# Memory Bank v2 Operations

## Command: `update memory bank`

Use this when significant changes happen or context gets stale.

### Procedure
1. Read Layer 0 (`00_project_snapshot.md`, `20_now.md`, `21_backlog.md`).
2. Read additional Layer 1 and Layer 2 files that are relevant to touched work.
3. Refresh core files to current truth.
4. Write or update ADRs for durable decisions.
5. Write a task log for significant implementation steps.
6. Validate:
   - All referenced ADR and task IDs exist.
   - Core files do not contain changelog-style history.
   - Budgets from `SCHEMA.md` are respected.
7. Update `Last Updated` fields in touched files.

### Expected Result
- Layer 0 gives an accurate project state in minutes.
- Decision and execution history is traceable via IDs.

## Command: `compact memory bank`

Use this manually for maintenance.

### Triggers
- Any file exceeds hard budget.
- Backlog has many stale completed items.
- Task logs older than 90 days are accumulating.

### Procedure
1. Identify oversized core files.
2. Remove historical narrative from core files.
3. Ensure old details live in ADR/task logs.
4. For each month older than 90 days, create or refresh `logs/digests/YYYY-MM.md`.
5. Replace verbose old references in core files with digest links.
6. Re-validate ID/link integrity and budgets.

### Expected Result
- Core remains concise.
- Older history stays discoverable through digests and linked logs.

## Consistency Checklist
- Every backlog item has an ID, status, and owner.
- Every ADR has status (`proposed`, `accepted`, `superseded`) and links.
- `20_now.md` and `21_backlog.md` do not conflict.
- `00_project_snapshot.md` reflects the same priorities as `20_now.md`.