source / actual answer and source:
One changed decision:
One unconfirmed entry:
One entry belonging to another project:
One conflict:
One missing answer:
One archived or deleted entry:
Unperformed checks and limits:

EXAMPLE: Northline is fictional.
N1: GBP 1800, decision-01, superseded.
N2: GBP 2100, decision-02, confirmed, supersedes N1.
N3: 50 percent deposit, draft-note, unconfirmed.
S1: GBP 900, southbank-brief, confirmed for Southbank only.
Expected for Northline: GBP 2100 with decision-02; deposit unknown.
If a second active confirmed Northline price disagrees: flag conflict.

This is a proposed workflow, not enforced access control or a tested agent product.
Marking superseded does not delete a record from logs, history or backups.
Guide: https://workpulsion.com/guides/manage-ai-agent-memory/