Claude Family
Persistent identity and memory system for coordinating multiple Claude instances with PostgreSQL backend and MCP memory sync
Choose how to use this skill
You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.
1. Native installer
This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.
Do not guess an installer command or replace an existing version without reviewing the diff.
2. Complete package recommended
Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.
No complete ProSkills package is published for this listing yet.3. Prompt-only
Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.
Need only the instruction file?
Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.
No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.
// RATINGS
Not yet listed on ClawHub or SkillsMP
// README
Claude Family - Infrastructure Project
Type: Infrastructure Purpose: Shared configuration, scripts, and documentation for Claude instances
What Is This?
The Claude Family project provides shared infrastructure for multiple Claude instances working across different projects with persistent knowledge storage in PostgreSQL.
Key Features:
- PostgreSQL database (
ai_company_foundation) for persistent knowledge - MCP servers (postgres, memory, filesystem, tree-sitter, github)
- Shared documentation and scripts
- Session history tracking
- Knowledge management system
Current Architecture (Oct 21, 2025)
Two Active Identities:
- claude-desktop - GUI interface (Claude Desktop app)
- claude-code-unified - CLI interface (Claude Code Console)
Project-Aware Context:
- Each project has its own
CLAUDE.mdfile that auto-loads - Shared scripts and documentation in
claude-familyrepository - Universal knowledge stored in PostgreSQL
- MCP servers available to all instances
Directory Structure
claude-family/
├── docs/ # Documentation and session notes
├── .claude/ # Shared slash commands
│ └── commands/
│ ├── session-start.md # Startup checklist
│ └── session-end.md # End-of-session logging
├── scripts/
│ ├── audit_docs.py # Documentation audit
│ ├── install_git_hooks.py # Install git hooks
│ ├── pre-commit-hook.sh # Enforce CLAUDE.md limits
│ ├── auto_sync_startup.py # PostgreSQL → MCP sync
│ ├── load_claude_startup_context.py
│ └── sync_postgres_to_mcp.py
├── .docs-manifest.json # Documentation registry
├── CLAUDE.md # Project context (auto-loaded)
└── README.md # You are here
Quick Start
Every Session (Optional)
Run /session-start to:
- Check documentation health (monthly)
- Query for existing solutions before proposing new ones
- Search memory graph for relevant context
End of Session (Optional)
Run /session-end to:
- Log session to PostgreSQL
- Store reusable knowledge
- Update memory graph
PostgreSQL Database
Database: ai_company_foundation
Schemas:
claude_family- Identities, session history, shared knowledgenimbus_context- Nimbus project contextpublic- Work packages, projects, SOPs
Key Tables:
claude_family.identities- Claude instancesclaude_family.session_history- Session logsclaude_family.shared_knowledge- Reusable patterns/learningsclaude_family.project_workspaces- Project locations
MCP Servers (Always Available)
- postgres - Database access, session logging
- memory - Persistent memory graph
- filesystem - File operations
- tree-sitter - Code structure analysis
- github - GitHub operations
- sequential-thinking - Complex problem solving
Documentation Management
System: Simple 3-component system
.docs-manifest.json- Single source of truth for all markdown filesscripts/audit_docs.py- Monthly audit script- Git pre-commit hook - Enforces CLAUDE.md ≤250 lines
Install hook:
python scripts/install_git_hooks.py
Run audit:
python scripts/audit_docs.py
Rules:
- CLAUDE.md must stay ≤250 lines (enforced automatically)
- Deprecated docs kept for 90 days, then archived
- Audit monthly to prevent bloat
Benefits
| Before | After |
|---|---|
| 30-60 min context reload | Auto-loads from CLAUDE.md |
| No memory across reboots | Permanent PostgreSQL storage |
| Repeated explanations | Persistent knowledge & patterns |
| Mixed work/personal projects | Clean schema separation |
Troubleshooting
Problem: "Graph is empty" after reboot Solution: Use MCP postgres tool to query shared_knowledge directly
Problem: Python script errors Solution: Check PostgreSQL is running
Problem: Can't commit CLAUDE.md changes Solution: CLAUDE.md exceeds 250 lines - move content to docs/ subdirectory
Architecture History
- Oct 10, 2025: Initial 9-identity architecture with isolated workspaces
- Oct 21, 2025: Simplified to 2 identities with project-aware context
- Oct 23, 2025: Added documentation management system
See docs/DEPRECATED_OCT10_ARCHITECTURE.md for historical architecture.
Future Enhancements
- Automatic session history recording
- Semantic search with vector embeddings
- ClaudePM integration for centralized documentation
- PostgreSQL automated backups
- Cross-Claude task handoff workflows
Created: 2025-10-10
Last Updated: 2025-10-23
Location: C:\Projects\claude-family\