Agent control · ConfigureOpen source · MIT

Dotfiles for your coding agent.

Outfitter launches your agent CLI from profiles: complete, reproducible loadouts you can layer, lint, version, and share.

$ npm install -g @ai-outfitter/outfitter

Artifact

Resolve, merge, launch.

Every run assembles a composite profile from your layers, shows you what merged, and launches the agent with exactly that configuration. Nothing ambient, nothing forgotten.

terminal · outfitter v0.8

$ outfitter run --profile engineer
resolving profile engineer
profile layer engineer github:ai-outfitter/default-profiles/profiles
merged controls
prepared composite profile /tmp/outfitter-engineer-pi-Gv76Wx
launching pi …
 
$ outfitter profile list
data-analyst
engineer
founder
media-editor

How it works

Profiles compose everything

Context, system prompts, models, skills, extensions, subagents, and DeepWork workflows: one YAML loadout per kind of work.

Layers, not forks

Project profiles inherit your home defaults and append to their prompts instead of replacing them. Personal taste underneath, project rules on top.

Share from any git repo

Teams publish profile catalogs; pin a tag, allowlist what you reviewed, and outfitter sync keeps everyone current. Org defaults merge below local settings; local always wins.

Run headless in CI

The same profile runs in GitHub Actions as a scheduled or event-driven agent (a PR reviewer, a task runner), even with no API keys, on GitHub Models.

Quickstart

Install.

terminal

$ npm install -g @ai-outfitter/outfitter

Or run it without installing via npx @ai-outfitter/outfitter (details on GitHub). Pi first; Claude Code supported with documented gaps; more adapters coming.