# Claude Toolkit
**Modular enhancements for Claude Code** - Use what you need, skip what you don't.
[](LICENSE)
## Philosophy
> Install one plugin or all of them. Each works independently.
This toolkit extracts the best ideas from three projects:
- **[wave-orchestration](https://github.com/sethdford/wave-orchestration)** - Parallel task execution
- **[claude-code-collab](https://github.com/sethdford/claude-code-collab)** - Multi-instance coordination
- **[Continuous-Claude-v3](https://github.com/sethdford/Continuous-Claude-v3)** - Memory, analysis, workflows
## Installation
### Quick Install (Recommended)
**Straight from GitHub — no clone needed:**
```bash
claude plugin marketplace add sethdford/claude-toolkit
claude plugin install compound-loop@claude-toolkit # or any plugin below
# Restart Claude Code to load.
```
**Or clone for local development:**
```bash
# 1. Clone the repository
git clone https://github.com/sethdford/claude-toolkit.git
cd claude-toolkit
# 2. Add as a local marketplace
claude plugin marketplace add .
# 3. Install plugins you want
claude plugin install tldr@claude-toolkit
claude plugin install memory@claude-toolkit
claude plugin install wave@claude-toolkit
claude plugin install workflows@claude-toolkit
claude plugin install team@claude-toolkit
claude plugin install dag@claude-toolkit
claude plugin install orchestra@claude-toolkit
claude plugin install compound-loop@claude-toolkit
# 4. IMPORTANT: Restart Claude Code to load plugins
# Exit current session and start a new one
```
### Installation Tiers
```
┌─────────────────────────────────────────────────────────────────────┐
│ TIER 1: Lite (Memory + Analysis) │
│ ───────────────────────────────────────────────────────────────── │
│ claude plugin install tldr@claude-toolkit │
│ claude plugin install memory@claude-toolkit │
├─────────────────────────────────────────────────────────────────────┤
│ TIER 2: Standard (+ Workflows) │
│ ───────────────────────────────────────────────────────────────── │
│ claude plugin install wave@claude-toolkit │
│ claude plugin install workflows@claude-toolkit │
├─────────────────────────────────────────────────────────────────────┤
│ TIER 3: Advanced Orchestration │
│ ───────────────────────────────────────────────────────────────── │
│ claude plugin install dag@claude-toolkit # Dependencies │
│ claude plugin install orchestra@claude-toolkit # Unified layer │
├─────────────────────────────────────────────────────────────────────┤
│ TIER 4: Full (+ Team Coordination) │
│ ───────────────────────────────────────────────────────────────── │
│ claude plugin install team@claude-toolkit │
│ # Run team server for cross-terminal coordination │
└─────────────────────────────────────────────────────────────────────┘
```
> **Note:** After installing plugins, you must restart Claude Code for them to take effect.
---
## Plugins
### tldr - Token-Efficient Code Analysis
**95% token savings** through 5-layer code analysis.
```bash
/tldr structure src/ # Code structure overview
/tldr context auth_handler # LLM-ready context for a function
/tldr impact process_data # What calls this function?
/tldr slice file.py func 42 # What affects line 42?
```
| Layer | What It Provides | Tokens |
|-------|------------------|--------|
| L1: AST | Functions, classes, signatures | ~500 |
| L2: Call Graph | Cross-file dependencies | +440 |
| L3: CFG | Control flow, complexity | +110 |
| L4: DFG | Data flow, variable tracking | +130 |
| L5: PDG | Program slicing, impact | +150 |
**Total: ~1,200 tokens vs 23,000 raw = 95% savings**
[Full documentation →](plugins/tldr/README.md)
---
### memory - Cross-Session Learning
Remember and recall across sessions. SQLite-based, no external dependencies.
```bash
/remember "JWT auth requires the secret in .env"
/recall authentication patterns
```
The system also learns automatically from your sessions via the memory-extractor hook.
| Feature | Description |
|---------|-------------|
| Semantic search | Find by meaning, not just keywords |
| Auto-extraction | Learns from your work automatically |
| Session handoffs | Resume where you left off |
[Full documentation →](plugins/memory/README.md)
---
### compound-loop - Self-Improving Loop
Every session leaves the next one **smarter**. A `/compound` skill, an independent `verifier` subagent that runs your tests itself, and a per-project `STATE.md` — enforcing **fail → investigate → verify → distill → consult**.
```bash
/compound-loop:compound # close out a task: verify, then distill a CONFIRMED lesson
/compound-loop:compound-nightly # optional unattended maintenance (schedule it nightly)
```
| Feature | Description |
|---------|-------------|
| Independent verifier | Runs tests/lint itself, instructed to reject — defeats "Verifier Theater" |
| Durable STATE.md | Per-project facts, rules, open failures, resume pointer |
| Gated distillation | A lesson is written only after a verifier PASS |
| Travels via memory | Cross-project rules land in `MEMORY.md`, auto-loaded next session |
[Full documentation →](plugins/compound-loop/README.md)
---
### wave - Parallel Task Orchestration
Execute complex tasks through waves of parallel agents.
```bash
/wave "Build REST API with auth, CRUD, and tests"
/wave-status
/cancel-wave
```
```
Wave 1 Wave 2 Wave 3
[Explore] [Plan] [Auth] [CRUD] [Tests] [Docs]
DONE!
```
| Feature | Description |
|---------|-------------|
| Parallel agents | Spawn multiple workers per wave |
| Iteration loop | Automatic retry until complete |
| Inter-agent signals | Agents communicate progress/discoveries |
| File claiming | Prevent edit conflicts |
[Full documentation →](plugins/wave/README.md)
---
### dag - Dependency-Aware Orchestration
Execute tasks with proper dependency ordering while maximizing parallelism.
```bash
/dag "Build REST API"
dag add design "Design schema"
dag add models "Build models" design # depends on design
dag add routes "Build routes" design # depends on design
dag add tests "Write tests" "models,routes" # depends on both
```
```
┌→ models ──┐
design ──┤ ├→ tests
└→ routes ──┘
```
| Pattern | Description |
|---------|-------------|
| Chain | A → B → C → D |
| Diamond | A → [B,C] → D |
| Fan-out | A → [B,C,D] |
| Fan-in | [A,B,C] → D |
[Full documentation →](plugins/dag/README.md)
---
### orchestra - Unified Orchestration
**Enterprise-grade orchestration** combining all coordination primitives.
```bash
/orchestra "Build checkout feature" full
orchestra add design "Design flow"
orchestra add cart "Cart" design
orchestra add payment "Payment" design
orchestra add confirm "Confirm" "cart,payment"
```
| Feature | Script | Description |
|---------|--------|-------------|
| **DAG** | Built-in | Task dependencies |
| **Wave** | Built-in | Parallel iteration |
| **Work Queue** | `work-queue.sh` | Dynamic task distribution |
| **Hierarchy** | `hierarchy.sh` | Nested sub-orchestrators |
| **Approval Gates** | `approval-gates.sh` | Human checkpoints |
| **Scheduler** | `scheduler.sh` | Deadlines & timeouts |
#### Advanced Features
**Work Stealing** - Fast agents grab more work automatically:
```bash
work-queue.sh init
work-queue.sh push task-1 "High priority" 1
work-queue.sh claim worker-1 # Gets highest priority
```
**Hierarchical Orchestration** - Break big problems into isolated sub-orchestrators:
```bash
hierarchy.sh create auth "Build authentication"
hierarchy.sh create payments "Build payments"
hierarchy.sh aggregate # Track all sub-orchestrators
```
**Human Approval Gates** - Block critical phases until approved:
```bash
approval-gates.sh create deploy "Deploy to production?"
approval-gates.sh approve deploy "admin"
```
**Time-Based Scheduling** - Deadlines, timeouts, delayed starts:
```bash
scheduler.sh deadline task-1 30m
scheduler.sh timeout task-1 5m
scheduler.sh overdue # List late tasks
```
[Full documentation →](plugins/orchestra/README.md)
---
### workflows - Development Workflows
Pre-built workflows for common development tasks.
```bash
/workflow # Interactive router
/fix bug "login fails" # Debug → implement → test → commit
/build feature "..." # Plan → validate → implement → PR
/explore # Understand codebase
/review # Code review
```
Uses 10 consolidated agents:
| Agent | Purpose |
|-------|---------|
| explorer | Codebase exploration |
| researcher | External research |
| planner | Planning & design |
| implementer | Implementation |
| tester | Testing |
| debugger | Bug investigation |
| reviewer | Code review |
| orchestrator | Multi-agent coordination |
| documenter | Documentation |
| reporter | Communication |
[Full documentation →](plugins/workflows/README.md)
---
### team - Multi-Instance Coordination
Multiple Claude Code instances working as a team.
```bash
# Terminal 1: Team Lead
/team create "Sprint 1"
/team assign worker-1 "Implement auth"
/team broadcast "Sprint started!"
# Terminal 2: Worker
/team join "Sprint 1"
/team status
/team complete "auth task"
```
**Requires:** Team server running locally.
[Full documentation →](plugins/team/README.md)
---
## Orchestration Guide
For detailed orchestration patterns and best practices, see [ORCHESTRATION.md](ORCHESTRATION.md).
### When to Use What
| Scenario | Plugin | Why |
|----------|--------|-----|
| Tasks with dependencies | **dag** | Respects ordering, maximizes parallelism |
| Many independent tasks | **wave** | All-parallel with iteration |
| Complex enterprise workflow | **orchestra** | All features + approval gates |
| Multi-terminal collaboration | **team** | Real-time coordination |
| Simple development task | **workflows** | Pre-built agent chains |
| Make work compound across sessions | **compound-loop** | Verified lessons + independent verifier, so each run sharpens the next |
### Feature Comparison
| Feature | wave | dag | orchestra | team |
|---------|------|-----|-----------|------|
| Task dependencies | ❌ | ✅ | ✅ | ❌ |
| Parallel execution | ✅ | ✅ | ✅ | ✅ |
| Inter-agent signals | ✅ | ❌ | ✅ | ✅ |
| Work stealing | ❌ | ❌ | ✅ | ❌ |
| Hierarchical | ❌ | ❌ | ✅ | ❌ |
| Approval gates | ❌ | ❌ | ✅ | ❌ |
| Deadlines | ❌ | ❌ | ✅ | ❌ |
| Cross-terminal | ❌ | ❌ | ✅* | ✅ |
| No external deps | ✅ | ✅ | ✅ | ❌ |
*Orchestra integrates with team when available
---
## Quick Start
```bash
# Clone and setup marketplace
git clone https://github.com/sethdford/claude-toolkit.git
cd claude-toolkit
claude plugin marketplace add .
# Install all plugins
claude plugin install tldr@claude-toolkit
claude plugin install memory@claude-toolkit
claude plugin install wave@claude-toolkit
claude plugin install workflows@claude-toolkit
claude plugin install dag@claude-toolkit
claude plugin install orchestra@claude-toolkit
claude plugin install team@claude-toolkit
claude plugin install compound-loop@claude-toolkit
# Initialize memory database
./plugins/memory/scripts/memory.sh init
# RESTART Claude Code, then test:
# /dag "Test dependencies"
# /orchestra "Test unified orchestration"
# /workflow
```
---
## Architecture
```
claude-toolkit/
├── plugins/ # Claude Code plugins
│ ├── tldr/ # Code analysis plugin
│ ├── memory/ # Cross-session memory
│ ├── wave/ # Parallel orchestration + signals
│ ├── workflows/ # /fix, /build, /explore
│ ├── dag/ # Dependency-aware orchestration
│ ├── orchestra/ # Unified orchestration layer
│ │ └── scripts/
│ │ ├── orchestra.sh # Main orchestrator
│ │ ├── work-queue.sh # Work stealing queue
│ │ ├── hierarchy.sh # Nested orchestrators
│ │ ├── approval-gates.sh # Human checkpoints
│ │ └── scheduler.sh # Deadlines & timeouts
│ ├── team/ # Multi-instance coordination
│ └── compound-loop/ # Self-improving loop (skill + verifier + STATE.md)
├── packages/ # Standalone tools
│ ├── tldr-core/ # TLDR CLI (Python)
│ ├── memory-core/ # Memory library (Python)
│ └── team-server/ # Team coordination server
├── agents/ # Consolidated agent definitions
├── docs/ # Documentation
├── examples/ # Example configurations
├── ORCHESTRATION.md # Orchestration guide
└── scripts/ # Installation & setup scripts
```
---
## Credits
This toolkit consolidates work from:
- **wave-orchestration** - Wave pattern, stop hook iteration
- **claude-code-collab** - Task tools, team coordination
- **Continuous-Claude-v3** - TLDR, memory, workflows, agents
Inspired by:
- [Ralph Wiggum Technique](https://ghuntley.com/ralph/) by Geoffrey Huntley
- [obra/superpowers](https://github.com/obra/superpowers) - Agent orchestration patterns
- [loop-engineering](https://github.com/cobusgreyling/loop-engineering) by Cobus Greyling - compound-loop primitives & the "Verifier Theater" failure mode
- [Anthropic](https://anthropic.com) - Claude Code, [sub-agents](https://docs.anthropic.com/en/docs/claude-code/sub-agents) & [memory-tool](https://platform.claude.com/docs/en/agents-and-tools/tool-use/memory-tool) docs
---
## License
MIT - See [LICENSE](LICENSE)