Two steps. Download the Blueprint as a reference card, then paste the audit prompt into your existing system and let it analyse itself.
Step 1: Download the May 2026 Blueprint as Reference
You're going to download the Blueprint, not open it as your main project yet. Use it as a reference card while you audit your existing system.
- Open the AI OS Blueprint we sent you. Check your email for the zip file.
- Unzip it somewhere on your computer.
- If you're an advanced AI user — have a scan through the structure.
The key files to look at:
| File | What It Teaches You |
|---|
CLAUDE.md | The full architecture, build path, autonomy ladder, and conventions |
context/ (6 files) | What a well-structured Context layer looks like |
reference/architecture.md | The 3-layer architecture explained in depth |
reference/decision-engine.md | The decision learning loop |
reference/autonomy-ladder.md | The 4-phase progressive autonomy framework |
reference/mcp-guide.md | Connecting data sources with zero code |
reference/scheduling.md | GitHub Actions workflows (canonical scheduling pattern, May 2026) |
reference/channels.md | Three-path mobile access (Slack, Telegram, Claude mobile app) |
.claude/commands/ | The 5 core slash commands, plus /audit-system and more |
Step 2: Paste the Audit Prompt Into Your Existing System
Open your existing setup. Wherever you'd normally chat with the AI, paste the prompt below into a new thread and send it. Don't edit it. It's tuned to work in Claude Code, Claude web, ChatGPT, or any AI tool with file access.
Our recommendation would be using VS Code & Claude Code, as per Module 2.
💬 Prompt:I've just been trained on a comprehensive AI Operating System architecture called the AI OS Blueprint, taught inside the 7 Figure AI Systems Accelerator. I want you to audit our current system against this architecture, identify what we're doing well, what we're missing, and create a migration plan.
Do NOT overwrite or delete anything. This is an audit and planning exercise only.
---
THE ACCELERATOR'S AI OS ARCHITECTURE (April 2026)
Your AI Operating System is built in VS Code with the Claude Code extension. Four tools form the stack: VS Code + Claude Code (workspace and AI), Supabase with pgvector (database with vector search), GitHub (version control + GitHub Actions for scheduled workflows), Vercel (hosting for sellable systems).
THE 3 LAYERS ON AN AI OS FOUNDATION
Each layer depends on the one below it.
AI OS (Foundation)
- A CLAUDE.md at the root of the project. The AI reads it at the start of every session. It is the system's job description and operating manual.
- Slash commands in .claude/commands/ for structured workflows: /prime, /build-context, /decide, /create-plan, /implement, /audit-system.
- The system is self-aware. It knows it is an AI OS, understands the architecture, and can guide you.
Layer 1: Context (The Brain)
Six context files teach the AI the business:
- personal-info.md: founder's role, goals, decision-making style, what stays human vs. automated
- companies.md: business portfolio, products, revenue models, how businesses connect
- team.md: team directory, communication frequency rules, delegation boundaries
- strategy.md: decision filters, current priorities, success criteria
- current-data.md: key metrics with current values, targets, trends
- integrations.md: every tool and service the business uses, mapped by connection priority
Layer 2: Data (The Nervous System)
Live data connections. Two paths: MCP for tools that support it (zero code), and sync scripts for tools that don't (custom code pulling into Supabase). Both are valid. Managed connectors on claude.ai cover Notion, Gmail, Google Drive, Google Calendar, and Slack out of the box. Everything else uses manual MCP via .mcp.json with tokens in .env.local.
The database is Supabase with pgvector for vector search. Hybrid search combines pgvector (semantic) with tsvector (keyword) and RRF (reciprocal rank fusion) for the best of both.
Priority order for agencies: (1) meeting transcripts, (2) CRM and sales, (3) financials, (4) team communication, (5) email, (6) project data, (7) marketing, (8) community.
Layer 3: Function (The Muscles)
Systems that act on Context plus Data:
- Daily Brief: morning intelligence report (financial pulse, pipeline, client health, team pulse, meeting follow-ups, priorities)
- Meeting Intelligence: transcripts synced, summarised, action items extracted
- Task Automation: meeting actions becoming tracked tasks
- Content Pipeline: ideas and drafts from business data
- Decision Engine: logs every significant decision, learns your patterns
Function automation runs on GitHub Actions workflows (May 2026 canonical pattern; replaces the earlier "Routines" guidance, which couldn't reach Supabase or run TypeScript sync scripts reliably). Workflows live in .github/workflows/, run on cron schedules, hold secrets in GitHub Settings, and can chain output steps into Slack (via incoming webhook), email, or Notion in one run.
Mobile access has three paths: (1) Slack bot, outbound via incoming webhook from a workflow, optional inbound @mention replies via Slack app + Cloudflare Worker; (2) Telegram bot via Anthropic plugin or serverless webhook; (3) Claude mobile app via Remote Control paired to your VS Code session.
THREE LEARNING LAYERS
Three complementary layers of learning sit across the architecture:
- CLAUDE.md: declarative rules and identity, edited deliberately
- Auto Memory: observed patterns saved to ~/.claude/projects/<project>/memory/, enabled by default in current Claude Code
- Decision Engine: structured decision logs via /decide, with expected-vs-actual outcome tracking
CORE FRAMEWORKS
1. Decision Learning Engine (5-step loop):
Decide, Log (what, why, options, expected outcome), Match (find similar past decisions), Learn (30-day outcome review), Earn (system earns autonomy through accuracy).
2. Progressive Autonomy Ladder (4 phases):
Phase 1: Inform (present data, human acts)
Phase 2: Recommend (suggest actions with reasoning)
Phase 3: Confirm (draft actions, human approves or rejects)
Phase 4: Autonomous (execute within guardrails)
Guardrails: financial transactions and external client comms ALWAYS require approval. Internal messages, calendar, and data analysis can graduate to autonomous. Trust earned through >80% acceptance for Recommend to Confirm, >90% for Confirm to Autonomous, over 50-100+ decisions.
3. "You Cannot Skip Layers":
Most people jump straight to Function ("automate this task"). Without Context, the system doesn't know why. Without Data, it can't measure if it worked. Build from the ground up.
CONVENTIONS
- Plans live in plans/ with dated filenames
- Outputs organised in outputs/ (briefs/, analyses/, decisions/)
- Decision logs in data/decisions/ with dated filenames
- AI model selection: Opus for strategic analysis, Sonnet for quality tasks, Haiku for classification or routing
- Context files are the source of truth. Update them when reality changes.
---
NOW AUDIT OUR SYSTEM
Please do the following:
1. MAP OUR CURRENT STRUCTURE
- List every file and directory in our project (or describe the tools and workflows we have if we are not file-based yet)
- Categorise each into which layer it belongs to: AI OS (foundation), Context, Data, Function, or Uncategorised
2. LAYER-BY-LAYER ASSESSMENT
For the foundation plus each of the 3 layers, assess:
- What we have that maps to this layer
- What's strong (doing well or better than the Blueprint)
- What's missing or incomplete
- What's misplaced (exists but is in the wrong location or structure)
3. CONTEXT LAYER DEEP DIVE
Compare our context or configuration against the 6 Blueprint context files. For each file, say whether we have something equivalent, whether it's thin or rich, and what's missing:
- personal-info.md: founder philosophy, decision-making style, human vs automated boundaries
- companies.md: business portfolio, revenue models, priority order
- team.md: team directory, relationship health rules, delegation boundaries
- strategy.md: decision filters, strategic priorities
- current-data.md: metrics with targets and trends
- integrations.md: map of all tools and connection status
4. FRAMEWORKS ASSESSMENT
- Decision Engine: are we logging decisions? Do we have a review loop? Are we tracking outcomes?
- Autonomy Ladder: is progressive autonomy built into our architecture? Do we have guardrails defined?
- Three learning layers: do we have CLAUDE.md (or equivalent)? Is Auto Memory in use? Is the Decision Engine running?
- Build path: did we build in the right order, or did we skip layers?
5. MIGRATION PLAN
Create a prioritised, step-by-step plan to upgrade. For each step:
- What to do (specific action)
- Why it matters (what it unlocks)
- Risk level (safe, needs care, or review first)
- Whether it is additive (adding something new) or restructuring (moving or renaming existing work)
Prioritise additive changes first (low risk, immediate value) before any restructuring.
6. QUICK WINS
List 3-5 things we can do RIGHT NOW (in this session) that would have the biggest impact on our system's quality, with zero risk of breaking anything.
IMPORTANT: Do not make any changes yet. Present the full audit and plan first. I'll decide what to action.
💡 Using a plain chatbot without file access? Paste in a description of your current setup instead. The prompt will still give you a useful audit, just thinner on the file-mapping step.
