Good morning.

Hi everyone, I’m Volodymyr, a product and engineering leader working across AI, cloud platforms and IoT. I’m hands-on with building and operating systems, and I use Claude Code and Codex extensively in my own development work.

I’m building taskPlane, an open-source workflow harness for AI-assisted software delivery inside Claude Code and Codex. It brings requirements, design, task planning, coordinated agents, verification and engineering reviews together in one workflow.

The idea is to make scope, decisions and evidence part of execution, rather than hoping an agent remembers the plan. Hooks support workflow checks, with clear criteria for when work is ready to start and what counts as done. A shared dashboard makes progress, dependencies, findings and outstanding decisions visible.

A lot of this comes from using taskPlane on real builds and dealing with the failures along the way: agents drifting from the plan, declaring work complete without enough evidence, or getting stuck in overly restrictive checks. I’m working toward a balance where governance is meaningful without slowing every step down.

I’d love to compare notes with others building agent harnesses or using coding agents for substantial projects, especially around hooks, recovery and verifying completed work.

taskPlane on GitHub