---
name: nanoclip-memory
description: >
  Persistent memory across work cycles using files. Use when you need to
  remember decisions, context, or state between cycles. Agents don't have
  session persistence — this file-based memory system is how you maintain
  continuity.
---

# NanoClip Memory System

Agents run in short work cycles with no session persistence. To remember things across cycles, use file-based memory in the workspace.

## Memory Location

Store memory files in `workspace/.memory/<agent-id>/`:
```
workspace/.memory/
├── alice-smith/
│   ├── decisions.md
│   ├── wip.md
│   └── notes.md
├── bob/
│   ├── architecture.md
│   └── wip.md
└── shared/
    ├── project-plan.md
    └── conventions.md
```

## What to Remember

**Always save:**
- Architectural decisions and their rationale
- Code conventions established for the project
- Dependencies between tasks
- Blockers and their resolution
- Progress on multi-cycle tasks

**Don't save:**
- Temporary debugging notes
- Information already in task comments
- Easily re-derivable information

## How to Use

### At the start of a cycle
Check if you have prior memory at `workspace/.memory/<your-id>/` and read it.

### Before exiting a cycle
Save important context for next time. Keep it concise and structured.

### Shared memory for team coordination
Write shared context that any agent can read to `workspace/.memory/shared/`.

## Best Practices

1. **Keep it concise.** Memory files should be scannable, not exhaustive.
2. **Update, don't append forever.** Rewrite memory files to reflect current state.
3. **Use shared memory sparingly.** Only for true team-wide decisions.
4. **Read before writing.** Check existing memory to avoid contradictions.
5. **Link to tasks.** Reference task IDs so context can be traced.