What it does

Loadout is a command-line tool that sits in front of AI coding agents. You write reusable pieces of guidance, group them into named kits called loadouts, and tie each kit to a type of project such as Rust, Next.js, Go or Python, or to a bare machine with no repository. When you start an agent through the load command, Loadout works out which kit applies and writes it to a local, gitignored file that the agent reads. Claude, Codex, Cursor, opencode and Copilot are supported, in the command line and in VS Code. Selection is deterministic rather than decided by a model, and committed project files such as AGENTS.md or CLAUDE.md are never edited. A local web studio lets you edit kits visually and review a diff before anything is written, and your kits can be synced across machines through a git repository. Loadout also provides six slash commands for a working process (explore, brainstorm, plan, implement, verify, ship), with bundled workflow styles you can pick or replace. A plan preview turns an agent's plan into a reviewable page where you can leave comments and paste them back to the agent.

Capabilities

Detects the project type and equips one matching kit of context automatically
Works with Claude, Codex, Cursor, opencode and Copilot
Writes only local, gitignored files; committed instruction files stay untouched
Local web studio with diff review before changes are applied
Syncs your kits across machines through a git repository
Plan preview page with comments you can paste back to the agent

When to use it

You switch between several stacks or machines and want each agent session to start with the right conventions.
You use more than one AI coding agent and want to keep a single set of personal instructions for all of them.
You want shared project instruction files left alone while keeping your own preferences separate.