Autoloop
repo →Problem
I kept running into the same problem: AI coding agents can implement one issue, but they can't reason about the queue. Issue #3 depends on #1 being merged first. Two issues touch the same file and need serial dispatch. Nobody was solving the orchestration layer. So I built Autoloop: a governed pipeline that triages, decomposes, orders, and implements, then hands every PR to a human for the final call.
Pipeline
Key features
- •Config-driven: works on any language or toolchain via
autoloop.toml - •Serial dispatch: respects dependency ordering, avoids merge conflicts
- •One-command setup:
autoloop initon any existing repo - •Protected paths: the loop can't modify itself or core infrastructure
- •Human gate: every PR requires human review before merge
Getting started
$ python autoloop/init.py \
--repo your-org/your-repo \
--target ~/Git/your-repo \
--reviewer your-github-handle \
--verify-cmd "npm run build"
✓ Created autoloop.toml
✓ Created .github/workflows/autoloop-cleanup.yml
✓ Created GitHub labels
✓ Ready to process issues