blackboard

Memory

Search first: /mnemonic:search {relevant_keywords} Capture after: /mnemonic:capture {namespace} "{title}"

Run /mnemonic:list --namespaces to see available namespaces from loaded ontologies.

Mnemonic Blackboard Skill

Cross-session handoff, persistent context, and agent coordination.

Trigger Phrases

• "blackboard", "cross-session", "session handoff", "shared context"

Coordination Overview

Scope	Mechanism	Tools
In-session	Claude Code native swarm	TeamCreate, SendMessage, TaskCreate, TaskUpdate
Cross-session	Mnemonic blackboard handoff	handoff/latest-handoff.md via hooks
Persistent knowledge	Mnemonic memories	*.memory.md files

---

Path Resolution

MNEMONIC_ROOT=$(tools/mnemonic-paths root)
BLACKBOARD=$(tools/mnemonic-paths blackboard)
HANDOFF_DIR=${BLACKBOARD}/handoff

---

Directory Structure

${BLACKBOARD}/
├── _legacy/                      # Pre-migration files (frozen)
├── sessions/
│   └── {session_id}/
│       ├── session-notes.md      # Session activity log
│       └── _meta.json            # Session lifecycle metadata
├── handoff/
│   ├── latest-handoff.md         # Cross-session context (overwritten each session end)
│   └── handoff-{session_id}.md   # Archived per-session handoffs
└── .archive/                     # Archived entries

---

Cross-Session Handoff

How It Works

1. Session end (hooks/stop.py): Writes summary to handoff/latest-handoff.md
2. Session start (hooks/session_start.py): Reads handoff/latest-handoff.md to restore context

This is automatic via hooks.

Handoff Format

# Session Handoff

**Session:** {session_id}
**Ended:** {timestamp}
**Project:** {org}/{project}

## What Was Accomplished
- Completed items

## In Progress
- Ongoing work with current state

## Blocked
- Blocker: reason

## Next Steps
- [ ] Action item 1
- [ ] Action item 2

## Important Context
- Key context for next session

---

Mnemonic Agents

Agent	Purpose	Invocation
memory-curator	Maintenance, deduplication, decay	Standalone or native swarm
mnemonic-search-subcall	Iterative memory search	Spawned via Task tool
compression-worker	Memory summarization	Spawned via Task tool
ontology-discovery	Entity discovery from codebase	Spawned via Task tool

---

Session Metadata

Each session creates _meta.json:

{
  "session_id": "abc-123",
  "started": "2026-01-15T10:00:00Z",
  "ended": "2026-01-15T11:30:00Z",
  "project": "zircote/mnemonic",
  "org": "zircote",
  "status": "ended"
}

---

Best Practices

• Native swarm for in-session work — don't write coordination state to blackboard
• Let hooks handle handoff — session lifecycle hooks manage cross-session context automatically
• Memories for persistence — if something should survive beyond the next session, capture it as a memory
• Keep handoffs actionable — focus on what the next session needs to continue work