To build the best possible skill from this material, I need a few things:

1. **What’s the source?** (Book title, author, or describe the framework)

2. **Who will use the generated skill?** (Your clients? Students? You personally?)

3. **What’s the transformation?** (What should someone be able to DO after going through this framework?)

4. **How deep?** Do you want:

- Quick diagnostic only (assess where someone is)

- Full guided journey (diagnose + prescribe all steps)

- Coaching tool (diagnose + prescribe + track progress)

5. **Any domain constraints?** (Industry, audience level, terminology preferences)

STARTING STATE → [Stage 1] → [Stage 2] → ... → [Stage N] → DESIRED OUTCOME

↑ ↑

“Where most Each stage has: “The promise

people begin” - Entry criteria of the framework”

- Key actions

- Common blockers

- Exit criteria (how you know you’re ready to advance)

- Estimated effort/time

[skill-name]/

├── [SKILL.md](http://skill.md/) # Main skill file (< 500 lines)

├── references/

│ ├── [framework-detail.md](http://framework-detail.md/) # Full framework with all stages, actions, criteria

│ ├── [diagnostic-engine.md](http://diagnostic-engine.md/) # Complete diagnostic question bank and scoring logic

│ └── [prescription-library.md](http://prescription-library.md/) # Action prescriptions organized by stage + blocker

├── assets/

│ └── (any templates needed)

└── companion/

├── [system-prompt.md](http://system-prompt.md/) # Standalone prompt for ongoing use without the skill

└── framework-summary.docx # Professional Word doc summarizing the framework

# [Framework Name] — AI Coaching Prompt

## Your Role

You are an AI coach trained in [Author]’s [Framework Name] methodology...

## The Framework

[Condensed framework with stages, criteria, and key principles]

## How to Interact

[Diagnostic flow + prescription logic in prompt form]

## Key Rules

[Author’s core principles that must be maintained]

{

“skill_name”: “[generated-skill-name]”,

“evals”: [

{

“id”: 1,

“prompt”: “[Simulated user at Stage 1 describing their situation]”,

“expected_output”: “Correctly identifies Stage 1, pinpoints likely blocker, prescribes stage-appropriate actions”,

“assertions”: [

“Identifies the user’s current stage correctly”,

“Names a specific blocker from the framework”,

“Provides exactly 3 actionable next steps”,

“Uses the author’s terminology”

]

},

{

“id”: 2,

“prompt”: “[Simulated user stuck between stages, unclear position]”,

“expected_output”: “Asks clarifying diagnostic questions to narrow stage, then prescribes”,

“assertions”: [

“Does not skip diagnostic when stage is ambiguous”,

“Asks targeted questions that differentiate between adjacent stages”,

“Eventually provides clear stage identification and prescription”

]

},

{

“id”: 3,

“prompt”: “[Simulated user at final stage, near completion]”,

“expected_output”: “Recognizes advanced position, provides completion/maintenance guidance”,

“assertions”: [

“Does not prescribe beginner actions to advanced user”,

“Acknowledges progress appropriately”,

“Provides next-level or maintenance guidance”

]

}

]

}

Full video transcript

Hey everybody, Darby here with GenAI University. We are going to be kicking things off today inside of our Claude Mastery Intensive series here on February 17th, 2026, going to be digging into how to use Claude to identify and build skills around frameworks that you can use as deliverables for yourself, your team, your clients, anyone that you’re working for — using Claude to help with using your frameworks, your own methodologies, and translating those into actionable deliverables.

I did make a post about today’s workshop to come to the table with a framework or with a methodology of some sort that we can use and put into action from this skill, which I know many of you here already have a lot of your skills, your methodologies built out in some way, shape, or form. So it’s probably just a matter of gathering some documents and some resources you’ve already created. But did anyone actually bring those here and is ready to implement in real time? If not, it’s cool too — I have a demonstration that we’re going to be walking through.

Some of what I’m going to show you guys today, we’re going to dive right into it so that we can see the flow of how to work with Claude. Much of this process is adaptable to whatever skill or framework or method that you would use to walk somebody through on a consistent basis to produce some sort of deliverable, some sort of report, some sort of audit that you do from a client discovery call. Maybe you interview somebody and then you produce a two or three page executive brief that walks them through a strategy that you’ll then guide them through implementing.

One example: an agency I work with, a PR and branding agency, developed a skill where they interview a client that’s onboarding with them, and then they deliver a four-page report that walks them through their positioning and how they’re going to help them implement, over the next 90 days, a roadmap toward getting more press for their brand. They work with a lot of authors and a lot of people in that regard. It’s the same process every time — they do the interview, transcribe it using Fathom, run that transcript through a skill, and it produces a four to six page report that they walk through with their clients and implement. We’ve done about 150 of those reports over the last two years.

One very specific part of their process — and what I’m going to walk y’all through today — is taking a framework or method that you already have. If you already have a book that you’ve written on a process, you probably already have your frameworks organized inside of the book. If you’re thinking about writing a book, you may have a framework you include inside it, or a skill you’ll use to accompany the book to help people implementing the concepts. What I’ve noticed with a lot of authors who create books with proven processes is a lot of them don’t have an AI to accompany their process — they give people the “how” and the framework, but people are left executing it fully themselves without AI tools to help support them inside that process.

So I’m going to show you how to do that with a book we’ve been using as a reference for an upcoming book intensive, and then we’ll have time for open floor questions and reviewing. I want to demonstrate one approach for taking an existing framework that’s been proven time and time again, extracting a specific deliverable, and then turning that into a skill you can use again and rinse and repeat for a ton of your other stuff.

A quick question from John: “I use Plaud to record many in-person sessions. Can I use the Fathom skill?” Not entirely sure what “use the Fathom skill” means exactly, but if recordings from Plaud are being sent over to a skill, that could work — happy to dig in with more context. Bruce is building a daily news and weekly deep dive academic professional research journal feed to help create valuable content for an audience — that can definitely become a skill too.

Without further ado, let’s dive in. So this is a book published by Chandler Bolt. We’re using this as a roadmap for many people going through our book-writing intensive. Chandler has been helping people for years write books — he’s helped hundreds, thousands of people go from blank page to published author. Inside this book, Chandler covers a lot of content. We’re not going to cover everything from the book. What I want to do is take a specific assessment that Chandler talks about — the book launch readiness assessment — and create a skill around it.

To do this: first, I’m going to download the book. Second, I’m going into my projects inside Claude and creating a new project — I’ll call it “published book skills.” If you’re starting this on your end, call it whatever methodology you want to build a skill around. Next, I’ll upload the book into the Claude chat thread and also upload it as a file into the knowledge base. Then I give Claude a conversational command: “Claude, I’m uploading a book for you from the author who has developed a framework for helping people go from blank page to published. I want to identify the book launch readiness assessment covered in this book and identify a way that I can turn this into a skill I can use for people going through the process of writing a book who are stuck at a specific point and need help going through this skill based on where they’re at right now to where they want to go with their book.”

That’s my starter conversational prompt — give Claude the framework with context, then tell it to analyze the framework and focus on the book launch readiness assessment we want to build a skill for. Claude is reading, assessing, identifying, and extracting the text and contents of the book. I’m using Opus 4.6 with extended thinking. Sometimes this can take a few minutes to run — I’m okay with that because of the quality of the outputs. You’ll see it compacting the conversation because the book is a large amount of content, which you might encounter too. You may find limitations with different Anthropic models — right now Opus 4.6 is definitely the best model you could use for this kind of work.

It’s ingesting all the context for the methods. Anyone here with their own methodology can build a project that does the same thing. The key is having the frameworks, the examples if you have them, anything relevant to your method and what you help people produce regularly, inside a project dedicated specifically to that.

To Hannah’s question about not having a methodology yet — you could use deep research to help develop one. Start with the big idea: “This is what I want to help people produce, help me create a method that gets people to this result,” as a starting point with Claude, and it will work through and create an instruction manual you can refine, add context to, and modify throughout the conversation. You probably won’t one-shot the perfect skill right away, especially when iterating on a method you know in your head but need help refining. Once you get to a point where it’s solid and repeatable, turn it into a skill, then start testing it inside your own process — that feedback loop of testing and refining is where it gets better over time.

Yolanda asked about best practices for creating skills, noting Darby created several skills from one book. Yes — that was a series of about 12 total skills created for people going through the book-writing intensive, from a deep analysis of the book that took several hours. Before jumping in and creating a ton of skills around everything possible, focus on the number one thing taking time right now that you’d like to streamline as a deliverable. The book launch readiness assessment was chosen because it doesn’t make sense to go through writing a book without first knowing if you’re ready to write one, or what to write about. Think about the order of operations your client needs to go through to get to the next step working with you, and where people run into friction or get set up for failure if they skip a step — focus skills on getting people started there.

Claude ran into a limitation partway through processing the book — a timeout, likely related to the size of the document — so it was rerun in a fresh chat. This is part of working with these tools: sometimes AI is “having a morning,” and the response is to adapt, rerun, or take a different approach.

Hannah mentioned a sales research prospect skill (built from the market sauce blueprint materials) that helped her generate an amazing proposal from a discovery call transcript. On the question of storing skills in GitHub versus directly in a Claude account: skills currently live directly inside Claude for this workflow, though it’s possible to add each skill as its own individual markdown file for reference in GitHub. On a related concern about many skills adding token overhead — uploading multiple skills does mean Claude has more to check through when a project or account holds several, though this wasn’t something extensively tested live.

To Coach John’s question about which plan — currently on the Max $100/month plan, having run into limitations on the $20/month plan; the $200/month tier offers roughly 20x more capacity than the $100 tier.

Recapping: the published PDF manuscript for Chandler Bolt’s “Published: The Proven Path from Blank Page to Bestselling Author” was given as context, with instructions about the goal. Claude identified the book launch readiness assessment as a strong first skill candidate, then, on a follow-up command, produced two deliverables: a fully populated sample assessment (using a test subject, “Jessica,” stuck mid-writing with a perfectionism blocker) and the actual skill markdown file — the Claude project instructions and the diagnostic engine behind the assessment. That skill file is what gets turned into a Claude project skill that anyone can use with their own book or methodology, and the same approach works for any book, which is why this example (rather than Darby’s own book) was used for the demo.

Attendees were invited to share a quick blurb about their own book and where they’re stuck, so the skill could be run live as a test case using their real situation.

Jumping back into the produced output: a “published pathway assessment,” current position diagnostic, ran as a live demonstration using the skill, producing a report someone stuck in the writing process could receive as a deliverable. Alongside the assessment example, the actual downloadable skill file was also produced, titled “Published Pathway Claude Skill” — triggered when a user says something like “run published pathway book assessment, where am I with my book?” It walks through Chandler Bolt’s framework phases and milestones, asking diagnostic questions when triggered.

Uploading the skill hit a snag: Claude’s skill upload requires proper YAML front matter (name and description) at the top of the markdown file, which wasn’t included automatically. This is a known issue — simply asking Claude directly, in the same chat, to reformat the skill so it’s properly structured for upload resolves it. After the fix, the skill uploaded successfully under Capabilities > Skills in account settings.

With the skill uploaded, the next step is setting up a dedicated Claude project containing both the skill instructions and the source framework, so the skill can be run repeatedly. Renaming the project to “published pathway assessment,” the workflow becomes: someone says they have a book and are stuck, they fill out an intake form or send a paragraph describing where they’re at, and that context gets pasted into the project to run the assessment — for example, “execute the published pathway assessment for Shelley, context = [notes].”

Running the assessment live for a participant (Shelley, stuck trying to narrow 120 case studies down to a workable number for her book on succeeding in small business), the skill produced an eight-page report: a diagnostic identifying the blocker as “the abundance trap” (no curation system, so every case feels important and none get cut), a four-criteria scoring rubric to score and select the top 10 cases per chapter (with the other 60 becoming pre-launch content like blog posts and LinkedIn stories), a 99-cent launch strategy from the book, and a full 90-day roadmap broken into phases, all grounded in specifics from Chandler’s book and tailored to Shelley’s stated blocker. The report includes concrete next steps, such as building the case-scoring rubric as a spreadsheet and having Claude execute the cutting/curation work directly rather than doing it manually.

The broader point: once a repeatable assessment or report skill like this exists, it can be run for many clients by feeding each person’s context into the same project — useful for anyone gathering client information (e.g., via intake forms) and needing to produce a consistent diagnostic or report at scale. It’s recommended to also keep copies of built skills in a Google Drive folder rather than relying solely on storage inside Claude.

To close the loop, the workshop also demonstrated using Claude’s own skill-builder capability to create a general-purpose “skill factory” skill — one that takes any expert framework or book, extracts it, and outputs a complete installable skill that diagnoses where someone is in the framework and prescribes next steps. Going through Claude’s skill creator (Capabilities > Skills > Create with Claude), a prompt describing “a skill for expert frameworks” was used, with Claude asking clarifying questions (desired output format — skill markdown plus reference files, whether to include interactive intake) before generating the skill. This “skill factory” is meant to be reusable any time a new book or framework needs to be mined for a deployable diagnostic skill.

On model usage and cost: a participant on the $20/month plan reported running out of usage credits after less than an hour into the workshop while attempting to follow along with Opus 4.6. Extensive outputs involving heavy reasoning consume significantly more usage on Opus 4.6 than on other models. The recommendation was to try Sonnet 4.5, Opus 4.5, or Haiku 4.5 for lighter runs of the same process, since Opus 4.6 burns through usage limits substantially faster.

On what files to include in a project: generally anything related to the skill, method, or framework itself. For the Chandler Bolt book assessment, that meant the book plus the instructions created from it. For another example (buyer’s brief recipes), it meant the source book for full framework context plus the exact instructions for the output, then running the skill against that.

Closing out, the workshop wrapped with a reminder that the next session will be at the same time next week, and thanks to everyone who joined.