You understand what skills are (Lesson 08). You've built your first one (Lesson 09). Theory is behind you. Now you build.
These exercises take you from dissecting existing skills to composing production-ready skill suites for real-world scenarios. Each module adds a layer of sophistication: first you read skills, then you write them, then you add examples and references, then you test and iterate, then you chain them into workflows, and finally you build complete skill libraries for business, education, or personal productivity.
The Big Idea behind all of this: a skill is a reusable instruction file that teaches an AI agent how to handle a specific type of task consistently. Building skills is the fundamental capability that separates "using AI" from "building with AI."
Download Exercise Files
Download Skills Exercises (ZIP)
After downloading, unzip the file. Each exercise has its own folder with an INSTRUCTIONS.md and any starter files you need.
If the download link doesn't work, visit the repository releases page directly.
Every exercise follows the same workflow:
You can work through exercises at your own pace. Each module builds on the previous one, so work in order within a module. You can skip ahead between modules if a particular topic interests you, but the full sequence gives you the strongest foundation.
Each exercise is marked with the tools it works best with:
Most exercises work with either tool. Where one is clearly better, the exercise notes will say so.
Use this framework for every skill you build across these exercises:
This framework applies whether you're building a simple email skill or a complex multi-step pipeline. Return to it whenever you get stuck.
Use this rubric to evaluate every skill you build. Score yourself honestly after each exercise:
A score of 18+ (out of 24) means you have a production-ready skill. Below 12 means you need another iteration round. Track your scores across exercises to see your growth.
Before you build skills, you need to read them fluently. These three exercises develop your ability to analyze existing skills, judge when a skill is the right tool, and understand what skills add beyond raw prompts.
1.1Anatomy of a Skill
The Problem: You have three sample skills of increasing complexity: a simple greeting formatter, a medium weekly report generator, and a complex content reviewer with examples and references. Your job is to take them apart and understand how each piece works.
Your Task: Open each SKILL.md and identify the five core components: frontmatter, trigger conditions, step-by-step instructions, output format, and rules/constraints. Write a 3-sentence summary for each skill. Then predict what Claude will produce for provided test prompts before running them. Finally, find the intentional weakness in each skill (language handling, empty sections, non-text content).
What You'll Learn:
Starter Prompt:
"Read the three sample skills in sample-skills/ and walk me through the anatomy of each one. Help me identify the frontmatter, triggers, instructions, output format, and constraints."
Reflection Questions:
The Problem: Not every task needs a skill. Building unnecessary skills wastes time; skipping useful ones means repeating yourself. You need to develop judgment about which tasks are "skill-worthy."
Your Task: Classify 15 provided scenarios as BUILD A SKILL, JUST PROMPT, or MAYBE using the four-criteria framework: Is it repeatable? Does it need consistent output? Can you write clear rules? Does it require background context? For every scenario you mark as skill-worthy, write a one-sentence trigger description. Then identify 3 tasks from your own work that deserve skills.
What You'll Learn:
Starter Prompt:
"Read scenarios.md and help me classify each scenario. For each one, let's decide together: build a skill, just prompt, or maybe? Use the four criteria from the decision framework."
Reflection Questions:
The Problem: You intuitively sense that skills produce better output than raw prompts, but you haven't measured the difference. This exercise makes the difference concrete and measurable.
Your Task: Generate a weekly status report from provided data twice: once with your best raw prompt (no skill), and once with the provided weekly-report skill. Score both outputs on a 7-criteria rubric (format, metrics, blocked items, conciseness, action language, tone, send-worthiness). Then improve the skill based on what you learned and run a third round.
What You'll Learn:
Starter Prompt:
"I'm going to generate a weekly report two ways. First, let me write my own prompt. Then we'll use the skill. Help me compare the outputs on the rubric in comparison.md."
Reflection Questions:
Now you build. These three exercises each produce a complete, working skill for a common task. The focus is on writing clear instructions, good trigger descriptions, and testing against real inputs.
The Problem: Every time you ask Claude to write an email, it sounds like Claude. You want emails that match YOUR voice: your greeting style, your formality level, your sentence patterns, your sign-off.
Your Task: Analyze 5 sample emails (provided) to identify your writing patterns. Then create a my-email-style/SKILL.md that captures those patterns as concrete rules. Test it against provided prompts, compare output to your real style, and iterate at least 2 rounds until Claude writes emails that sound like you.
What You'll Learn:
Starter Prompt:
"Analyze the 5 emails in my-emails/ and identify my writing patterns. What's my greeting style, formality level, sentence length, and sign-off? We'll turn these into a SKILL.md."
Reflection Questions:
The Problem: You keep asking Claude to organize files, and every time you re-explain your preferences from scratch: folder structure, naming conventions, handling duplicates, creating changelogs.
Your Task: Define your file organization rules, then encode them into a file-organizer/SKILL.md. Test against provided sample files first, then against edge cases designed to break simple rules: files with no extension, very long names, special characters, empty files, deeply nested duplicates. Improve and re-test.
What You'll Learn:
Starter Prompt:
"Help me define my file organization rules. I'll answer your questions about folder structure, naming conventions, and edge cases. Then we'll write it as a SKILL.md and test it."
Reflection Questions:
The Problem: You regularly receive CSV files with inconsistent formatting: mixed date formats, random capitalization, duplicate rows, missing values. You want Claude to clean them the same way every time.
Your Task: Study two provided messy CSV files and catalog every data quality issue. Write specific cleaning rules (not "fix the dates" but "convert all dates to YYYY-MM-DD; if ambiguous, assume MM/DD/YYYY"). Build data-cleaner/SKILL.md with rules for dates, names, phone numbers, email addresses, and deduplication. Require a preview before changes and a change log after. Test on both files, then test on a surprise third file you haven't seen.
What You'll Learn:
Starter Prompt:
"Open messy-data/customers.csv and messy-data/transactions.csv. Help me catalog every data quality issue. We'll write specific cleaning rules for each issue type."
Reflection Questions:
Rules tell Claude what to do. Examples show Claude what "good" looks like. These exercises teach you to use examples, templates, and tone samples to dramatically improve skill output consistency.
The Problem: You write monthly stakeholder reports. The format, tone, and structure should be identical every month, but Claude gives you slightly different formatting, section ordering, and detail levels each time.
Your Task: Run the same report data through two versions of a skill: v1 (rules only, no examples) and v2 (same rules plus an example of a perfect report). Compare the outputs. Then build your own v3 with two examples plus anti-examples showing what NOT to produce. Test on new data.
What You'll Learn:
Starter Prompt:
"Run report-data/february-data/ through skill-v1/SKILL.md and save the output. Then run the same data through skill-v2/SKILL.md. Let's compare them side by side."
Reflection Questions:
The Problem: Raw meeting notes need to become polished minutes. The output should follow an exact template, extract action items consistently, and exclude off-topic chatter. You need a skill with a separate template file that's easy to update.
Your Task: Build a complete skill folder: SKILL.md (instructions referencing the template), templates/minutes-template.md (the output format), and examples/ (an input/output pair). Your SKILL.md tells Claude to read the template, study the example, then transform raw notes following specific rules for inclusion, exclusion, and action item extraction. Test on three provided meetings. Then extend the skill to produce two outputs: formal minutes and a quick Slack summary.
What You'll Learn:
Starter Prompt:
"Let's build a meeting minutes skill with a separate template file. Read the raw meetings in raw-meetings/ and the starter template in starter-template.md. Help me create the full skill folder structure."
Reflection Questions:
The Problem: Writing student or employee feedback is hard because tone matters enormously. "Be warm but honest" means different things to different people. You need a skill that captures YOUR specific version of warm-but-honest.
Your Task: Study four tone examples (harsh, soft, corporate, ideal) and build a skill that references the ideal example as the target tone and the others as anti-patterns. Define feedback structure (strengths, areas for improvement, specific examples, encouragement) with rules requiring specificity and personalization. Test on five student profiles. The ultimate test: can a classmate tell which feedback was skill-generated vs. hand-written?
What You'll Learn:
Starter Prompt:
"Read the four tone examples in tone-examples/. Let's analyze what makes ideal.md work and what's wrong with the other three. Then we'll build a feedback skill that nails the right tone every time."
Reflection Questions:
Real-world skills often need to consult external documents: brand guides, policy documents, teaching standards. These exercises teach you to build skills that reference material too long to embed directly in the SKILL.md.
The Problem: A company (fictional "NovaTech") has a detailed brand guide that governs all communications. Instead of re-reading the guide every time you create content, you build a skill that automatically applies the brand voice.
Your Task: Study the provided brand guide, then build a brand-voice-skill/ with a SKILL.md that tells Claude to read the brand guide from references/ before writing anything. Define when to activate (any NovaTech content), how to apply voice rules, how to handle different content types (technical vs. marketing), and include a brand compliance checklist. Test on five content tasks. Then extend the skill with a compliance review mode that scores existing content against the brand guide.
What You'll Learn:
Starter Prompt:
"Read brand-guide-source/brand-guide.md thoroughly. Then help me build a skill that enforces NovaTech's brand voice. The skill should reference the brand guide and include a compliance checklist."
Reflection Questions:
The Problem: Your company has an AI Usage Policy. Before any AI-generated content goes public, it must pass compliance. Checking manually is slow and inconsistent. You build a skill that automates the check.
Your Task: Build a policy-checker-skill/ that reads content, reads the policy document from references/, checks each policy requirement, and produces a compliance scorecard with PASS/FAIL per requirement. Flag specific violations with quotes from the content and suggest fixes. Test on three content samples (at least one should pass, at least one should fail). Then add a "fix it" mode that rewrites violating content to bring it into compliance.
What You'll Learn:
Starter Prompt:
"Read policy-document/ai-usage-policy.md. Let's build a skill that checks any content against this policy and produces a compliance scorecard. I want PASS/FAIL for each requirement with specific quotes if it fails."
Reflection Questions:
The Problem: You're creating course materials that must align with a teaching standards framework (the "Seven Pillars of AI-Driven Development"). Instead of manually checking alignment for every lesson, you build a skill that enforces it automatically.
Your Task: Build a curriculum-skill/ with SKILL.md, a reference to the Seven Pillars standards document, and a lesson plan template. The skill should generate aligned lesson plans, create exercises targeting specific pillars, review existing content for alignment, and produce an alignment matrix. Test by generating three lesson plans, then run a gap analysis across all three to identify which pillars are well-covered and which are underrepresented.
What You'll Learn:
Starter Prompt:
"Read standards/seven-pillars.md and templates/lesson-plan.md. Build a skill that generates lesson plans aligned with the Seven Pillars and can check existing content for coverage gaps."
Reflection Questions:
Building a skill is half the work. Making it reliable is the other half. These exercises develop your ability to find where skills break, measure improvement, and get feedback from others.
The Problem: You have an invoice processing skill that works on standard invoices. But invoices in the real world are messy: handwritten notes, multiple currencies, missing fields, unusual formats. Your job is to break the skill, then fix it.
Your Task: Test the provided invoice skill against normal invoices (should work) and tricky invoices (will break). Document every failure. Then create 3 new invoice files specifically designed to break the skill in ways the provided tricky set didn't. Fix the skill to handle all edge cases and verify the fix works across every test invoice.
What You'll Learn:
Starter Prompt:
"Test invoice-skill/SKILL.md on the invoices in invoices/normal/ first. Then try the ones in invoices/tricky/. Let's document every failure and figure out why it breaks."
Reflection Questions:
The Problem: You changed a skill, but did it actually get better? Without measurement, you're guessing. This exercise teaches you to use a fixed rubric and fixed test cases to isolate the effect of skill changes.
Your Task: Run five customer emails through v1 of an email response skill and score each response on the provided rubric. Identify the three biggest weaknesses. Build v2 addressing those weaknesses. Re-run the exact same five emails through v2 using the exact same rubric. Compare: Did scores improve? Did fixing one thing break another?
What You'll Learn:
Starter Prompt:
"Run all 5 emails in customer-emails/ through email-response-skill-v1/SKILL.md. Score each response using rubric.md. Let's record the scores in scorecard.md."
Reflection Questions:
5.3User Testing: Does Your Skill Work for Others?
The Problem: A skill that works for you might confuse someone else. The ultimate quality test is handing your skill to another person and seeing if they can use it without help.
Your Task: Package your best skill from previous exercises with a README, reference files, and sample test prompts. Swap with a classmate. Test their skill: run their test prompts, then create 2 new prompts of your own. Score on output correctness, quality, usability, and suggestions for improvement. Receive their feedback on your skill and create a v2.
If working solo: wait 24 hours, then re-read your SKILL.md with fresh eyes. The time gap creates enough mental distance to see your own skill's weaknesses.
What You'll Learn:
Starter Prompt:
"Help me package my best skill for someone else to use. I need a README explaining what the skill does, how to use it, and 3 sample test prompts."
Reflection Questions:
Individual skills are useful. Skills that work together are powerful. These exercises teach you to chain skills into pipelines, organize them into libraries, and build skill sets for teams.
The Problem: A single task often involves multiple steps: raw meeting notes become formatted minutes, which become action items, which become team notifications. Instead of running each skill manually, you build a pipeline that chains them together.
Your Task: Build (or reuse from earlier exercises) three skills: Meeting Minutes, Action Item Extractor, and Task Notifier. Then create a pipeline-skill/SKILL.md that orchestrates all three: reads raw notes, passes output from each step to the next, and produces a final summary. Test on provided meeting notes. Then add error handling: what if there are no action items? What if an item has no assignee?
What You'll Learn:
Starter Prompt:
"Let's build a meeting notes pipeline. We need three skills that chain together: raw notes into minutes, minutes into action items, action items into team notifications. Start by showing me the data flow."
Reflection Questions:
The Problem: You've built skills across multiple exercises. They're scattered in different folders with different structures. You need to organize them into a reusable personal library.
Your Task: Inventory every skill you've built. Standardize each to a consistent structure (SKILL.md, README.md, optional examples/templates/references folders). Create a LIBRARY.md index categorizing skills by type (Writing, Data, Workflow). Pick your top 3 and polish them to production-ready: complete documentation, at least 2 test cases each, all edge cases handled. Package them so someone else could use them without explanation.
What You'll Learn:
Starter Prompt:
"Help me inventory all the skills I've built across these exercises. For each one, note: name, what it does, completeness level (draft/tested/production-ready), and whether it needs other files."
Reflection Questions:
The Problem: A design agency has 5 roles that each need AI skills: Creative Director, Senior Designer, Junior Designer, Project Manager, and Admin. The skills must work together, using consistent terminology and cross-referencing where relevant.
Your Task: Build one skill for each role: Creative Brief, Design Feedback, Asset Checklist, Status Update, and Client Invoice. Each must be usable by someone unfamiliar with it. Skills should share consistent project and team terminology and cross-reference each other where relevant (the invoice skill references project names from the status update skill). Create a LIBRARY.md indexing all five with usage guides.
What You'll Learn:
Starter Prompt:
"We're building AI skills for a 5-person design agency. Each role gets one skill. Let's start with the Creative Director's Creative Brief skill, since other skills will reference its outputs."
Reflection Questions:
These exercises simulate building skills for actual business needs. The quality bar is higher: a non-technical person should be able to use each skill by reading only its README.
The Problem: You need a production-ready skill for extracting data from invoices in multiple formats (text, structured, informal). It must handle multiple currencies, detect anomalies (negative amounts, missing dates, duplicates), generate both a data file and a summary report, and include clear error messages when extraction fails.
Your Task: Build a complete invoice-processor/ skill. Test against 10 provided invoices of varying complexity. Compare your extraction against expected output for the first 5 (provided), then do a blind test on invoices 6-10. Polish until a bookkeeper could use it with zero additional instructions.
What You'll Learn:
Starter Prompt:
"Let's build a production invoice processor. Start with a basic SKILL.md, test it against the first 5 invoices in invoices/, and compare to expected-output.csv. We'll iterate from there."
Reflection Questions:
The Problem: A single topic needs to become a complete content package: blog post, LinkedIn posts, Twitter/X thread, and email newsletter blurb. All outputs must convey the same core message but be adapted to each channel's length, tone, and formatting conventions.
Your Task: Build a content-pipeline-skill/ that takes a topic brief (subject, key points, audience, CTA) and produces all four content types. The skill must reference provided channel guidelines and avoid patterns shown in anti-examples. Test on 3 content briefs and verify cross-channel message consistency.
What You'll Learn:
Starter Prompt:
"Read channel-guidelines.md and anti-examples.md. Then let's build a content pipeline skill that takes a brief and produces blog post, LinkedIn, Twitter, and email content. Start with content-briefs/ brief #1."
Reflection Questions:
The Problem: You need a skill that conducts structured research on any topic and produces a comprehensive analysis document with citations, key themes, disagreements, and a confidence assessment of the findings.
Your Task: Build a research-skill/ that accepts a research question, defines scope, gathers information from provided documents or web search, synthesizes findings, and produces a structured research memo. The output must follow the provided analysis template and include a confidence assessment. Test across three research questions with different source requirements (documents only, web only, combined).
What You'll Learn:
Starter Prompt:
"Read analysis-template.md and the first research question in research-questions/. Build a research skill that can gather from source documents, web search, or both, and produce a structured analysis."
Reflection Questions:
These capstones bring everything together. Each asks you to build a complete, interconnected skill suite for a real scenario. Choose one based on your interests, or tackle all three.
ABusiness Operations Suite
The Mission: Build 6 interconnected skills for Pixel Perfect Design Studio, a 5-person graphic design agency. Skills must share consistent context, cross-reference each other, and include a pipeline orchestrator.
Skills to Build:
Quality Bar: All skills share terminology, each has 2+ test cases, the pipeline chains at least 3 skills, and a non-technical person could use any skill from its README alone.
What You'll Learn:
The Mission: Build 6 skills that power a course delivery system. These help instructors create, deliver, and assess content consistently while maintaining alignment with a teaching standards framework.
Skills to Build:
Reference Materials: Seven Pillars standards document, tone guide, and a sample lesson plan are provided.
What You'll Learn:
The Mission: This capstone is different. Instead of a provided scenario, you build skills for YOUR real life. Audit your own recurring tasks, select the top 5, build complete skills for each, and measure the actual time saved.
Your Process:
What Makes This Special: Unlike Capstones A and B, these skills solve YOUR actual problems. This is where skill-building stops being academic and becomes genuinely useful. The skills you build here are ones you'll keep using after this course.
What You'll Learn: