antidrift

Company brain for you and your AI agents. Shared knowledge, shared skills, no more passing files around.

Works with Claude Code, Claude Cowork / Desktop, OpenAI Codex, Cursor, and Google Antigravity.

The Problem

Every time you start a new session, your AI agent has no idea who you are, what your company does, who your customers are, or how you work. You repeat yourself constantly. Context drifts. Knowledge gets lost.

The Fix

Antidrift is a git repo that gives your AI agents full context on your company. Product, customers, stack, positioning, decisions, standards — all of it. Clone it on any machine and your agent knows everything.

Two things in the box:

1. The brain — markdown files organized by department. Your agent reads them automatically.
2. Skills — slash commands that do real work: /write, /review, /icp, /prep, /tps, and more.

Install

npm install -g @antidrift/cli

Or via Python:

pip install antidrift

Works on macOS, Linux, and Windows (CMD, PowerShell, Git Bash).

Quick Start

antidrift init

That's it. Your agent loads the brain and knows where everything is. Type /ingest <path> to start building context from your existing files.

What's In the Brain

your-brain/
├── CLAUDE.md              # Company overview, priorities, navigation
├── AGENTS.md              # Same content — for Codex users
├── product/               # What you're building, roadmap
├── customers/             # Account data, meeting notes, agreements
├── marketing/             # Positioning, brand voice, campaigns, ICPs
├── sales/                 # Pipeline, prospects, outreach
├── engineering/           # Architecture, coding standards, specs
├── finance/               # Revenue, expenses
├── decisions/             # Why we made the choices we made
└── projects.md            # Project registry with repo URLs

Add a department by creating a directory. Nest as deep as you want. Your agent discovers it all.

Skills

Skills are slash commands. Type / to see them.

Core (ships with every brain)

Skill	What it does
/ingest <path>	Import files and directories into the brain
/push	Commit and push changes (syncs brain files, merges additively)
/refresh	Pull latest changes from remote
/onboard	Walk a new person through everything
/publish <skill>	Share a skill with the community

Community Skill Packs

antidrift skills add essentials         # decision, recap, write, prep, followup, status, search
antidrift skills add engineering        # review, spec, changelog, standards, tps
antidrift skills add customer-research  # icp, voc, twins
antidrift skills add email              # inbox (classify, move, archive, clean)
antidrift skills add legal              # legal document generator
antidrift skills add all              # everything

Pack	Skills
essentials	/decision — log decisions with context and reasoning
	/recap — what changed since last time
	/write — write anything using brand voice + brain context
	/prep — prep for a customer call from account data
	/followup — draft follow-up emails after meetings
	/status — brain health, freshness, gaps
	/search — search across the entire brain
engineering	/review — code review using team standards from the brain
	/spec — technical spec using architecture context
	/changelog — generate changelog from git history
	/standards — generate coding standards from existing code
	/tps — TPS report from git history (yes, that TPS report)
customer-research	/icp — build research-backed Ideal Customer Profiles
	/voc — Voice of Customer analysis in their actual language
	/twins — create AI twin personas for market testing
email	/inbox — classify, label, move, archive, and clean your inbox (requires mcp-gmail)
legal	/legal — generate NDAs, subscription agreements, PSAs from templates

Browse what's available:

antidrift skills list

How It Works

Your Agent Reads the Brain Automatically

Every directory has a brain file (CLAUDE.md for Claude Code, AGENTS.md for Codex). Your agent loads it when working in that area. Only what's relevant gets loaded — not the whole brain at once.

Skills Are Just Markdown

A skill is a directory with a SKILL.md file:

---
name: my-skill
description: What this skill does and when to use it
---

Instructions for your agent go here. Markdown.

No build step. No framework. No config.

Two Layers of Skills

Company skills live in the brain repo at .claude/skills/. Everyone gets them via git.

Personal skills live on your machine at ~/.claude/skills/. Never shared unless you choose to. Personal skills override company skills if they share a name.

When a personal skill is useful for the team: copy it to the brain repo and /push.

Keeping the Brain Efficient

Brain files are indexes, not documents. Keep them under 250 lines — summaries and pointers.

Heavy reference material goes in _reference/ subdirectories. Your agent won't read it unless asked.

marketing/
├── CLAUDE.md              # Summary — auto-loaded (~150 lines)
├── pricing-structure.md   # Active reference
└── _reference/            # Heavy files — only on demand
    ├── twin-tests/
    └── voc/

Supported Platforms

Antidrift works across all major AI coding platforms on the same brain.

Platform	Brain File	Config File
Claude Code	CLAUDE.md	.mcp.json
Claude Cowork / Desktop	CLAUDE.md	claude_desktop_config.json
OpenAI Codex	AGENTS.md	—
Cursor	AGENTS.md	—
Google Antigravity	GEMINI.md	—

• antidrift init creates CLAUDE.md, AGENTS.md, and GEMINI.md
• /push syncs them before every commit
• /ingest creates all brain files for every department
• Community skills compile to the right format on install
• Teams can mix platforms — everything stays in sync

Skill Compiler

Convert skills between platforms:

antidrift cross-compile .claude/skills/my-skill --to codex
antidrift cross-compile .agents/skills/my-skill --to claude

Connect Services

Connect external services as MCP servers — your AI agent gets live access to your tools.

antidrift connect google                # Google Workspace (all-in-one)
antidrift connect gmail                 # Gmail only
antidrift connect drive                 # Drive, Docs, Sheets
antidrift connect calendar              # Calendar only
antidrift connect stripe                # Stripe
antidrift connect attio                 # Attio CRM
antidrift connect github                # GitHub

Platform targeting

By default, connect installs for Claude Code. Use flags to target other platforms:

antidrift connect stripe --cowork       # Claude Cowork / Desktop
antidrift connect stripe --all          # All detected platforms
antidrift connect stripe --claude-code  # Claude Code (explicit)

Packages

Package	What It Does
@antidrift/cli	Unified CLI
@antidrift/core	Brain + core skills + setup
@antidrift/skills	Community skill registry
@antidrift/mcp-google	Google Workspace — all-in-one (Sheets, Docs, Drive, Gmail, Calendar)
@antidrift/mcp-gmail	Gmail (search, read, send, reply, labels, archive, drafts)
@antidrift/mcp-drive	Drive + Docs + Sheets (files, folders, docs, spreadsheets)
@antidrift/mcp-calendar	Calendar (upcoming, search, create, today)
@antidrift/mcp-stripe	Stripe (invoices, customers, subscriptions, charges, payment links)
@antidrift/mcp-attio	Attio CRM (people, companies, deals, tasks, notes)
@antidrift/mcp-github	GitHub (repos, issues, PRs, actions, releases, traffic)

Learn More

antidrift.io

Contributing

See CONTRIBUTING.md for guidelines on sharing skills and developing core.

License

MIT