Switch agents.
Keep your setup.
One source for your harness configuration, rendered for Claude Code, Codex, Cursor CLI, Factory Droid, OpenCode, and Pi.
Configure once. Use across harnesses.¶
Keep shared instructions and rules together. Loadout writes each harness's native configuration, so adding another harness doesn't mean rebuilding your setup.
Give each project the context it needs.¶
Keep project-specific context out of your global instructions and token budget. Reuse kits for instructions, skills, and tools, and update projects from a shared source.
Compose it your way.¶
Combine organisation, team, and personal sources. Keep global defaults and project configuration separate. Use profiles and ordered fragments to assemble the setup you need.
A shared source, native configuration¶
A project names its harnesses and the instruction fragments they should read:
Write the instructions in loadout/instructions/project.md and testing.md, then run:
Loadout composes AGENTS.md, writes a CLAUDE.md that imports it with @AGENTS.md, and
renders the project's permission rules into each harness's format. Edit the source once, sync it,
and use either harness.
Follow the complete walkthrough.
Start with the setup you need¶
- For your own harnesses: global setup applies configuration across projects.
- For a shared repository: project setup keeps shared sources in git and your personal permission rules local.
- For several similar projects: kits let you maintain a common configuration without filling every session with unrelated instructions.
Loadout renders configuration. Your harnesses still decide how to load and enforce it, and some features have different support across harnesses. The support reference explains those differences.