Engineering tool · 2026

Agentflow

A CLI workflow that makes AI-assisted development bounded, reviewable and recoverable.

Status
In active development
Role
Product, workflow, architecture and Go
Stack
Go, Cobra, YAML, Git
Source
Private
Synthetic Agentflow workflow with gated development stages

Problem

Speed without control creates expensive rework

AI coding tools can produce changes quickly, but a long unconstrained run makes intent, failure points and rollback hard to inspect. Agentflow turns delivery into explicit stages with artifacts and decisions between them.

Bounded stages

Planning, implementation, checks and remediation have explicit inputs and outputs.

Human gates

Acceptance stays with the operator at points where judgment matters.

Checkpoints

Git-backed checkpoints make recovery a normal workflow action.

Diagnostics

Structured run state, doctor/repair commands and artifacts make failures inspectable.

Workflow

A loop with visible evidence

Task moves through planning, implementation and checks to human review, with checkpoints and artifacts
Review and rollback are part of the happy path, not emergency procedures.

Current state

A tested orchestration layer, not an autonomy claim

The Go repository implements staged runs, guards, checkpoints, checks, review/fix loops, diagnostics and CI support. Its test suite passed during the August 2026 review. The tool orchestrates a CLI and human decisions; it does not promise autonomous or defect-free software delivery.

The source repository and internal run data remain private. No productivity percentage is claimed without a controlled comparison.

Next project

GophKeeper

A learning project exploring secure client-server storage and synchronization in Go.

Read case study →