| name | fluctlight-memory |
| description | Use FluctlightDB project brains for durable memory, recall, and handoffs between Cursor, Claude Code, and Codex in this monorepo. Call when you need prior decisions, conventions, cross-agent context, or to leave a handoff for another tool. |
FluctlightDB project memory
This repo uses FluctlightDB as the shared project brain ({{project_id}}).
Layout
.fluctlight/project/ — shared hub (decisions, handoffs, conventions)
.fluctlight/agents/<tool>/ — per-agent spoke (session-local notes)
When to use
- Before big changes — recall conventions and prior decisions for the current subdirectory.
- After completing work — write a handoff if another agent may continue.
- When unsure — recall by cue instead of re-deriving from chat history.
Python (embedded)
from fluctlightdb import connect_project
pb = connect_project(agent="claude")
print(pb.session_context())
pb.remember("API uses JWT in Authorization header", scope="project", context="architecture")
hits = pb.recall("authentication conventions", scope="all")
pb.handoff("Finished auth middleware", next_steps=["Add integration tests"], files=["src/auth.py"])
MCP (if configured)
Tools: fluctlight_recall, fluctlight_remember, fluctlight_handoff, fluctlight_list_handoffs, fluctlight_session_context, fluctlight_status.
CLI: fluctlight-project handoffs, fluctlight-project doctor.
Handoff protocol
Handoffs are stored in the project brain and .fluctlight/handoffs.jsonl (deterministic inbox). Use list_handoffs() or fluctlight_list_handoffs — not only fuzzy recall.
Rules
- Prefer project scope for team-wide facts; agent scope for tool-specific session notes.
- Do not store secrets or credentials in brains.
- Keep handoff summaries short and actionable (status, next steps, touched files).