Claude Code Orchestrator Kit

Professional automation and orchestration system for Claude Code

Complete toolkit with 39 AI agents, 38 skills, 25 slash commands, auto-optimized MCP, Beads issue tracking, Gastown multi-agent orchestration, ready-to-use prompts, and quality gates for building production-ready projects with Claude Code.

English | Русский

---

Table of Contents

• Overview
• Key Innovations
• Quick Start
• Installation
• Architecture
• Agents Ecosystem
• Skills Library
• Slash Commands
• MCP Configuration
• Claude Code Settings
• Prompts
• Project Structure
• Usage Examples
• Best Practices
• Gastown Setup
• Contributing
• License

---

Overview

Claude Code Orchestrator Kit transforms Claude Code from a simple assistant into an intelligent orchestration system. Instead of doing everything directly, Claude Code acts as an orchestrator that delegates complex tasks to specialized sub-agents, preserving context and enabling indefinite work sessions.

What You Get

Category	Count	Description
AI Agents	39	Specialized workers for bugs, security, testing, database, frontend, DevOps
Skills	39	Reusable utilities for validation, reporting, automation, senior expertise
Commands	25	Health checks, SpecKit, Beads, Gastown, process-logs, worktree, releases
MCP Servers	6	Auto-optimized: Context7, Sequential Thinking, Supabase, Playwright, shadcn, Serena

Key Benefits

• Context Preservation: Main session stays lean (~10-15K tokens vs 50K+ in standard usage)
• Specialization: Each agent is expert in its domain
• Indefinite Work: Can work on project indefinitely without context exhaustion
• Quality Assurance: Mandatory verification after every delegation
• Senior Expertise: Skills like code-reviewer, senior-devops, senior-prompt-engineer

---

Key Innovations

1. Orchestrator Pattern

The Core Paradigm: Claude Code acts as orchestrator, delegating to specialized sub-agents.

┌─────────────────────────────────────────────────────────────────┐
│                     MAIN CLAUDE CODE                             │
│                   (Orchestrator Role)                            │
├─────────────────────────────────────────────────────────────────┤
│  1. GATHER CONTEXT    │  2. DELEGATE        │  3. VERIFY        │
│  - Read existing code │  - Invoke agent     │  - Read results   │
│  - Search patterns    │  - Provide context  │  - Run type-check │
│  - Check recent commits│  - Set criteria    │  - Accept/reject  │
└─────────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────────┐
│                    SPECIALIZED AGENTS                            │
├──────────────┬──────────────┬──────────────┬───────────────────┤
│  bug-hunter  │  security-   │  database-   │  performance-     │
│  bug-fixer   │  scanner     │  architect   │  optimizer        │
│  dead-code-  │  vuln-fixer  │  api-builder │  accessibility-   │
│  hunter      │              │  supabase-   │  tester           │
│              │              │  auditor     │                   │
└──────────────┴──────────────┴──────────────┴───────────────────┘

2. Inline Skills (New Architecture)

Evolution from Orchestrators: We replaced heavy orchestrator agents with lightweight inline skills.

Old Approach	New Approach
Separate orchestrator agent per workflow	Inline skill executed directly
~1400 lines per workflow	~150 lines per skill
9+ orchestrator calls	0 orchestrator calls
~10,000+ tokens overhead	~500 tokens
Context reload each call	Single session context

Example: /health-bugs now uses bug-health-inline skill:

Detection → Validate → Fix by Priority → Verify → Repeat if needed

3. Senior-Level Skills

Professional-grade skills for complex tasks:

Skill	Expertise
code-reviewer	TypeScript, Python, Go, Swift, Kotlin code review
senior-devops	CI/CD, Docker, Kubernetes, Terraform, Cloud
senior-prompt-engineer	LLM optimization, RAG, agent design
ux-researcher-designer	User research, personas, journey mapping
systematic-debugging	Root cause analysis, debugging workflows

4. Auto-Optimized MCP Configuration

No manual switching required! Claude Code automatically optimizes context usage:

• Single .mcp.json with all servers — no manual switching needed
• Automatic deferred loading via ENABLE_TOOL_SEARCH=auto:5
• 85% context reduction for MCP tools (loaded on-demand via ToolSearch)
• Transparent to user — just works without configuration

5. SpecKit Integration

Specification-driven development workflow with Phase 0 Planning:

• Executor assignment (MAIN vs specialized agent)
• Parallel agent creation via meta-agent
• Atomicity: 1 Task = 1 Agent Invocation

6. Beads Issue Tracking (Optional)

Beads by Steve Yegge — git-backed issue tracker for AI agents:

• Persistent tasks: Survives session restarts, tracked in git
• Dependency graph: blocks, blocked-by, discovered-from
• Multi-session: Work across multiple Claude sessions without losing context
• 8 workflow formulas: bigfeature, bugfix, hotfix, healthcheck, etc.
• Initialize: Run /beads-init in your project

7. Gastown Multi-Agent Orchestration (Optional)

Gastown by Steve Yegge — multi-agent workspace manager that dispatches tasks to AI worker processes (polecats):

You (human)
  │
  ├─ /work "Fix login bug"          ← Give task via Claude Code
  │    │
  │    ├─ bd create (bead)           ← Creates task in Beads
  │    └─ gt sling PREFIX-xxx RIG   ← Dispatches to Gastown
  │         │
  │         └─ Daemon (automatic)
  │              ├─ Spawns Polecat (AI worker in isolated git worktree)
  │              ├─ Polecat: branch → implement → test → commit
  │              ├─ Refinery: merge queue → develop
  │              └─ Witness: health monitoring
  │
  ├─ /status                         ← Check progress
  └─ git push                        ← Ship changes

Key Features:

• Parallel AI workers: Multiple polecats work simultaneously in isolated git worktrees
• Multi-runtime: claude (default), codex, gemini — all subscription-based, no API billing
• A/B testing: /work --ab "task" sends same task to 2 runtimes, compare results
• Self-healing: Daemon manages Dolt DB, restarts crashed agents, monitors health
• Auto-provisioning: /onboard connects any project with a single command

Included commands:

Command	Purpose
/onboard	Connect project to Gastown (one-time setup)
/work "task"	Dispatch task to AI polecat
/status	Show convoys, agents, pending tasks
/upgrade	Safely upgrade gt/bd binaries

Agent instruction templates for all runtimes are provided in .claude/templates/:

• CLAUDE.md — Claude Code instructions with Gastown workflow
• AGENTS.md — Codex-compatible instructions
• GEMINI.md — Gemini-compatible instructions

Initialize: Run /onboard in your project directory. See Gastown Setup Guide below.

---

Quick Start

Option 1: npm Install

npm install -g claude-code-orchestrator-kit
cd your-project
claude-orchestrator  # Interactive setup

Option 2: Clone Repository

git clone https://github.com/maslennikov-ig/claude-code-orchestrator-kit.git
cd claude-code-orchestrator-kit

# Configure environment (optional, for Supabase)
cp .env.example .env.local
# Edit .env.local with your credentials

# Restart Claude Code - ready!

Option 3: Copy to Existing Project

# Copy orchestration system to your project
cp -r claude-code-orchestrator-kit/.claude /path/to/your/project/
cp claude-code-orchestrator-kit/.mcp.json /path/to/your/project/
cp claude-code-orchestrator-kit/CLAUDE.md /path/to/your/project/

---

Installation

Prerequisites

• Claude Code CLI installed
• Node.js 18+ (for MCP servers)
• Git (for version control features)

Environment Variables

Create .env.local (git-ignored) with your credentials:

# Supabase (optional)
SUPABASE_PROJECT_REF=your-project-ref
SUPABASE_ACCESS_TOKEN=your-token

# Sequential Thinking (optional)
SEQUENTIAL_THINKING_KEY=your-smithery-key
SEQUENTIAL_THINKING_PROFILE=your-profile

Verify Installation

# Check that .mcp.json and .claude/settings.json exist
ls -la .mcp.json .claude/settings.json

# Try a health command in Claude Code
/health-bugs

---

Architecture

Component Overview

┌────────────────────────────────────────────────────────────────┐
│                        CLAUDE.md                                │
│                  (Behavioral Operating System)                  │
│                                                                 │
│  Defines: Orchestration rules, delegation patterns, verification│
└────────────────────────────────────────────────────────────────┘
                              ↓
┌────────────────────────────────────────────────────────────────┐
│                         AGENTS                                  │
│            (39 specialized workers)                             │
├────────────────────────────────────────────────────────────────┤
│  health/       development/   testing/      database/          │
│  ├─bug-hunter  ├─llm-service  ├─integration ├─database-arch   │
│  ├─bug-fixer   ├─typescript   ├─performance ├─api-builder     │
│  ├─security-   ├─code-review  ├─mobile      ├─supabase-audit  │
│  ├─dead-code   ├─utility-     ├─access-     │                  │
│  └─reuse-      └─skill-build  └─ibility     │                  │
│                                                                 │
│  infrastructure/  frontend/     meta/        research/         │
│  ├─deployment     ├─nextjs-ui   ├─meta-agent ├─problem-invest  │
│  ├─qdrant         ├─fullstack   └─skill-v2   └─research-spec   │
│  └─orchestration  └─visual-fx                                  │
└────────────────────────────────────────────────────────────────┘
                              ↓
┌────────────────────────────────────────────────────────────────┐
│                         SKILLS                                  │
│            (37 reusable utilities)                              │
├────────────────────────────────────────────────────────────────┤
│  Inline Orchestration:        Senior Expertise:                 │
│  ├─bug-health-inline          ├─code-reviewer                  │
│  ├─security-health-inline     ├─senior-devops                  │
│  ├─deps-health-inline         ├─senior-prompt-engineer         │
│  ├─cleanup-health-inline      ├─ux-researcher-designer         │
│  └─reuse-health-inline        └─systematic-debugging           │
│                                                                 │
│  Utilities:                   Creative:                         │
│  ├─validate-plan-file         ├─algorithmic-art                │
│  ├─run-quality-gate           ├─canvas-design                  │
│  ├─rollback-changes           ├─theme-factory                  │
│  ├─parse-git-status           └─artifacts-builder              │
│  └─generate-report-header                                       │
└────────────────────────────────────────────────────────────────┘
                              ↓
┌────────────────────────────────────────────────────────────────┐
│                        COMMANDS                                 │
│            (21 slash commands)                                  │
├────────────────────────────────────────────────────────────────┤
│  /health-bugs      /speckit.specify    /worktree              │
│  /health-security  /speckit.plan       /push                  │
│  /health-deps      /speckit.implement  /translate-doc          │
│  /health-cleanup   /speckit.clarify    /process-logs          │
│  /health-reuse     /speckit.constitution                       │
│  /health-metrics   /speckit.taskstoissues                      │
└────────────────────────────────────────────────────────────────┘

---

Agents Ecosystem

39 Specialized Agents

Health (10 agents)

Agent	Purpose
bug-hunter	Detect bugs, categorize by priority
bug-fixer	Fix bugs from reports
security-scanner	Find security vulnerabilities
vulnerability-fixer	Fix security issues
dead-code-hunter	Detect unused code
dead-code-remover	Remove dead code safely
dependency-auditor	Audit package dependencies
dependency-updater	Update dependencies safely
reuse-hunter	Find code duplication
reuse-fixer	Consolidate duplicated code

Development (6 agents)

Agent	Purpose
llm-service-specialist	LLM integration, prompts
typescript-types-specialist	Type definitions, generics
cost-calculator-specialist	Token/API cost estimation
utility-builder	Build utility services
skill-builder-v2	Create new skills
code-reviewer	Comprehensive code review

Testing (6 agents)

Agent	Purpose
integration-tester	Database, API, async tests
test-writer	Write unit/contract tests
performance-optimizer	Core Web Vitals, PageSpeed
mobile-responsiveness-tester	Mobile viewport testing
mobile-fixes-implementer	Fix mobile issues
accessibility-tester	WCAG compliance

Database (3 agents)

Agent	Purpose
database-architect	PostgreSQL schema design
api-builder	tRPC routers, auth middleware
supabase-auditor	RLS policies, security

Infrastructure (5 agents)

Agent	Purpose
infrastructure-specialist	Supabase, Qdrant, Redis
qdrant-specialist	Vector database operations
quality-validator-specialist	Quality gate validation
orchestration-logic-specialist	Workflow state machines
deployment-engineer	CI/CD, Docker, DevOps

Frontend (3 agents)

Agent	Purpose
nextjs-ui-designer	Modern UI/UX design
fullstack-nextjs-specialist	Full-stack Next.js
visual-effects-creator	Animations, visual effects

Other (6 agents)

Agent	Purpose
meta-agent-v3	Create new agents
technical-writer	Documentation
problem-investigator	Deep problem analysis
research-specialist	Technical research
article-writer-multi-platform	Multi-platform content
lead-research-assistant	Lead qualification

---

Skills Library

38 Reusable Skills

Inline Orchestration (5 skills)

Execute health workflows directly without spawning orchestrator agents. All skills include Beads integration for issue tracking.

Skill	Invocation	Purpose	Version
health-bugs	/health-bugs	Bug detection & fixing with history enrichment	3.1.0
security-health-inline	/health-security	Security vulnerability scanning & fixing	3.0.0
deps-health-inline	/health-deps	Dependency audit & update	3.0.0
cleanup-health-inline	/health-cleanup	Dead code detection & removal	3.0.0
reuse-health-inline	/health-reuse	Code duplication consolidation	3.0.0

Senior Expertise (6 skills)

Professional-grade domain expertise:

Skill	Expertise
code-reviewer	TypeScript, Python, Go, Swift, Kotlin review
senior-devops	CI/CD, containers, cloud, infrastructure
senior-prompt-engineer	LLM optimization, RAG, agents
ux-researcher-designer	User research, personas
ui-design-system	Design tokens, components
systematic-debugging	Root cause analysis

Validation & Quality (6 skills)

Skill	Purpose
validate-plan-file	JSON schema validation
validate-report-file	Report completeness
run-quality-gate	Type-check/build/tests
calculate-priority-score	Bug/task prioritization
setup-knip	Configure dead code detection
rollback-changes	Restore from changes log

Reporting & Formatting (6 skills)

Skill	Purpose
generate-report-header	Standardized report headers
generate-changelog	Changelog from commits
format-markdown-table	Well-formatted tables
format-commit-message	Conventional commits
format-todo-list	TodoWrite-compatible lists
render-template	Variable substitution

Parsing & Extraction (4 skills)

Skill	Purpose
parse-git-status	Parse git status output
parse-package-json	Extract version, deps
parse-error-logs	Parse build/test errors
extract-version	Semantic version parsing

Creative & UI (6 skills)

Skill	Purpose
algorithmic-art	Generative art with p5.js
canvas-design	Visual art in PNG/PDF
theme-factory	Theme styling for artifacts
artifacts-builder	Multi-component HTML artifacts
webapp-testing	Playwright testing
frontend-aesthetics	Distinctive UI design

Automation Workflows (2 skills)

Skill	Purpose	Version
process-logs	Automated error log processing with Beads integration	1.8.0
process-issues	GitHub Issues processing with similar issue search	1.1.0

Other (4 skills)

Skill	Purpose
git-commit-helper	Commit message from diff
changelog-generator	User-facing changelogs
content-research-writer	Research-driven content
lead-research-assistant	Lead identification

---

Slash Commands

21 Commands

Health Monitoring (6 commands)

Command	Purpose
/health-bugs	Bug detection and fixing workflow
/health-security	Security vulnerability scanning
/health-deps	Dependency audit and updates
/health-cleanup	Dead code detection and removal
/health-reuse	Code duplication elimination
/health-metrics	Monthly ecosystem health report

Example:

/health-bugs
# Scans → Categorizes → Fixes by priority → Validates → Reports

SpecKit (9 commands)

Command	Purpose
/speckit.analyze	Analyze requirements
/speckit.specify	Generate specifications
/speckit.clarify	Ask clarifying questions
/speckit.plan	Create implementation plan
/speckit.implement	Execute implementation
/speckit.checklist	Generate QA checklist
/speckit.tasks	Break into tasks
/speckit.constitution	Define project constitution
/speckit.taskstoissues	Convert tasks to GitHub issues

Beads (2 commands)

Command	Purpose
/beads-init	Initialize Beads in project
/speckit.tobeads	Import tasks.md to Beads

Gastown (4 commands)

Command	Purpose
/onboard [rig-name]	Connect project to Gastown (one-time setup)
/work [--agent|--ab|--all] "task"	Dispatch task to AI polecat
/status	Show convoys, agents, pending tasks
/upgrade [gt|bd|all]	Safely upgrade Gastown/Beads binaries

Other (4 commands)

Command	Purpose
/process-logs	Automated error log processing and fixing
/push [patch|minor|major]	Automated release with changelog
/worktree	Git worktree management
/translate-doc	Translate documentation (EN↔RU)

---

MCP Configuration

Unified Auto-Optimized Setup

No more manual switching! The kit uses a single .mcp.json with automatic optimization.

How It Works

1. Single config file (.mcp.json) contains all MCP servers
2. Auto-deferred loading via .claude/settings.json:

   {
     "env": { "ENABLE_TOOL_SEARCH": "auto:5" }
   }
   

3. On-demand loading — Claude loads tools only when needed via ToolSearch
4. 85% context savings compared to loading all tools upfront

Included MCP Servers

Server	Purpose	Tools
Context7	Up-to-date library documentation	resolve-library-id, query-docs
Sequential Thinking	Structured reasoning for complex tasks	sequentialthinking
Supabase	Database operations, migrations, RLS	Tables, SQL, migrations, edge functions
Playwright	Browser automation & testing	Screenshots, navigation, form filling
shadcn/ui	UI component library integration	Registry search, component examples
Serena	Semantic code analysis (LSP)	Symbol search, references, refactoring

Environment Variables

Set in .env.local for Supabase integration:

SUPABASE_PROJECT_REF=your-project-ref
SUPABASE_ACCESS_TOKEN=your-token

---

Claude Code Settings

.claude/settings.json

Project-level Claude Code configuration for enhanced workflow:

{
  "plansDirectory": "./docs/plans",
  "env": {
    "ENABLE_TOOL_SEARCH": "auto:5"
  }
}

Settings Explained

Setting	Value	Purpose
plansDirectory	./docs/plans	Where Claude saves implementation plans when using Plan Mode
ENABLE_TOOL_SEARCH	auto:5	Auto-enables deferred MCP tool loading for servers with >5 tools

Benefits

• Plan Mode Integration: Plans are saved to docs/plans/ for version control and review
• Automatic Context Optimization: MCP tools load on-demand instead of upfront
• No Manual Configuration: Works transparently — just install and use

---

Prompts

Ready-to-use prompts for setting up various features in your project. Copy, paste, and let Claude Code do the work.

Prompt	Description
setup-health-workflows.md	Health workflows with Beads integration (/health-bugs, /health-security, etc.)
setup-error-logging.md	Complete error logging system with DB table, logger service, auto-mute rules

Quick Start: Health Workflows

# 1. Install Beads CLI
npm install -g @anthropic/beads-cli

# 2. Initialize in your project
bd init

# 3. Run in Claude Code
/health-bugs

How to Use:

1. Copy the prompt content to your chat with Claude Code
2. Answer any questions Claude asks about your project specifics
3. Review the generated code before committing

See prompts/README.md for full documentation.

---

Project Structure

claude-code-orchestrator-kit/
├── .claude/
│   ├── agents/                 # 39 AI agents
│   │   ├── health/             # Bug, security, deps, cleanup
│   │   ├── development/        # LLM, TypeScript, utilities
│   │   ├── testing/            # Integration, performance, mobile
│   │   ├── database/           # Supabase, API, architecture
│   │   ├── infrastructure/     # Qdrant, deployment, orchestration
│   │   ├── frontend/           # Next.js, visual effects
│   │   ├── meta/               # Agent/skill creators
│   │   ├── research/           # Problem investigation
│   │   ├── documentation/      # Technical writing
│   │   ├── content/            # Article writing
│   │   └── business/           # Lead research
│   │
│   ├── skills/                 # 37 reusable skills
│   │   ├── bug-health-inline/  # Inline orchestration
│   │   ├── code-reviewer/      # Senior expertise
│   │   ├── validate-plan-file/ # Validation utilities
│   │   └── ...
│   │
│   ├── commands/               # 25 slash commands
│   │   ├── health-*.md         # Health monitoring
│   │   ├── speckit.*.md        # SpecKit workflow
│   │   ├── work.md             # Gastown: dispatch tasks
│   │   ├── status.md           # Gastown: show status
│   │   ├── upgrade.md          # Gastown: safe upgrade
│   │   ├── onboard.md          # Gastown: connect project
│   │   └── ...
│   │
│   ├── templates/              # Instruction file templates
│   │   ├── CLAUDE.md           # Claude Code template
│   │   ├── AGENTS.md           # Codex template
│   │   └── GEMINI.md           # Gemini template
│   │
│   ├── schemas/                # JSON schemas
│   └── scripts/                # Quality gate scripts
│
├── mcp/                        # Legacy MCP configs (reference only)
│   └── ...
│
├── prompts/                    # Ready-to-use setup prompts
│   ├── README.md
│   └── setup-error-logging.md
│
├── docs/                       # Documentation
│   ├── FAQ.md
│   ├── ARCHITECTURE.md
│   ├── TUTORIAL-CUSTOM-AGENTS.md
│   └── ...
│
├── .mcp.json                   # Unified MCP configuration
├── CLAUDE.md                   # Behavioral Operating System
└── package.json                # npm package config

---

Usage Examples

Example 1: Bug Fixing Workflow

# Run complete bug detection and fixing
/health-bugs

# What happens:
# 1. Pre-flight validation
# 2. Bug detection (bug-hunter agent)
# 3. Quality gate validation
# 4. Priority-based fixing (critical → low)
# 5. Quality gates after each priority
# 6. Verification scan
# 7. Final report

Example 2: Code Review

# Invoke code-reviewer skill
/code-reviewer

# Provides:
# - Automated code analysis
# - Best practices checking
# - Security scanning
# - Review checklist

Example 3: Release Automation

# Auto-detect version bump
/push

# Or specify type
/push minor

# Actions:
# 1. Analyze commits since last release
# 2. Bump version in package.json
# 3. Generate changelog entry
# 4. Create git commit + tag
# 5. Push to remote

Example 4: Parallel Feature Development

# Create worktrees
/worktree create feature/new-auth
/worktree create feature/new-ui

# Work in parallel
cd .worktrees/feature-new-auth
# ... changes ...

# Cleanup when done
/worktree cleanup

---

Best Practices

1. Use Auto-Optimized MCP

The kit automatically optimizes context usage — no configuration needed:

• MCP tools load on-demand via ToolSearch
• ~85% context reduction compared to loading all tools upfront

2. Run Health Checks Weekly

/health-bugs      # Monday
/health-security  # Tuesday
/health-deps      # Wednesday
/health-cleanup   # Thursday
/health-metrics   # Monthly

3. Use Library-First Approach

Before writing code >20 lines, search for existing libraries:

• Check npm/PyPI for packages with >1k weekly downloads
• Evaluate maintenance status and types support
• Use library if it covers >70% of functionality

4. Follow Orchestration Rules

1. GATHER CONTEXT FIRST - Read code, search patterns
2. DELEGATE TO SUBAGENTS - Provide complete context
3. VERIFY RESULTS - Never skip verification
4. ACCEPT/REJECT LOOP - Re-delegate if needed

5. Keep Credentials Secure

# Never commit .env.local
echo ".env.local" >> .gitignore

---

Gastown Setup

Prerequisites

1. Go 1.21+ installed (go version)
2. Gastown and Beads CLI installed:

   go install github.com/steveyegge/gastown/cmd/gt@latest
   go install github.com/steveyegge/beads/cmd/bd@latest
   

3. Gastown town initialized:

   gt init ~/gt
   

4. Daemon running as systemd user service:

   gt daemon enable-supervisor
   systemctl --user start gastown-daemon
   loginctl enable-linger $USER  # Auto-start on boot
   

Important: Daemon Service Customization

The default gt daemon enable-supervisor template does NOT include PATH. You must add Environment lines manually:

# Edit the service file
nano ~/.local/share/systemd/user/gastown-daemon.service

Add under [Service]:

Environment="GT_TOWN_ROOT=/home/YOUR_USER/gt"
Environment="GT_ROOT=/home/YOUR_USER/gt"
Environment="PATH=/home/YOUR_USER/go/bin:/home/YOUR_USER/.local/bin:/usr/local/bin:/usr/bin:/bin"
Environment="HOME=/home/YOUR_USER"

Then reload and restart:

systemctl --user daemon-reload
systemctl --user restart gastown-daemon

Daemon Configuration

Configure Dolt management in ~/gt/mayor/daemon.json:

{
  "heartbeat": { "enabled": true, "interval": "3m" },
  "patrols": {
    "dolt_server": {
      "enabled": true,
      "port": 3307,
      "host": "127.0.0.1",
      "user": "root",
      "data_dir": "/home/YOUR_USER/gt/.dolt-data",
      "log_file": "/home/YOUR_USER/gt/daemon/dolt-server.log",
      "auto_restart": true
    },
    "deacon": { "enabled": true, "interval": "5m", "agent": "deacon" },
    "refinery": { "agent": "refinery", "enabled": true, "interval": "5m", "rigs": [] },
    "witness": { "agent": "witness", "enabled": true, "interval": "5m", "rigs": [] }
  },
  "type": "daemon-patrol-config",
  "version": 1
}

Important: time.Duration fields like health_check_interval must be integers (nanoseconds) in Go, NOT strings like "30s". Omit them to use defaults (30 seconds).

Connect a Project

Run /onboard from your project directory in Claude Code:

/onboard

This will:

1. Run gt rig add <name> <path> — auto-provisions 30+ components
2. Update daemon.json — adds project to witness/refinery patrols
3. Restart daemon — picks up new configuration
4. Run gt doctor --fix — diagnoses and auto-fixes issues
5. Copy slash commands — /work, /status, /upgrade, /onboard
6. Copy instruction templates — CLAUDE.md, AGENTS.md, GEMINI.md

Daily Workflow

# Give task to AI agent
/work "Fix the login validation bug"

# Check progress
/status

# Use specific runtime
/work --agent codex "Refactor auth module"

# A/B test: same task to 2 agents
/work --ab "Optimize database queries"

# Find available tasks
bd ready

# Visual dashboard
gt dashboard --open

Troubleshooting

Problem	Diagnosis	Solution
Daemon not starting	systemctl --user status gastown-daemon	Check PATH in service file
Dolt unreachable	gt dolt status	Restart daemon, check daemon.json
Doctor failures	gt doctor --fix --rig <name>	Auto-fixes most issues
Polecat stuck	gt convoy list	gt convoy cancel <id>

Upgrading

/upgrade        # Upgrade both gt and bd
/upgrade gt     # Upgrade Gastown only
/upgrade bd     # Upgrade Beads only

The /upgrade command handles the full cycle: stop daemon, upgrade binaries, verify service file, check daemon.json, restart, run doctor.

---

Documentation

Document	Description
FAQ	Frequently asked questions
Architecture	System design diagrams
Tutorial: Custom Agents	Create your own agents
Use Cases	Real-world examples
Performance	Token optimization
Migration Guide	Add to existing projects
Commands Guide	Detailed command reference

---

Contributing

Adding New Agents

1. Create file in .claude/agents/{category}/workers/
2. Follow agent template structure
3. Add to this README

Adding New Skills

1. Create directory .claude/skills/{skill-name}/
2. Add SKILL.md following format
3. Add to this README

Adding MCP Servers

1. Add server to .mcp.json
2. Document in README under MCP Configuration section

---

Attribution

SpecKit by GitHub

Commands /speckit.* adapted from GitHub's SpecKit.

• License: MIT License
• Copyright: GitHub, Inc.

Beads by Steve Yegge

Beads issue tracking integration adapted from Steve Yegge's Beads.

• Description: Distributed, git-backed graph issue tracker for AI agents
• License: MIT License
• Copyright: Steve Yegge
• Commands: /beads-init, /speckit.tobeads
• Templates: .beads-templates/ directory with 8 workflow formulas

Gastown by Steve Yegge

Multi-agent orchestration integration from Steve Yegge's Gastown.

• Description: Multi-agent workspace manager with isolated git worktrees
• License: MIT License
• Copyright: Steve Yegge
• Commands: /onboard, /work, /status, /upgrade
• Templates: .claude/templates/ directory with instruction files for all runtimes

---

Acknowledgments

Built with:

• Claude Code by Anthropic
• Context7 by Upstash
• Supabase MCP
• Smithery Sequential Thinking
• Playwright
• shadcn/ui

---

Stats

• 39 AI Agents
• 39 Reusable Skills
• 25 Slash Commands
• 6 MCP Servers (auto-optimized)
• 3 Runtime templates (Claude, Codex, Gemini)
• v1.4.20 Current Version

---

Author

Igor Maslennikov

• GitHub: @maslennikov-ig
• Website: aidevteam.ru

---

License

MIT License - see LICENSE file.

---

Star this repo if you find it useful!