GitHub-native task execution layer inspired by GSD

[davide-troiani]

Hi all,

I wanted to share an experiment I have been working on, inspired by GSD-style workflows:

https://github.com/ai-is-gonna/get-tasks-done

The problem I am focusing on is reviewability.
As AI coding agents become more capable, the risk is whether their work stays small, explicit, traceable and safe for a human to review.

Get Tasks Done implements a GitHub-native execution model:

spec → planned task → GitHub issue → branch → pull request → validation record → human review

The goal is to make agentic coding fit inside normal engineering delivery instead of bypassing it.

A few design choices I am exploring:

• stronger task decomposition during the planning
• GitHub issues as the execution unit
• one branch and one PR per planned task
• explicit validation evidence before review
• human approval as the final merge
• compatibility with different coding agents since it is built on GSD

I see it more as an adjacent execution layer / workflow experiment around task granularity, review gates, and GitHub-based delivery control. Actually it is helping me A LOT in my projects.

Curious how others here think about this direction.

Cheers

[MicheleBe]

Thanks for sharing. I am particularly interested in stronger task decomposition during the planning. I feel that is the right direction as well as a finer phase decomposition

[davide-troiani]

Glad you agree!
As a tech lead I see it working pretty good, using this approach I have much better control over technical decisions and the code changes which solves A LOT of headaches.

I'm maintaining the repo alone now, hope it gets contribution from more people. It's hard though, because such concepts are often ignored by vibe coders and tech leads / seniors are harder to reach.