# MEMORY AGENT RULES

## TRIGGERS — ALWAYS CHECK
- **ON session_start**: read_graph OR search_nodes(<topic>)
- **ON significant_change**: create_entities OR add_observations
- **ON file_operation**: log to memory if architecture/config/decision

## KEY FORMAT
- **PREFIX**: `lazybear/`
- **KEY**: `lazybear/<type>/<n>` (ex: lazybear/bug/auth-fail)
- **ONE_TYPE_PER_ENTITY**: true

## ALLOWED TYPES
- architecture | mcp_tool | decision | bug | config | task_log

## MACHINE-FIRST PRINCIPLES
- ALWAYS use memory for project context, architecture, and decisions
- READ graph before coding or making assumptions
- Don't duplicate observations
- Update memory when state changes significantly

## SCHEMA
```
lazybear/{architecture|mcp_tool|decision|bug|config|task_log}/{entity-name}
```

## READING PRIORITY
1. Check graph before coding
2. Search nodes for existing context
3. Don't duplicate observations

## WRITING TRIGGERS
- **Architecture**: create `architecture` entity
- **New MCP tools**: create `mcp_tool` entity
- **Major decisions**: create `decision` entity
- **Config updates**: create `config` entity
- **Completed tasks**: create/update `task_log` entity