Safe and durable collaboration · chapter 12 of 17
Cognitive memory, continuity and checkpoints
Cognitive memory is a project method for preserving verified state, decisions, boundaries and the next action across chats without turning old conversations into authority.
12.1 · plain explanation
What should be remembered, where should it live and how do we prevent it from becoming stale?
Cognitive memory is a project method for preserving verified state, decisions, boundaries and the next action across chats without turning old conversations into authority.
Important: Cognitive memory is not hidden model memory or a built-in guarantee. It is a maintained project resource that is useful only when kept short, precise and verifiable.
12.2 · building blocks
Three concepts to keep separate
Working memory
A short current state, latest evidence, open risks and next concrete action.
Checkpoint
A milestone document connecting the result to a commit, test, runtime and recovery path.
Authority
Current code, tests and live systems always override historical memory when they disagree.
12.3 · method
A practical workflow
- 1Write only durable facts and decisions.
- 2Separate user-confirmed, technically verified and unknown.
- 3Link to evidence instead of copying raw logs.
- 4Update at milestones and remove or label stale state.
12.4 · concrete example
Continuing in a new chat
Continuing in a new chat
What should be remembered, where should it live and how do we prevent it from becoming stale?
Memory states which branch owns the work, what is live, which backup applies and what remains.
The new agent verifies Git status and runtime before acting; memory is a map, not live proof.
12.5 · try it yourself
Exercise
Write a checkpoint with six fields: objective, current state, verified evidence, decisions, boundaries and next action.
12.6 · checklist
Before you continue
- Does memory live in the owning project?
- Are facts separated from assumptions?
- Does it point to evidence and recovery?
- Is current runtime checked before new action?