nrslib/takt

TAKT Agent Koordination Topology - Define how AI agents coordinate, where humans intervene, and what gets recorded — in YAML

What it solves

TAKT is designed to stop the "babysitting" of AI coding agents. It addresses common issues where agents forget instructions, pollute their own context, or skip critical review steps in long-running development tasks. Instead of relying on a single agent to manage a complex process, TAKT enforces a structured, repeatable development workflow from the outside.

How it works

TAKT is a CLI tool that orchestrates multiple AI agents using YAML-defined workflows. It separates the development process into explicit steps (e.g., planning, implementation, review, and fixing). Each step is assigned a specific persona, set of permissions, and transition rules.

Key technical mechanisms include:

  • Isolated Worktrees: Tasks are executed in isolated Git worktrees to prevent polluting the main working directory.
  • Workflow Engine: A YAML-based system that manages transitions between steps (e.g., routing a failed review back to the implementation step).
  • Provider Agnostic: It coordinates various AI providers such as Claude Code, Cursor, GitHub Copilot CLI, and SDK-based options like Codex and OpenCode.
  • Auto-routing: An optional feature that routes specific workflow steps to the most appropriate model based on cost, performance, or reasoning requirements.

Who it’s for

TAKT is for developers who use AI coding agents for substantial work and want to move from ad-hoc prompting to professional, reproducible software engineering processes. It is also suitable for teams wanting to integrate AI agents into CI/CD pipelines via GitHub Actions.

Highlights

  • Declarative Workflows: Define the entire dev loop (Plan $\rightarrow$ Implement $\rightarrow$ Review $\rightarrow$ Fix) in YAML.
  • Context Management: Prevents context bloat by giving each agent only the information needed for its specific role.
  • Multi-Agent Coordination: Supports a variety of CLI and SDK-based AI agents as interchangeable providers.
  • Traceability: Maintains logs and reports for every step, ensuring the path from task to Pull Request is auditable.
  • Integration: Built-in support for GitHub and GitLab issues to queue tasks directly from project management tools.