$ man context-wiki/context-repository

A context repository is a collection of markdown files in a GitHub repo that contain everything your AI systems need to know. Products, personas, messaging frameworks, voice guides, workflow documentation, partner details. You update these files once, and every AI agent that touches your repo gets the latest version automatically. No copy-pasting context into chat windows. No re-explaining your business every session. The repo IS the context. My entire GTM OS repo is a context repository. Every skill file, every voice guide, every partner research folder. It is not a separate thing I maintain alongside the work. It is the work itself.

PATTERN

Stage 1: I do, AI helps with drafting. You write the content, AI suggests edits. Most people live here. Stage 2: AI monitors and surfaces drift. AI watches your repo for inconsistencies. Messaging changed but the email templates did not update? AI flags it. Stage 3: AI makes changes, I approve via PR. AI proposes updates, you review and merge. This is the sweet spot. 90% efficiency with 100% control. You are not handing over the keys. You are reviewing the work. Stage 4: AI does everything, no human review. Full autopilot. Risky for anything customer-facing. Useful for internal automation like dashboards and data syncs. I operate at Stage 3 for most workflows. Claude writes the code, the content, the campaign copy. I review it, adjust it, approve it. The context repository makes Stage 3 possible because Claude has enough context to produce work worth reviewing.

PATTERN

At minimum: a CLAUDE.md (environment defaults and code style), a voice guide (how you sound), product docs (what you sell), ICP definitions (who you sell to), persona tiers (who you talk to), and workflow documentation (how you operate). Beyond the minimum: partner folders with research, skills for every repeated workflow, rules files for file-specific patterns, content drafts and published archives, and campaign copy templates. The more you put in, the less you have to explain. The less you explain, the faster you move. My repo has 40+ skills, a full voice system, and research folders for every partner. Claude does not need a briefing. It reads the files.

PRO TIP

Markdown is plain text with simple formatting. It is readable by humans, parseable by AI, version-controllable in Git, and renderable as HTML. There is no proprietary format to worry about. No lock-in to a platform. No conversion step. You write it in any text editor, commit it to Git, and every AI tool can read it natively. I have seen teams try to use Notion or Google Docs as their context layer. The problem is access. AI agents cannot easily read Notion pages mid-workflow. They can read a markdown file in your repo instantly. Keep it simple. Keep it in the repo.