ai-setup

Why we included this project

Teams that lean on coding agents find out fast that hand-written context files go stale. A refactor renames a module, a new dependency lands, and the CLAUDE.md or AGENTS.md you wrote last month is now describing a repository that no longer exists. Caliber treats those files as something to audit and regenerate rather than maintain by hand. It scores your current setup against the actual filesystem, then proposes diffs you can accept, refine in chat, or decline. Because it covers Claude Code, Cursor, Codex, OpenCode, and GitHub Copilot at once, one command keeps every agent on the team working from the same current instructions, skills, and MCP server configs. Everything runs locally through your own API key and writes nothing without a backup and an undo path, so it is safe to adopt on a shared codebase. If you are tired of correcting an agent that keeps guessing at old paths and outdated dependencies, this is a practical way to keep agents grounded in what the code actually looks like today.