Client Intelligence OS

A structured "brain" for one advertising client/account. Open the folder in any LLM agent (Claude Code, Cursor, Codex, ChatGPT) and it starts every conversation with full account + business context — so the answers to any question are grounded.

This is the core of the whole system. The more context the LLM has, the better it manages the ads. This template is used by everyone:

• In-house teams / solo managers of one company → use this folder on its own.
• Agencies → the Agency Intelligence OS wraps many of these (one per client) under clients/. Same OS, just structured for multiple accounts (see Connection below).

What this OS is (and isn't)

This OS is just two things: the context structure + the onboarding. Nothing else. Connectors, credentials, fetch scripts, and reporting tools live separately (the Setup/ folder + their own explainers) so the template stays clean and standardized for any client.

How the context gets built: fetch first, then ask

Onboarding (_setup/onboarding.md) fetches the account reality first via the connectors (Google Ads / Meta / GA — credential-based scripts that live separately in Setup/connectors/): campaign structure, spend, conversion actions, and recent spikes. Then it asks the human only what the data can't tell — how conversions are really tracked (margin vs revenue, primary action), what caused anomalies, and the business context. If the connectors aren't set up, it simply asks. Result: the richest, most standardized client context we can produce, the same way every time.

What you get

A 5-folder brain, deliberately simple:

| Folder | What lives here | Changes | |---|---|---| | 00-context/ | One file per domain: business + unit economics, audience, offer & messaging, channels, measurement & tracking, competitors, creative, constraints, history, operations, tech-stack, naming conventions. | Rarely | | 01-goals-budget/ | North-star (north-star.md), targets (KPIs, guardrails, real-goal mapping — targets.md), budget (budget.md), 90-day roadmap (roadmap.md). | Monthly/quarterly | | 02-reporting/ | Self-contained reports (one folder each: data + script + HTML). | Often | | 03-notes/ | Raw captures (calls, threads). Distilled into 00-context/ via the update ritual. | Constantly | | 04-automations/ | Recurring scripts / scheduled tests (e.g. A/B tests from the AB-Test skill). | As needed |

Plus three front-door files:

• CLIENT.md — the 5-second identity card.
• AGENTS.md — short router that tells any agent how to use the folder.
• CLAUDE.md — one-line stub pointing at AGENTS.md.

Quick start

1. Create the folder — either run

   npx @ads-repo/project-intelligence-os <your-company>

   or copy template/ and rename it to your client (e.g. acme/). The npx route also wires the onboard-client skill into Claude Code, Codex and Cursor so it's a slash command from the first message.

2. Open the folder in your LLM agent and say "Run the onboarding." (No account setup needed. If you've connected your accounts via Setup/, it also imports conversion actions, campaigns + spend so you type less.)

3. It picks a scope (paid-advertising by default, or broader CMO/all-channel), fetches what the accounts and your website already know, then runs a Core interview (~20-25 questions, many pre-filled). ~20-25 min → populated CLIENT.md, all 00-context/ files, goals, budget, roadmap.

4. Go deeper any time with "deep-dive on <domain>"; after any working session say "Update context" — the one ritual to learn (documented in the client's AGENTS.md §3).

Tiered capture: Core now, Deep-dive later

Every file has a Core ([C]) layer filled at onboarding and a Deep-dive ([D]) layer marked ⬜ deep-dive pending that fills over time — so the brain gets rich without a 90-minute setup.

The facts that make automations smarter

• 00-context/01-business.md — unit economics (margin, target CAC, profitability requirement) so decisions optimize profit, not just revenue.
• 00-context/05-measurement.md — every conversion action catalogued + the tracking model (revenue vs margin, primary action, attribution).
• 01-goals-budget/targets.md — the real goal per campaign / product line / funnel stage (not just the platform's optimization setting).
• 00-context/12-naming-conventions.md — the naming schema + a machine dictionary so the agent can decode any campaign name and point automations at the right place.

These turn the brain into something an LLM can act on without re-scanning the live account every time.

Connection to the Agency Intelligence OS

The Agency OS keeps one client brain per client under clients/<name>/, all using this exact schema — it embeds a copy of this template/ at clients/_TEMPLATE. The two ship as separate, self-contained downloads and are kept identical by a small sync script in the Agency OS (sync-template.sh), so the schema never drifts (see Agency-Intelligence-OS/clients/README.md).

Keep it detailed but not complicated. Five folders, one ritual, two smart context files. That's the whole system.

Copyright (c) 2026 Tomas Kliment & Denis Capko / adsrepo.com. All rights reserved. See LICENSE for terms — https://www.adsrepo.com/license