The quality of your context determines the quality of your AI outputs.
Every AI agent, copilot, and assistant you use is only as good as the context you give it. Most teams dump unstructured docs into a folder and wonder why AI sounds generic. Context OS gives you a better way — a structured, opinionated system for organizing everything AI needs to know about you and your company.
The result: AI that writes in your voice, thinks with your frameworks, and makes decisions like someone who actually works there.
AI tools are powerful, but they start from zero every time. They don't know your voice, your strategy, your customers, or how you make decisions. So you get:
- Generic outputs that could have been written by anyone
- Repeated prompting — re-explaining the same context over and over
- Inconsistent quality — some outputs are great, most are mediocre
- Knowledge trapped in people's heads — when someone leaves, the context leaves with them
The fix isn't better prompts. It's better context.
Context OS is a starter kit — a ready-to-fork repository structure that organizes your company's knowledge into a format AI agents can actually use. You fill in the templates with your information, and every AI tool you connect to it immediately gets smarter about your business.
Built for Claude Code, Claude Cowork, and Cursor. The context is structured in plain markdown, so it works with any AI tool that can read files — but these three get first-class support.
├── context/ How AI should behave (voice, roles, personal style)
├── docs/ What's true about your company (handbook, strategy, products)
├── knowledge/ Deep reference material (study guides, domain expertise)
├── agent-docs/ Task-specific agent configurations
├── pipeline/ Work-in-progress (research → scratchpad → outputs)
├── records/ Historical archives (transcripts, customer notes)
├── sources/ Trusted people and sources indexes
├── prompts/ Reusable prompt templates
├── scripts/ Automation scripts
├── tests/ Context smoke tests
├── CLAUDE.md Claude Code entry point (loaded automatically)
├── AGENTS.md Cross-platform agent config
└── SETUP.md Setup guide
Most knowledge bases are flat — a pile of docs with no hierarchy. AI agents waste tokens reading irrelevant content, miss important context, and produce inconsistent results.
Context OS is designed around three principles:
-
Prescriptive context (
context/) tells agents HOW to act — your voice, your decision-making frameworks, your leadership style. This is what makes outputs sound like you instead of sounding like AI. -
Reference context (
docs/,knowledge/) tells agents WHAT IS true — your business model, your customers, your strategy. This is what makes outputs accurate instead of generic. -
Progressive disclosure — agents scan summaries first, then load full files only when needed. Every directory has a
README.md(what's here) andINDEX.md(file listing with summaries). No token waste.
The result is AI that has the right context at the right time, without being overwhelmed by everything at once.
- Fork or clone this repo
- Open
CLAUDE.md— replace[YOUR COMPANY]with your company name and fill in the placeholders - Define your voice — see
context/voice/writing-style-context-v2.mdfor the GrowthX voice guide - Add your basics — fill in
docs/company/mission-and-vision.mdwith your mission and vision - Explore roles — browse 15 pre-built AI personas in
context/roles/(CEO, Marketing, Sales, Finance, etc.) - Test it — open Claude Code, Claude Cowork, or Cursor pointed at your repo and ask it a question about your company
For the full setup guide, see SETUP.md.
AI agents read CLAUDE.md first (loaded automatically). Based on the task, they load a specific agent config from agent-docs/ which tells them exactly which context files to read. Agents only load what's relevant — no wasted tokens on irrelevant context.
Every directory has README.md (what's here, why) and INDEX.md (file listing with summaries). Start at docs/start-here.md for onboarding.
Work flows forward only: research/ → scratchpad/ → outputs/. Raw research gets refined into working drafts, then polished into final deliverables. This mirrors how good thinking actually works.
| Tool | Configuration |
|---|---|
| Claude Code | CLAUDE.md (loaded automatically) |
| Claude Cowork | CLAUDE.md (loaded automatically) |
| Cursor | .cursor/rules/ and .cursor/skills/ |
| Other AI tools | AGENTS.md (cross-platform standard) |
- Format:
descriptive-name-v1.md(lowercase, hyphens, version suffix) - Minor updates: Edit in place, update
last_updatedin metadata - Major changes: Create new version (
-v2), move old to/archive
- SETUP.md — Full setup guide with context engineering primer
- agent-docs/context-engineering-guide.md — How to write content that AI agents can effectively use
- docs/context-routing.md — Detailed routing rules for context loading