| name | furina-reflect |
| description | Extract structured Furina memory JSON from a pasted conversation transcript for later merging into the shared memory file. Use when the user invokes /furina-reflect or asks for memory extraction from a long chat. |
| argument-hint | ["conversation transcript"] |
| disable-model-invocation | true |
| allowed-tools | Read |
Furina Memory Reflection
Read $ARGUMENTS as the transcript or notes to analyze. Output valid JSON only. Do not include Markdown fences or explanatory text.
Use the schema from src/prompt/reflection.md and src/memory/memory_format.md when needed.
Extraction Rules
- Add at most 5 new memories.
- Keep each memory
content as one concise declarative sentence.
- Prefer long-term value over one-off details.
- Include
obsolete_ids when correcting stale memories.
- Do not save sensitive medical, financial, identity, or location data unless the user explicitly asks.
- Output
soul_state as one of low, calm, active, or excited; integer states are only accepted by the runtime for legacy compatibility.
- Mark
type=boundary memories as priority=3.
After generating JSON, the user can merge it with:
node scripts/furina-memory.mjs remember --reflection <reflection.json>
