🇯🇵 日本語ドキュメント

Task Agent Koordination Tool - Multi-agent orchestration system for Claude Code and OpenAI Codex.

Note: This project is developed at my own pace. See Disclaimer for details.

TAKT is built with TAKT (dogfooding).

• Claude Code or Codex must be installed and configured

TAKT supports both Claude Code and Codex as providers; you can choose the provider during setup.

npm install -g takt

# Run a task (will prompt for workflow selection and optional worktree creation)
takt "Add a login feature"

# Add a task to the queue
takt /add-task "Fix the login bug"

# Run all pending tasks
takt /run-tasks

# Watch for tasks and auto-execute
takt /watch

# Review worktree results (merge or delete)
takt /review-tasks

# Switch workflow
takt /switch

TAKT uses YAML-based workflow definitions. Place them in:

• ~/.takt/workflows/*.yaml

name: default
max_iterations: 10

steps:
  - name: plan
    agent: planner
    provider: claude         # Optional: claude or codex
    model: opus              # Claude: opus/sonnet/haiku, Codex: gpt-5.2-codex/gpt-5.1-codex
    instruction_template: |
      {task}
    transitions:
      - condition: done
        next_step: implement

  - name: implement
    agent: coder
    provider: codex
    model: gpt-5.2-codex     # Codex model example
    instruction_template: |
      {task}
    transitions:
      - condition: done
        next_step: review
      - condition: blocked
        next_step: ABORT

  - name: review
    agent: architect
    model: sonnet            # Model alias (no provider = uses global default)
    transitions:
      - condition: approved
        next_step: COMPLETE
      - condition: rejected
        next_step: implement

TAKT ships with several built-in workflows:

Switch between workflows with takt /switch.

• coder - Implements features and fixes bugs
• architect - Reviews code and provides feedback
• supervisor - Final verification and approval
• planner - Task analysis and implementation planning
• ai-reviewer - AI-generated code quality review
• security - Security vulnerability assessment

Define custom agents in .takt/agents.yaml:

agents:
  - name: my-reviewer
    prompt_file: .takt/prompts/reviewer.md
    allowed_tools: [Read, Glob, Grep]
    provider: claude             # Optional: claude or codex
    model: opus                  # Claude: opus/sonnet/haiku or full name (claude-opus-4-5-20251101)
    status_patterns:
      approved: "\\[APPROVE\\]"
      rejected: "\\[REJECT\\]"

  - name: my-codex-agent
    prompt_file: .takt/prompts/analyzer.md
    provider: codex
    model: gpt-5.2-codex         # Codex: gpt-5.2-codex, gpt-5.1-codex, etc.

You can specify models using either aliases or full model names:

Aliases (recommended for simplicity):

• opus - Claude Opus 4.5 (highest reasoning capability)
• sonnet - Claude Sonnet 4.5 (balanced, best for most tasks)
• haiku - Claude Haiku 4.5 (fast and efficient)
• opusplan - Opus for planning, Sonnet for execution
• default - Recommended model for your account type

Full model names (recommended for production):

• claude-opus-4-5-20251101
• claude-sonnet-4-5-20250929
• claude-haiku-4-5-20250101

Available Codex models:

• gpt-5.2-codex - Latest agentic coding model (default)
• gpt-5.1-codex - Previous generation
• gpt-5.1-codex-max - Optimized for long-running tasks
• gpt-5.1-codex-mini - Smaller, cost-effective version
• codex-1 - Specialized model aligned with coding preferences

~/.takt/
├── config.yaml          # Global config (provider, model, workflows, etc.)
├── workflows/           # Workflow definitions
└── agents/              # Agent prompt files

.takt/                   # Project-level config
├── agents.yaml          # Custom agent definitions
├── tasks/               # Pending task files (.yaml, .md)
├── completed/           # Completed tasks with reports
├── worktrees/           # Git worktrees for isolated task execution
├── reports/             # Execution reports (auto-generated)
└── logs/                # Session logs (incremental)
    ├── latest.json      # Pointer to current/latest session
    ├── previous.json    # Pointer to previous session
    └── {sessionId}.json # Full session log per workflow run

Configure default provider and model in ~/.takt/config.yaml:

# ~/.takt/config.yaml
language: en
default_workflow: default
log_level: info
provider: claude         # Default provider: claude or codex
model: sonnet            # Default model (optional)
trusted_directories:
  - /path/to/trusted/dir

Model Resolution Priority:

1. Workflow step model (highest priority)
2. Custom agent model
3. Global config model
4. Provider default (Claude: sonnet, Codex: gpt-5.2-codex)

When TAKT prompts for additional input during execution (e.g., "Please provide more details"), use the -r flag to continue the conversation:

# First run - agent might ask for clarification
takt "Fix the login bug"

# Resume the same session to provide the requested information
takt -r "The bug occurs when the password contains special characters"

The -r flag preserves the agent's conversation history, allowing for natural back-and-forth interaction.

When running takt "task", you are prompted to:

1. Select a workflow - Choose from available workflows (arrow keys, ESC to cancel)
2. Create a worktree (optional) - Optionally run the task in an isolated git worktree

This interactive flow ensures each task runs with the right workflow and isolation level.

Create your own workflow by adding YAML files to ~/.takt/workflows/:

# ~/.takt/workflows/my-workflow.yaml
name: my-workflow
description: My custom workflow

max_iterations: 5

steps:
  - name: analyze
    agent: ~/.takt/agents/my-agents/analyzer.md
    instruction_template: |
      Analyze this request: {task}
    transitions:
      - condition: done
        next_step: implement

  - name: implement
    agent: ~/.takt/agents/default/coder.md
    instruction_template: |
      Implement based on the analysis: {previous_response}
    pass_previous_response: true
    transitions:
      - condition: done
        next_step: COMPLETE

Agents are specified using file paths in workflow definitions:

# Use built-in agents
agent: ~/.takt/agents/default/coder.md
agent: ~/.takt/agents/magi/melchior.md

# Use project-local agents
agent: ./.takt/agents/my-reviewer.md

# Use absolute paths
agent: /path/to/custom/agent.md

Create custom agent prompts as Markdown files:

# ~/.takt/agents/my-agents/reviewer.md

You are a code reviewer focused on security.

## Your Role
- Check for security vulnerabilities
- Verify input validation
- Review authentication logic

## Output Format
- [REVIEWER:APPROVE] if code is secure
- [REVIEWER:REJECT] if issues found (list them)

TAKT supports batch task processing through task files in .takt/tasks/. Both .yaml/.yml and .md file formats are supported.

# Quick add (no worktree)
takt /add-task "Add authentication feature"

# Interactive mode (prompts for worktree, branch, workflow options)
takt /add-task

YAML format (recommended, supports worktree/branch/workflow options):

# .takt/tasks/add-auth.yaml
task: "Add authentication feature"
worktree: true                  # Run in isolated git worktree
branch: "feat/add-auth"         # Branch name (auto-generated if omitted)
workflow: "default"             # Workflow override (uses current if omitted)

Markdown format (simple, backward compatible):

# .takt/tasks/add-login-feature.md

Add a login feature to the application.

Requirements:
- Username and password fields
- Form validation
- Error handling for failed attempts

YAML task files can specify worktree to run each task in an isolated git worktree, keeping the main working directory clean:

• worktree: true - Auto-create at .takt/worktrees/{timestamp}-{task-slug}/
• worktree: "/path/to/dir" - Create at specified path
• branch: "feat/xxx" - Use specified branch (auto-generated as takt/{timestamp}-{slug} if omitted)
• Omit worktree - Run in current working directory (default)

When a worktree task completes successfully, TAKT automatically commits all changes (auto-commit). Use takt /review-tasks to review, try-merge, or delete completed worktree branches.

takt /run-tasks

• Tasks are executed in alphabetical order (use prefixes like 001-, 002- for ordering)
• Completed tasks are moved to .takt/completed/ with execution reports
• New tasks added during execution will be picked up dynamically

takt /watch

Watch mode polls .takt/tasks/ for new task files and auto-executes them as they appear. The process stays resident until Ctrl+C. This is useful for:

• CI/CD pipelines that generate task files
• Automated workflows where tasks are added by external processes
• Long-running development sessions where tasks are queued over time

takt /review-tasks

Lists all takt/-prefixed worktree branches with file change counts. For each branch you can:

• Try merge - Attempt merge into main (dry-run check, then actual merge)
• Merge & cleanup - Merge and remove the worktree
• Delete - Remove the worktree and branch without merging

TAKT writes session logs incrementally to .takt/logs/. Logs are saved at workflow start, after each step, and at workflow end — so even if the process crashes mid-execution, partial logs are preserved.

• .takt/logs/latest.json - Pointer to the current (or most recent) session
• .takt/logs/previous.json - Pointer to the previous session
• .takt/logs/{sessionId}.json - Full session log with step history

Agents can read previous.json to pick up context from a prior run (e.g., when resuming with takt "続けて").

Available variables in instruction_template:

Each workflow step requires three key elements:

1. Agent - A Markdown file containing the system prompt:

agent: ~/.takt/agents/default/coder.md    # Path to agent prompt file
agent_name: coder                          # Display name (optional)

2. Status Rules - Define how the agent signals completion. Agents output status markers like [CODER:DONE] or [ARCHITECT:REJECT] that TAKT detects to drive transitions:

status_rules_prompt: |
  Your final output MUST include a status tag:
  - `[CODER:DONE]` if implementation is complete
  - `[CODER:BLOCKED]` if you cannot proceed

3. Transitions - Route to the next step based on status:

transitions:
  - condition: done        # Maps to status tag DONE
    next_step: review      # Go to review step
  - condition: blocked     # Maps to status tag BLOCKED
    next_step: ABORT       # End workflow with failure

Available transition conditions: done, blocked, approved, rejected, improve, answer, always. Special next_step values: COMPLETE (success), ABORT (failure).

Step options:

import { WorkflowEngine, loadWorkflow } from 'takt';  // npm install takt

const config = loadWorkflow('default');
if (!config) {
  throw new Error('Workflow not found');
}
const engine = new WorkflowEngine(config, process.cwd(), 'My task');

engine.on('step:complete', (step, response) => {
  console.log(`${step.name}: ${response.status}`);
});

await engine.run();

This project is a personal project developed at my own pace.

• Response times: I may not be able to respond to issues immediately
• Development style: This project is primarily developed using "vibe coding" (AI-assisted development) - use at your own risk
• Pull requests:
  • Small, focused PRs (bug fixes, typos, docs) are welcome
  • Large PRs, especially AI-generated bulk changes, are difficult to review

See CONTRIBUTING.md for more details.

Docker environment is provided for testing in other environments:

# Build Docker images
docker compose build

# Run tests in container
docker compose run --rm test

# Run lint in container
docker compose run --rm lint

# Build only (skip tests)
docker compose run --rm build

This ensures the project works correctly in a clean Node.js 20 environment.

• Workflow Guide - Create and customize workflows
• Agent Guide - Configure custom agents
• Changelog - Version history
• Security Policy - Vulnerability reporting
• Blog: TAKT - AI Agent Orchestration - Design philosophy and practical usage guide (Japanese)

MIT - See LICENSE for details.