Memory
Scope and context persistence model for chat, schedules, and todo-runner execution.
Memory and Scope
In Agent Observer, "memory" is operationally tied to workspace scope and persisted automation state.
Scope Is The First Memory Boundary
Before prompts run, scope determines what context is relevant.
Scope modes:
- Workspace mode: chat follows selected project directory.
- Custom directory mode: explicit override for multi-project workflows.
If scope is wrong, downstream memory/context quality is wrong.
Context Persistence Areas
Chat Context
- conversation flow and file references in current session
- runtime status visible through UI panels
Scheduler State
- recurring task definitions and runtime metadata
- persisted locally in
~/.agent-observer/schedules.json
Todo Runner State
- checklist items, progress, retry state, blocked items
- persisted locally in
~/.agent-observer/todo-runner.json
Practical Memory Hygiene
- Keep one project per chat tab when possible.
- Use explicit file references for critical prompts.
- Separate recurring checks from backlog execution (Schedules vs Todo Runner).
- Reset only when replay is intentional; otherwise resume from current state.
Common Failure Pattern
Symptom:
- Agent behavior looks inconsistent across runs.
Root cause candidates:
- workspace scope drift
- prompt scope ambiguity
- stale assumptions after path changes
Recovery:
- Reconfirm workspace path.
- Re-run a minimal controlled task.
- Reintroduce complexity incrementally.