The deep dive

One brain. Every AI tool your team uses.

Your team shouldn't have to pick one AI tool. Write context once, and it works everywhere - Claude Code, Codex, Gemini, and Cursor.

The problem

Your engineer uses Claude Code. The new hire is trying Codex. A teammate prefers Gemini. The contractor uses Cursor. Each tool has its own context file format - CLAUDE.md, AGENTS.md, GEMINI.md - and none of them talk to each other.

The result: every tool gets a different slice of company knowledge. Or worse, no knowledge at all. The team fragments. Context drifts. Someone writes a CLAUDE.md and the Codex users never see it.

How it works

Antidrift maintains all three context file formats simultaneously. Edit any one of them, run /push, and the cross-compiler syncs the others.

What happens when you /push
# You edit engineering/CLAUDE.md
# Run /push
# Antidrift automatically updates:

engineering/CLAUDE.md    → Claude Code
engineering/AGENTS.md    → Codex, Cursor (partial)
engineering/GEMINI.md    → Gemini

There's no manual syncing. No copy-paste between files. No "did you update the AGENTS.md too?" One edit, one push, every tool gets the update.

Supported tools

Claude Code Full support Reads CLAUDE.md, runs skills
Codex Full support Reads AGENTS.md, runs skills
Cursor Partial Reads AGENTS.md, but auto-load is unreliable
Gemini Full support Reads GEMINI.md
Any markdown-aware tool Works If it reads markdown context files, the brain works

The cross-compiler

Each AI tool has slightly different conventions for context files. CLAUDE.md uses one format. AGENTS.md uses another. The cross-compiler translates between them so you never think about it.

Under the hood, antidrift uses an intermediate representation (IR) - a universal format that compiles to any platform's native format. Skills and context both go through this pipeline:

Compile pipeline
CLAUDE.md → IR → AGENTS.md
CLAUDE.md → IR → GEMINI.md
AGENTS.md → IR → CLAUDE.md

# Works in any direction. Edit any file,
# the others update on /push.

Community skills are stored in IR format too. When you install a skill, it compiles to your tool's native format automatically. A skill written for Claude Code works in Codex without modification.

No lock-in

The brain is a git repo full of markdown files. There's no database, no vendor account, no proprietary format. If you stop using antidrift tomorrow, you still have:

Switch tools, switch vendors, fork the brain - it's yours. That's the point.

Try it

$ npx @antidrift/cli init

One command creates a brain that works with every AI tool on your team.