Turn Claude Code into a personal AI assistant — with persistent memory, seamless session chaining, and structured workflows.
Claude is already smart. What it lacks is structure, memory, and habits.
Decisions, context, and preferences persist across sessions — not just within one.
Skills encode your best workflows. No more re-explaining the same steps every time.
The Evaluator-Optimizer pattern catches issues that single-pass generation misses.
Claude's context window is finite. Session Chaining makes it infinite — your state is saved automatically and the next session picks up where you left off.
Use claude-loop as a drop-in replacement for claude. When context gets critical, your work is preserved and the next session continues seamlessly.
# Just use claude-loop instead of claude claude-loop # When context fills up: # → State saved automatically # → "New session with context? [Y/n]" # → Press Y — picks up where you left off
Your message flows through four stages — from intent to structured output.
Context that persists — from one conversation to months of knowledge.
Current conversation and temporary results. Lifetime: 1 session.
Past session handoffs, routing decisions, and project states. Lifetime: weeks to months.
Routing rules, preferences, glossary, contacts, and domain knowledge. Lifetime: permanent.
Reusable workflows defined as markdown files — instructions, not code.
Save notes and ideas quickly. Works with Obsidian, local files, or inline.
Progressive summarization. Extract key takeaways from any text.
Generate polished, structured content — articles, posts, emails, concepts.
Structured ReAct analysis. Investigate topics with iterative reasoning.
4-axis substance analysis: meaning, facts, framing, context. Signal vs. noise.
Adversarial testing from 3 perspectives. Find weaknesses, blind spots, and hidden assumptions.
Output quality orchestration with triage levels, external verification, and quality scoring.
Save decisions, context, and next steps. Next session picks up seamlessly.
Skills chain together into powerful workflows.
Write, evaluate, improve — like a human review cycle.
Write, attack, harden — adversarial stress-testing.
Write, verify with triage, revise — structured quality control.
Capture, process, output, verify — continuous learning.
The starter kit gives you the building blocks. The full system goes much deeper.
| Dimension | Starter Kit | Full System |
|---|---|---|
| Skills | 7 core skills | 20+ domain-specific skills |
| Routing | 7 keyword mappings + mode detection | 100+ mappings with disambiguation |
| Memory | Template files | Months of accumulated context |
| Obsidian | 1 vault (optional) | 6 connected vaults, 930+ notes |
| Database | Empty schema | Research items, datasets, usage stats |
| Workflow Chains | 3 feedback loops | 15+ multi-step chains |
| Self-Improvement | Manual updates | Automated routing analysis |
# Clone the repo git clone https://github.com/janrummel/claude-orchestrator-starter.git cd claude-orchestrator-starter # Copy to your Claude config cp CLAUDE.md.example ~/.claude/CLAUDE.md cp -r orchestrator/ ~/.claude/orchestrator/ cp -r memory/ ~/.claude/memory/ mkdir -p ~/.claude/orchestrator/projects # Set up hooks (statusbar, context monitor, session start) cp hooks/session-start.sh.example ~/.claude/hooks/session-start.sh cp hooks/context-monitor.js ~/.claude/hooks/context-monitor.js cp hooks/context-statusline.js ~/.claude/hooks/context-statusline.js cp hooks/settings.json.example ~/.claude/settings.json chmod +x ~/.claude/hooks/session-start.sh # Optional: install session chaining wrapper cp hooks/claude-loop.sh ~/.local/bin/claude-loop chmod +x ~/.local/bin/claude-loop # Start Claude Code claude-loop
All memory files, routing logs, and project states live in ~/.claude/. No server, no cloud upload, no analytics.
Everything is markdown. Export, backup, version-control, or delete anytime. You own your data.
The core kit needs zero API keys. Only Claude Code itself requires an Anthropic account.