I'm nrslib, the maker of TAKT. I built TAKT because I got tired of babysitting AI coding agents. They're powerful, but in long-running work they forget instructions, blur the line between implementing and reviewing, and I kept repeating the same feedback over and over. That wears you down. Adding more rules to prompts or CLAUDE.md helps, but it can't *enforce* a process — whether the rules are followed is still left to the agent. TAKT flips this around: the workflow controls the agents from the outside. You define plan → implement → review → fix loops in YAML. Each step gets its own persona, policies, and output contracts, so context stays focused instead of polluted. Reviews can't be silently skipped — findings route work back to fix steps, and human judgment can be requested when it matters. Tasks run in isolated worktrees, and every step leaves logs and reports, so the path from task to PR stays traceable. A side effect I like: since quality comes from the process rather than the model alone, review loops lift output quality even with weaker models. It works with Claude Code, Codex, OpenCode, Cursor, GitHub Copilot CLI, and Kiro. And TAKT is built with TAKT itself — every PR goes through its own review workflow. TAKT took off in Japan first — 1.2k+ GitHub stars, 18k+ npm downloads a month, and a 700-member Discord, with engineers and companies writing their own guides for it. Today I'm excited to bring it to the rest of the world. It's open source (MIT). Try it in 5 minutes: npm install -g takt I'd love to hear how you run your AI coding workflows — I'm here all day to answer questions!

Forum Threads

p/takt-3

•

21d ago

How do you keep AI coding agents from going off the rails on long tasks?

I'm nrslib, maker of TAKT (launching here Aug 9).

I built TAKT because agents kept forgetting my instructions in long-running work implementing and reviewing blurred together, and I kept repeating the same feedback over and over.

View all

TAKT

Stop babysitting AI coding agents — reviews can't be skipped

Stop babysitting AI coding agents — reviews can't be skipped

Forum Threads

How do you keep AI coding agents from going off the rails on long tasks?

Forum Threads

How do you keep AI coding agents from going off the rails on long tasks?