colonizer
install

modules

add what you need. the rest stays off.

colonizer runs a coding agent in a sandbox for each task. modules are the parts you choose around it: which agent, which models, the skills it can load, chores on a schedule, and add-ons. you turn them on in the cockpit, and most start off.

settlers

the coding agent that works inside each colony. you pick one; every one speaks the same protocol, so nothing else changes.

model providers

where the model answers come from. keys stay on your machine; a colony only ever holds a placeholder.

skill packs

libraries of instructions a claude code settler can reach for. pinned, read-only, outside the code the pull request is made from.

token savers

switches on the claude code settler that spend fewer tokens on the same work. all off by default.

loops

chores that run on a schedule. each is off until you switch it on and name the repositories it may touch.

security

point bug hunters at your own repository, and screen what a colony publishes.

add-ons

notes, notifications, a stand-in to answer questions, logs for your own dashboards, and more.

code hosts

where the work comes from and where the pull request goes.

what the badges mean

  • built-in ships with colonizer and is on without you doing anything.
  • optional ships with colonizer, off until you switch it on.
  • tests only ships with colonizer, but is covered by tests only: it has not yet done real work in a colony.
  • planned not usable yet.

keeping them current

every bundled skill pack is pinned by sha256. a daily job checks each one against its upstream and proposes the new pins, with the skills added, removed and changed; staging runs every check a real install does, so an update that fails them never becomes a proposal. it never merges: a person reads and merges.

checked against the harness at ccf738f (v0.2.7) on 2026-10-06. the harness is the source of truth: Colonizer-dev/harness.