DEV Community

Matt Cretzman
Matt Cretzman

Posted on • Originally published at blog.mattcretzman.com

Master Library + Client Forks: A Deployment Pattern for Multi-Client AI Skills

A fractional CMO walks into Monday morning with five clients on her roster. Each one sells something different. Each one targets a different ICP. Each one has a different competitor set, a different industry, a different incumbent vendor problem. And every Monday she has to brief herself before five different sales conversations.

The math gets ugly fast. Most fractional consultants run three to five clients at a time. Average monthly retainers for fractional sales leaders climbed to $9,651 in 2024. 72% of CEOs plan to increase their use of fractional executives in the next twelve months.

Which means every fractional operator has the same operational problem: how do you build leverage across multiple clients without flattening them all into the same generic playbook?

You don't want a separate, hand-built workflow for each client. That doesn't scale. You also don't want one generic workflow that ignores everything that makes each client different. That doesn't deliver.

The answer is a deployment pattern most operators haven't named yet. Master Library + Client Forks.

The pattern

Inside Skill Refinery, every account can have multiple libraries. A library is a folder. Each library can hold any number of skills — structured pieces of expert IP you invoke from ChatGPT, Claude, or Copilot via MCP.

Two layers:

Layer 1 — Master Library. One library called Master Skills (Blueprints). Holds the canonical, generalized version of every skill. Industry-agnostic, ICP-agnostic, offer-agnostic. The blueprint.

Layer 2 — Client Libraries. One library per client, named after the client. Contents start identical to the Master — a clone — then get personalized. Same skill structure, but tuned to the client's industry, ICP, regulatory framework, offer language, named accounts and competitors.

The Master is the source of truth. The Client Library is the working copy.

If you've worked in software, this is the same shape as a base class with subclasses, or a config file with environment overrides. The methodology lives in one place. Personalization lives in another. Inheritance is implicit but disciplined.

Why this beats the alternatives

Two wrong-shape patterns most operators stumble into:

One big library. Everything goes in one library. Skills accumulate conditional logic — if client is in healthcare, do X; if financial services, do Y. Within six months they're bloated, fragile, impossible to reason about. Every change risks breaking three other clients' workflows.

Skill per client. Every skill gets cloned once per client and lives in isolation. No master. When you learn something new, you have to update five skills manually. You won't. Drift sets in. Quality varies wildly across clients.

Master Library + Client Forks fixes both. The master is the single place where methodology lives. Forks are the only place where personalization lives. They never get confused.

Implementation — three steps

1. Create the Master Library
   Name: "Master Skills (Blueprints)"
   Access: internal
   Audience: staff
   Description: "Canonical generalized skills. Forked into client 
                 libraries for personalization."

2. Drop generalized skills into the master
   - Strip client-specific naming
   - Strip tool dependencies that aren't universal
   - Generic enough for any operator to pick up Monday morning

3. Create one library per client. Clone master skills into each.
   Library name = client name
   Personalize:
     - Industry context
     - ICP definition
     - Named competitors / incumbents
     - Offer language
     - Tone-of-voice rules
Enter fullscreen mode Exit fullscreen mode

Onboarding a new client onto every skill in the master takes about thirty minutes. Linear scaling with client count.

The leverage moment

The reason the pattern matters is what happens at month three.

You'll learn something about how to run a sales intelligence briefing that you didn't know on day one. Maybe LinkedIn job descriptions are the highest-signal incumbent vendor check ever invented. Maybe Form 4 SEC filings are the most underused data point in B2B prep.

When that learning hits, you don't update five client skills.

You update the master. Once.

Then you propagate the relevant change to client libraries that need it — and only the ones that need it. Some changes are universal. Some only matter for clients in specific industries.

Compound leverage. Every engagement makes the master better. Every master improvement makes every future engagement better. The skills don't degrade. They get sharper over time.

The IP shift

The pattern works because of something most consultants get wrong about IP.

Your IP isn't a single document or a single methodology. It's the set of repeatable, generalized patterns plus the specific knowledge of how to deploy them in a given context.

Generalized methodology without context is a textbook.

Context without generalized methodology is tribal knowledge that walks out the door when you do.

A master library captures the methodology. A client library captures the context. Together they're an actual operating system for fractional work.

You're not selling hours anymore. You're selling deployments of your skill stack.

What to build first

Pick one skill. The one you do most often. The one you'd hate to lose if you had a memory wipe.

For most fractionals that's pre-meeting research, weekly client check-in synthesis, or outbound sequence drafting.

Generalize it into a master skill. Fork it into a single client library for one engagement. Run it for two weeks. See whether the master gets sharper. See whether the fork stays personalized. See whether the second client onboarding takes thirty minutes instead of three hours.

The pattern compounds quietly until it doesn't. By client four or five it's the only way you can imagine running the practice.

The tools are off the shelf. Skill Refinery, an AI like Claude or ChatGPT, the libraries you set up yourself in the SR admin UI. No engineering required.

Keep Building,
Matt


Originally published at blog.mattcretzman.com.

About the author: Matt Cretzman builds AI agent systems through Stormbreaker Digital. Ventures include Skill Refinery. Writing at mattcretzman.com.

Top comments (0)