Defined stages
Planning, implementation, checks and fixes each have a clear input and result.
Engineering tool
Agentflow
A Go CLI that guides AI-assisted development through explicit stages, checks and human decisions.
Problem
Long automated runs are hard to review and recover
When an AI coding tool changes too much at once, it becomes difficult to see where it went wrong or return to a known state. Agentflow splits the work into stages and records the result of each one.
Human approval
The operator accepts changes at the points that require judgment.
Git checkpoints
Each important stage can be restored without rebuilding the whole run.
Diagnostics
Run state, doctor/repair commands and saved artifacts help locate failures.
Workflow
From task to review — and back to a checkpoint when needed
Current state
The core development loop is implemented
The Go CLI runs staged tasks with guards, Git checkpoints, automated checks, review/fix loops and diagnostics. A person still approves changes and decides when to retry or roll back.
The source code and internal run data are private.