The Agency OS Orchestrator

Back to all skills

What it does

Sequences the eight Agency OS builds and hands each one to the skill that owns it. Build 1 (Brand DNA) runs once per workspace. Builds 2 to 8 (ICA, Offer, Funnel, Content, Ads, Sales, Delivery) run once per product, and a workspace can hold several products running at different stages. After every build it writes the build status, regenerates agency-dashboard.html, opens it, offers you a slide deck, and tells you what is next. It writes no deliverables itself: it routes, gates, and keeps state.

Say this to start

This skill has no button. You start it by saying what you want. Any of these will do it:

When to reach for it

• Brand new workspace, no idea where to start
• You want the builds run in the right order without deciding each time
• Setting up a new client workspace from scratch
• Adding a second or third product to an existing brand
• Coming back after a break and needing to know what is done
• You asked for a build but its prerequisite is missing

When NOT to use it

If you actually want	Use this instead
You already know which single build you want to run	the build skill directly (brand-dna, ideal-customer-avatar, offer-architect, funnel-architect, content-system-architect, meta-lead-system, sales-conversion-engine, client-delivery-engine)
Winning clients for your own agency through outbound	agency-sales-system
Pressure-testing a deliverable that is already written	critique
Keeping or expanding the clients you already have	client-retention or referrals

Before you start

What you need	Why
Nothing for the first run	it creates the brand/ folder, the _default workspace and your CLAUDE.md config silently on first load	Required
Build 1 (Brand DNA) complete or lite	the prerequisite gate reroutes every product build back to Brand DNA until it exists	Required
A GoHighLevel account	GHL is the default deployment platform for funnels, pipelines and automations. Without it you still get every document, but nothing deploys	Optional
Web research access (Firecrawl, Perplexity, or plain web search)	Builds 2 to 7 run a blocking market research pass first. Without a tool it writes an error stub and continues	Optional
For a client workspace: their name, industry, website and social handles	the new-client flow asks for all four before it creates the workspace	Optional

How it runs

1. Silent setupChecks for CLAUDE.md and creates or updates the Agency OS section without telling you. Creates brand/ if missing. On a brand new install it asks once whether you want English or Deutsch.
2. Workspace selectionA fresh install goes straight to _default. Once you have completed builds it asks whether this session is for your agency or a client. Choosing client asks for the name, industry, website and social handles, then creates brand/client--[name]/.
3. Welcome and next actionReads the .status.yml files and shows where each product stands, then suggests exactly one next action. It is not a menu. Next up is the product furthest behind in the sequence.
4. Prerequisite gateBefore any build it checks that build's prerequisite for that specific product. If Build 4 is missing and you ask for Build 6, it offers to run the missing build first and resumes yours afterwards. The gate cannot be skipped.
5. Market research passFor Builds 2 to 7 it researches competitor signal, industry state and platform trends before the build starts. It asks whether you want to name your top three competitors, skip naming them, or skip the research. Results append to market-intel.md.
6. Run the buildHands off to the owning skill with every previous schema output passed as context, so you never re-explain your business. Builds 1 and 8 skip the research step.
7. CloseoutAuto-scores the output with critique (Build 1 gets a completeness check instead), writes .status.yml, regenerates and opens the dashboard, then asks whether you want the slides. That question blocks: it waits for your answer.
8. HandoffIf your profile is lite, it asks one or two profile questions to fill a gap. If the product is still generalist it shows one line suggesting /set-niche. Then it tells you the next build and stops.

What you get

• brand/[workspace]/agency-dashboard.html, regenerated and opened after every build
• brand/[workspace]/build-overview.md plus .docx, a snapshot of progress
• brand/[workspace]/.status.yml and a per-product .status.yml recording what is complete and what is next
• agency-deck.pptx, cumulative, only if you say yes at the opt-in (one deck per product in multi-product mode)
• An Agency OS section written into your CLAUDE.md
• Every build artifact from the skills it routes to, saved as .md and .docx where applicable

Honest limits

Where people go wrong

The mistake	Do this instead
Treating the welcome screen as a menu and picking any build you like	Accept the suggested next action or name a specific product and build. The suggestion is the one that keeps the sequence intact.
Running Build 1 again for every new product	Brand DNA is brand-scope. It runs once per workspace. Products start at Build 2.
Typing folder-style slugs when naming a product	Type the product name in plain English. It converts the name to a folder name itself and shows you the name, never the slug.
Assuming a new client workspace means delivery work	New client means a new workspace to build for. Delivery (Build 8) only opens once that product has its Offer, or when you run /deliver-client explicitly.
Skipping Brand DNA because the interview looks long	Run /first-win instead. The lite profile passes the gate and fills in build by build.