How to build skills for AI — reusable, portable instruction sets that produce consistent results every time, without you in the room.
Where we've been — and why this week changes everything.
You have the tools. Now it's time to turn them into systems.
This week we go hands-on. You'll learn how to actually build skills — the structured, reusable instruction sets that turn one-off prompts into systems other people can run. The difference between "I use AI" and "I built something with AI" is a skill.
"A prompt is something you type. A skill is something you ship. This week, you learn to ship."
The fundamental distinction that separates casual AI use from real systems.
A prompt is a one-off instruction. You type it, get a response, and it's done. A skill is a reusable operating instruction: it includes the context, constraints, prerequisites, and output format so the AI produces consistent results every time — whether you run it or someone else does.
Write me a marketing email for our new product launch.
You are a B2B SaaS copywriter. The product is [X], targeting [Y]. Use the AIDA framework. Keep it under 200 words. Include one CTA. Tone: professional but warm. Output: subject line + body.
A skill produces the same quality output on run 1 and run 100. A prompt depends on context and luck.
A skill can be handed to a colleague, a scheduled agent, or an n8n workflow. A prompt lives in your head.
Skills compose into systems. Prompts are isolated events that don't build on each other.
When a skill produces bad output, you can inspect which layer failed. With a prompt, you just try again.
Not everything needs to be a full skill. Think of it as a spectrum: on one end, a raw question you type into chat. On the other, a fully specified instruction set with role, context, prerequisites, constraints, and output format. Most people live at the prompt end. The goal is to move your repeating tasks toward the skill end.
"If you find yourself typing the same kind of request more than three times, it's time to turn it into a skill."
A five-layer framework for building skills that work every time.
Every reliable skill follows the same five-layer structure. This isn't theory — it's the pattern behind every skill that actually ships. Master these five layers and you can turn any repeating task into a system.
# [Skill Name] ## Role You are a [specific role with expertise level]. You [key behavior or perspective]. ## Context - [Company/team/project background] - [Industry or domain context] - [Any standards or references to follow] ## Prerequisites Before producing any output, ask the user for: 1. [Required input #1] 2. [Required input #2] 3. [Required input #3] If any prerequisite is unclear, ask for clarification. Do not guess. ## Constraints - [Word count / length limit] - [Tone and voice] - [Things to avoid] - [Format restrictions] ## Output Format Produce the result as: - [Section 1: description] - [Section 2: description] - [Section 3: description] Use [formatting style]. Do not include [excluded elements].
# Meeting Prep Brief ## Role You are an executive assistant who prepares concise, actionable meeting briefs. ## Context - Company: [Your company name] - Meeting briefs should enable the attendee to walk in prepared in under 2 minutes of reading. - Prioritize decisions needed and open questions over background summary. ## Prerequisites Before producing the brief, ask for: 1. Who is the meeting with? (names and roles) 2. What is the meeting about? (agenda or topic) 3. What is our goal? (decide, inform, explore, or close) 4. Any relevant history? (past meetings, open items, blockers) ## Constraints - Maximum 250 words - No filler or pleasantries - Use bullet points, not paragraphs - Flag any risks or sensitivities ## Output Format 1. **Meeting**: [who + when] 2. **Our Goal**: [one sentence] 3. **Key Context**: [3-5 bullets] 4. **Decisions Needed**: [numbered list] 5. **Open Questions**: [numbered list] 6. **Watch Out For**: [risks or sensitivities, if any]
"A skill that doesn't need you in the room is a skill that's finished. If someone else can't run it and get the same quality output, keep refining."
Three ways to run your skills — and how to pick the right one for each task.
A skill is interface-agnostic — it works whether you run it in chat, through an integration, or via code. The question isn't which interface is "best." It's which interface matches the frequency, complexity, and autonomy level of the task.
Exploration, brainstorming, one-off runs, testing new skills. You're in the loop for every step.
Recurring tasks, scheduled workflows, MCP connections, Dispatch. Runs without you once configured.
Complex logic, custom data transforms, high-volume processing, API integrations. Maximum control.
I need to analyze customer feedback, so let me build a Python script with the Anthropic API, set up a database, and deploy it to AWS.
I need to analyze customer feedback, so let me write a skill in chat, test it on 10 examples, then schedule it in Dispatch to run weekly on my inbox.
Most people jump straight to code because it feels more "real." Don't. The sweet spot for 90% of business tasks is chat to prove the concept, then integrations to automate it. Code is the exception, not the rule.
"The best interface is the simplest one that does the job. If you're writing code to do something Dispatch can handle, you've over-engineered it."
A step-by-step walkthrough of turning your most common tasks into production-ready skills.
Theory is done. Let's build. The process is always the same: identify a repeating task, extract the five layers, test until consistent, then deploy to the right interface.
Start with the task you do most often with AI. Not the most complex — the most frequent. Frequency is what makes a skill valuable. A skill you use daily returns 10x more value than a brilliant skill you use once a month.
Client responses, internal updates, follow-ups. High frequency, high variation.
Meeting notes, report summaries, research briefs. Structured output from unstructured input.
Social posts, blog outlines, internal docs. Repeating format with variable content.
Code reviews, proposal feedback, competitive analysis. Structured critique with criteria.
For your chosen task, answer these questions — they map directly to the skill template:
I do this task regularly with AI: [Describe the task in detail — what you ask for, what good output looks like, what mistakes you've had to correct] Help me extract the five layers of a reusable skill: 1. Role — Who should the AI be? What expertise level and perspective? 2. Context — What background does the AI need every time? What doesn't change between runs? 3. Prerequisites — What information must I provide (or the AI must ask for) before it starts? 4. Constraints — What rules apply? Length, tone, format, things to avoid? 5. Output Format — What should the result look like? Headers, sections, structure? Write the complete skill using the standard template format. Include [placeholders] for the parts that change each time.
Run the skill five times with different inputs. Each time, evaluate: did the output meet your standards without correction? If not, identify which layer was weak and tighten it. Common failure modes:
Once the skill produces consistent results in chat, move it to the right interface. Save it as a Project instruction for daily use, add it to Dispatch for on-demand automation, or schedule it for recurring tasks.
"The first skill takes 20 minutes to build. The second takes 10. By the fifth, you're writing them in 5 minutes because you've internalized the pattern. That's the point."
"Week 1 you learned to talk to AI. Week 2 you learned to use it. Week 3 you connected it to everything. Week 4 you learned the doctrine. Week 5 you built the systems. That's the complete toolkit."
Real questions from the live office hours — practical answers you can use today.
Convert your top tasks into reusable skills, test them for consistency, and start your personal skill pack. Everything you need is in the homework above.