Doyaken

A standalone multi-project autonomous agent CLI that works with any AI coding agent. Install once, use on any project.

Aliases: doyaken, dk

Why Doyaken?

• One install, all projects - Global installation works across all your repos
• Any AI agent - Claude, Codex, Gemini, Copilot, Cursor, or OpenCode
• Batteries included - 40+ skills, 25+ prompts, 8-phase workflow out of the box
• Zero config - Works immediately after dk init

Key Features

Multi-Agent Configuration Sync

Automatically generates and syncs configuration files for all major AI coding agents:

dk sync  # Generates all agent files from single source of truth

| Generated File | Agent/Tool | |---------------|------------| | AGENTS.md | Codex, OpenCode (industry standard) | | CLAUDE.md | Claude Code | | .cursorrules | Cursor (legacy) | | .cursor/rules/*.mdc | Cursor (modern rules) | | GEMINI.md | Google Gemini | | .github/copilot-instructions.md | GitHub Copilot | | .opencode.json | OpenCode |

All files point to .doyaken/ as the single source of truth. Edit once, sync everywhere.

Prompt Library (25+ Methodologies)

Battle-tested prompts for common development tasks:

| Category | Prompts | |----------|---------| | Code Quality | quality, refactor, debugging, errors | | Reviews | review-architecture, review-security, review-debt, review-performance | | Planning | planning, diagnose, research-features, research-competitors | | Development | api-rest, ci, git, docs |

# Use as slash commands in Claude Code
/quality      # Apply quality methodology
/debugging    # Debug current issue
/planning     # Create implementation plan

Vendor-Specific Prompts

Pre-configured prompts for popular frameworks and services:

• Frameworks: Next.js, Rails, React
• Databases: PostgreSQL, Redis, Supabase
• Platforms: Vercel, DigitalOcean, Dokku
• Tools: GitHub Actions, Figma

8-Phase Workflow

Structured task execution with configurable timeouts:

EXPAND → TRIAGE → PLAN → IMPLEMENT → TEST → DOCS → REVIEW → VERIFY

Each phase has dedicated prompts in .doyaken/prompts/phases/ that guide the agent through systematic task completion.

Skills System (40+ Built-in)

Reusable, composable skills with MCP integration:

dk skills                    # List all skills
dk skill periodic-review     # Run comprehensive codebase review
dk skill audit-security      # OWASP-based security audit
dk skill ci-fix              # Diagnose and fix CI failures

Claude Code Hooks

Automatic hooks that enhance Claude Code sessions:

| Hook | Purpose | |------|---------| | check-quality.sh | Run linters before commits | | check-security.sh | Scan for security issues | | task-context.sh | Inject current task context | | format-on-save.sh | Auto-format code | | protect-sensitive-files.sh | Prevent editing secrets | | inject-base-prompt.sh | Add project context to prompts |

Slash Command Generation

Auto-generates Claude Code slash commands from skills and prompts:

dk commands  # Regenerate .claude/commands/

Creates commands like /workflow, /security, /testing that you can invoke directly in Claude Code.

Periodic Codebase Reviews

Automated comprehensive reviews covering:

• Code quality and maintainability
• Security vulnerabilities (OWASP)
• Technical debt assessment
• Performance analysis
• UX audit
• Documentation gaps

dk skill periodic-review  # Full review with auto-fix

Additional Features

• Self-Healing: Automatic retries, model fallback, crash recovery
• Parallel Agents: Multiple agents work simultaneously with lock coordination
• Project Registry: Track projects by path, git remote, domains
• MCP Integration: Connect to GitHub, Linear, Slack, Jira
• Test Automation: test-chaos and test-user skills for robust testing

Installation

Option 1: npm (Recommended)

# Install globally
npm install -g @doyaken/doyaken

# Or use npx without installing
npx doyaken --help

Option 2: curl (Per-User)

curl -sSL https://raw.githubusercontent.com/mitchellfyi/doyaken-cli/main/install.sh | bash

Option 3: Clone & Install

git clone https://github.com/mitchellfyi/doyaken-cli.git
cd doyaken-cli
./install.sh

Quick Start

# Initialize a new project
cd /path/to/your/project
dk init

# Create a task
dk tasks new "Add user authentication"

# Run the agent
dk run 1     # Run 1 task
dk           # Run 5 tasks (default)

# Check status
dk status    # Project status
dk tasks     # Show taskboard
dk doctor    # Health check

Commands

| Command | Description | |---------|-------------| | dk | Run 5 tasks in current project | | dk run [N] | Run N tasks | | dk task "<prompt>" | Create and immediately run a single task | | dk init [path] | Initialize a new project | | dk tasks | Show taskboard | | dk tasks new <title> | Create a new task | | dk tasks view <id> | View a specific task | | dk skills | List available skills | | dk skill <name> | Run a skill | | dk sync | Sync all agent configuration files | | dk commands | Regenerate slash commands | | dk mcp status | Show MCP integration status | | dk mcp configure | Generate MCP configs | | dk status | Show project status | | dk list | List all registered projects | | dk manifest | Show project manifest | | dk config | Show/edit configuration | | dk upgrade | Upgrade doyaken to latest version | | dk upgrade --check | Check for available updates | | dk doctor | Health check | | dk help | Show help |

Note: doyaken and dk are interchangeable.

Multi-Agent Support

Doyaken supports multiple AI coding agents. Use --agent to switch between them:

dk --agent claude run 1      # Use Claude (default)
dk --agent cursor run 1      # Use Cursor
dk --agent codex run 1       # Use OpenAI Codex
dk --agent gemini run 1      # Use Google Gemini
dk --agent copilot run 1     # Use GitHub Copilot
dk --agent opencode run 1    # Use OpenCode

Supported Agents & Models

| Agent | Command | Models | Install | |-------|---------|--------|---------| | claude (default) | claude | opus, sonnet, haiku | npm i -g @anthropic-ai/claude-code | | cursor | cursor agent | claude-sonnet-4, gpt-4o | curl https://cursor.com/install -fsS \| bash | | codex | codex exec | gpt-5, o3, o4-mini | npm i -g @openai/codex | | gemini | gemini | gemini-2.5-pro, gemini-2.5-flash | npm i -g @google/gemini-cli | | copilot | copilot | claude-sonnet-4.5, gpt-5 | npm i -g @github/copilot | | opencode | opencode run | claude-sonnet-4, gpt-5 | npm i -g opencode-ai |

Specifying Models

dk --agent codex --model o3 run 1
dk --agent gemini --model gemini-2.5-flash run 2

Agent Workflow

The agent operates in 8 phases for each task:

| Phase | Timeout | Purpose | |-------|---------|---------| | EXPAND | 2min | Expand brief prompt into full task specification | | TRIAGE | 2min | Validate task, check dependencies | | PLAN | 5min | Gap analysis, detailed planning | | IMPLEMENT | 30min | Execute the plan, write code | | TEST | 10min | Run tests, add coverage | | DOCS | 5min | Sync documentation | | REVIEW | 10min | Code review, create follow-ups | | VERIFY | 3min | Verify task management, commit |

Parallel Execution

Run multiple agents simultaneously:

dk run 5 &
dk run 5 &
dk run 5 &

Agents coordinate via lock files in .doyaken/locks/ and will not work on the same task.

Skills

Skills are reusable prompts with YAML frontmatter that declare tool requirements:

# List available skills
dk skills

# Run a skill
dk skill github-import --filter=open

# Show skill info
dk skill github-import --info

Built-in Skills

Quality & Audits: | Skill | Description | |-------|-------------| | setup-quality | Set up quality gates, CI, and git hooks | | check-quality | Run all quality checks and report issues | | audit-security | OWASP-based security audit | | audit-deps | Audit dependencies for vulnerabilities | | audit-debt | Technical debt assessment | | audit-performance | Performance analysis | | audit-ux | User experience audit |

Development: | Skill | Description | |-------|-------------| | review-codebase | Comprehensive codebase review | | research-features | Discover next best feature to build | | ci-fix | Diagnose and fix CI/CD failures | | workflow | Run the 8-phase task workflow | | sync-agents | Sync agent config files to project |

Integrations (require MCP servers): | Skill | Description | Requires | |-------|-------------|----------| | github-import | Import GitHub issues as tasks | GitHub MCP | | github-sync | Sync task status to GitHub | GitHub MCP | | github-pr | Create PR from recent commits | GitHub MCP | | notify-slack | Send Slack notifications | Slack MCP | | mcp-status | Check MCP integration status | - |

Creating Custom Skills

Create a .md file in ~/.doyaken/skills/ (global) or .doyaken/skills/ (project):

---
name: my-skill
description: What this skill does
requires:
  - github                     # MCP servers needed
args:
  - name: filter
    description: Filter option
    default: "open"
---

# My Skill Prompt

Instructions for the AI agent...

MCP Integration

Doyaken supports MCP (Model Context Protocol) tools for external integrations:

# Show integration status
dk mcp status

# Generate MCP configs for enabled integrations
dk mcp configure

Enabling Integrations

Edit .doyaken/manifest.yaml:

integrations:
  github:
    enabled: true
  linear:
    enabled: false
  slack:
    enabled: false
  jira:
    enabled: false

After enabling, run dk mcp configure to generate the MCP configuration.

Supported Integrations

| Integration | Description | Required Env Var | |-------------|-------------|------------------| | GitHub | Issues, PRs, repos | GITHUB_TOKEN | | Linear | Issues, projects | LINEAR_API_KEY | | Slack | Messages, channels | SLACK_BOT_TOKEN | | Jira | Issues, sprints | JIRA_API_TOKEN |

Skill Hooks

Auto-run skills at specific workflow points:

# In manifest.yaml
skills:
  hooks:
    before-triage:
      - github-import       # Sync issues before starting
    after-verify:
      - github-sync         # Update issues after completion

Project Structure

After running dk init, your project will have:

your-project/
├── .doyaken/
│   ├── manifest.yaml        # Project configuration
│   ├── tasks/
│   │   ├── 1.blocked/       # Blocked tasks
│   │   ├── 2.todo/          # Ready to start
│   │   ├── 3.doing/         # In progress
│   │   └── 4.done/          # Completed
│   ├── prompts/
│   │   ├── library/         # 25+ methodology prompts
│   │   └── phases/          # 8-phase workflow prompts
│   ├── skills/              # Project-specific skills
│   ├── hooks/               # Claude Code hooks
│   ├── logs/                # Execution logs
│   ├── state/               # Session recovery
│   └── locks/               # Parallel coordination
├── .claude/
│   └── commands/            # Auto-generated slash commands
├── .cursor/
│   └── rules/               # Cursor modern rules (.mdc)
├── AGENTS.md                # Multi-agent instructions
├── CLAUDE.md                # Claude Code config
├── .cursorrules             # Cursor legacy config
├── GEMINI.md                # Gemini config
└── TASKBOARD.md             # Generated task overview

Keeping Agent Files in Sync

When you update prompts or skills, sync all agent configuration files:

dk sync      # Regenerate all agent files
dk commands  # Regenerate slash commands
dk upgrade   # Update doyaken itself

Project Manifest

Configure your project in .doyaken/manifest.yaml:

project:
  name: "my-app"
  description: "My awesome app"

git:
  remote: "[email protected]:user/my-app.git"
  branch: "main"

domains:
  production: "https://my-app.com"
  staging: "https://staging.my-app.com"

quality:
  test_command: "npm test"
  lint_command: "npm run lint"

agent:
  name: "claude"
  model: "opus"
  max_retries: 2

Task System

Task Priority

Tasks use the naming format PPP-SSS-slug.md:

| Priority | Code | Use For | |----------|------|---------| | Critical | 001 | Blocking, security, broken | | High | 002 | Important features, bugs | | Medium | 003 | Normal work | | Low | 004 | Nice-to-have, cleanup |

Example: 002-001-add-user-auth.md = High priority, first in sequence

Creating Tasks

# Via CLI
dk tasks new "Add user authentication"

# Manually create in .doyaken/tasks/todo/
# Use the template format

Environment Variables

| Variable | Default | Description | |----------|---------|-------------| | DOYAKEN_AGENT | claude | AI agent to use | | DOYAKEN_MODEL | agent-specific | Model for the selected agent | | DOYAKEN_AUTO_TIMEOUT | 60 | Auto-select menu options after N seconds (0 to disable) | | AGENT_DRY_RUN | 0 | Preview without executing | | AGENT_VERBOSE | 0 | Detailed output | | AGENT_QUIET | 0 | Minimal output | | AGENT_MAX_RETRIES | 2 | Retries per phase | | TIMEOUT_EXPAND | 300 | Expand phase timeout (seconds) | | TIMEOUT_TRIAGE | 180 | Triage phase timeout (seconds) | | TIMEOUT_PLAN | 300 | Plan phase timeout (seconds) | | TIMEOUT_IMPLEMENT | 1800 | Implement phase timeout (seconds) | | TIMEOUT_TEST | 600 | Test phase timeout (seconds) | | TIMEOUT_DOCS | 300 | Docs phase timeout (seconds) | | TIMEOUT_REVIEW | 600 | Review phase timeout (seconds) | | TIMEOUT_VERIFY | 300 | Verify phase timeout (seconds) | | DOYAKEN_HOME | ~/.doyaken | Global installation directory |

Troubleshooting

# Health check
dk doctor

# View logs
ls -la .doyaken/logs/

# Reset stuck state
rm -rf .doyaken/locks/*.lock
mv .doyaken/tasks/doing/*.md .doyaken/tasks/todo/

Development

Setup

git clone https://github.com/mitchellfyi/doyaken-cli.git
cd doyaken-cli

# Install git hooks
npm run setup

# Run all checks
npm run check

Quality Scripts

| Script | Description | |--------|-------------| | npm run lint | Lint shell scripts with shellcheck | | npm run validate | Validate YAML files | | npm run test | Run test suite | | npm run check | Run all quality checks | | npm run setup | Install git hooks |

Git Hooks

The repository includes git hooks for quality assurance:

• pre-commit: Lints staged shell scripts and YAML files
• pre-push: Runs the full test suite

To bypass temporarily: git commit --no-verify

Requirements

• At least one AI coding agent CLI installed:
  • Claude Code (default)
  • Cursor CLI
  • OpenAI Codex
  • Google Gemini CLI
  • GitHub Copilot CLI
  • OpenCode
• yq - YAML processor (required for project registry)
• Bash 3.2+ (macOS default works)
• Git
• macOS or Linux
• Node.js 16+ (for npm install)

License

MIT