Skip to content

sipag

Autonomous dev agents that evolve with your project.

sipag generates project-aware review agents, ships work through isolated Docker containers, and learns from failures. You write the assignment; workers do the work.

Three commands for humans

sipag configure                           # Configure agents + commands for .claude/
sipag dispatch <PR_URL>                   # Launch a Docker worker for a PR
sipag tui                                 # Monitor all workers

Everything else (sipag ps, sipag logs, sipag kill) is for managing workers from the command line.

How it works

sipag configure      Configure review agents + commands
       ↓
create PR            Describe the work in the PR body
       ↓
sipag dispatch       Launch a Docker worker
       ↓
worker executes      clone → read PR → claude → push
       ↓
review + merge       You decide what ships

Workers run in isolated Docker containers. Claude has full autonomy inside them — no approval dialogs, no interruptions. The container is the safety boundary.

Install

brew tap Dorky-Robot/sipag
brew install sipag

curl -fsSL https://raw.githubusercontent.com/Dorky-Robot/sipag/main/scripts/install.sh | bash

cargo install --path sipag

Part of the dorky robot stack

kubo (think)  →  sipag (do)  →  GitHub PRs (review)
                    ↑
tao (decide)  ─────┘

sipag is the execution layer. kubo handles chain-of-thought planning; tao surfaces suspended decisions.

Get started → How it works →