The Full Monty
This is where everything comes together.
You’ve built individual skills, composed them into a marketplace, and worked with pre-built tools. Now you’ll use three different skill layers in a real workflow — demonstrating how skills at different abstraction levels compose without conflict.
The Layered Ecosystem
Skills don’t just add capabilities — they operate at different layers of the development process. Understanding these layers helps you choose the right tool and combine skills effectively.
| Layer | Purpose | Example |
|---|---|---|
Harness skills |
Shape HOW you work — process, methodology, guardrails |
superpowers (brainstorming, TDD, debugging workflows) |
Domain skills |
Provide WHAT you work with — domain-specific capabilities |
conference-toolkit (abstract-writer, conf-finder, trip-report) |
Workflow skills |
Track WHERE the work stands — state, progress, coordination |
beads (issue tracking, dependency management) |
The key insight: these layers don’t conflict because they operate at different abstraction levels.
A harness skill tells Claude how to approach a task. It shapes the process: "use TDD" or "brainstorm systematically."
A domain skill gives Claude what to produce: conference abstracts, API clients, documentation.
A workflow skill tracks progress: what’s done, what’s blocked, what depends on what.
When you invoke multiple skills together, Claude resolves them correctly because each operates at its own layer. A /brainstorming session can guide you toward using /conf-finder, and both can update a beads issue — no collision, no confusion.
Setting Up the Layers
Let’s install all three layers.
Install Superpowers (Harness Layer)
Superpowers is an official Anthropic plugin that provides process-oriented skills: brainstorming (structured ideation), TDD (test-first development), and debugging (systematic root-cause analysis).
These are "harness" skills — they shape how you work, not what you build.
claude plugin install superpowers@claude-plugins-official
Initialize Beads (Workflow Layer)
Beads is a Dolt-powered issue tracker that lives in your repository. Issues persist across conversation compaction and session restarts — they’re real data, not ephemeral context.
Create a test project and initialize beads:
mkdir -p ~/capstone-project && cd ~/capstone-project
git init
bd init
This creates a .bd/ directory with a Dolt database. Issues, comments, and metadata are tracked in Git-style versioned tables.
The Capstone Workflow
Scenario: You’ve been invited to submit a talk to a conference next month. You need to plan it, write it, and track the work.
We’ll use all three layers to complete this from start to finish.
Step 1: Create a Work Item (Workflow Layer)
Start by tracking the work. Create a beads issue:
bd create --title="Submit talk to upcoming conference" --description="Find a relevant conference, write an abstract, submit it" --type=task --priority=2
Beads assigns an ID (e.g., 1). This issue now tracks your progress. You can update it, comment on it, and close it when done — even if the conversation gets compacted or you restart Claude Code.
Step 2: Brainstorm the Topic (Harness Layer)
You have an idea — "something about Agent Skills" — but you need to refine it. Use superpowers to brainstorm systematically:
/brainstorming I want to give a talk about how Agent Skills change the way developers interact with AI coding tools. Help me explore angles and narrow down a focus.
Superpowers will guide you through a structured brainstorming session. It asks clarifying questions:
-
Who’s the audience? (Developers, engineering managers, AI/ML practitioners?)
-
What’s the outcome? (Practical takeaways, conceptual understanding, or both?)
-
What’s the scope? (Intro-level overview or deep-dive implementation?)
As you answer, it proposes angles:
-
"Skills as a new abstraction layer for AI agents"
-
"From prompts to protocols: how skills change the contract"
-
"Teaching AI how you work: process skills vs. domain skills"
You narrow down to: "Agent Skills: Teaching AI How to Work With You" — focusing on the layered ecosystem and real workflow examples.
This is the harness layer in action: superpowers didn’t write code or produce a deliverable. It shaped your process, helping you think clearly and land on a strong angle.
Step 3: Find the Right Conference (Domain Layer)
Now that you have a topic, find a venue. Use the domain layer:
/conference-toolkit:conf-finder Find technology conferences in the next 3 months that accept talks about AI developer tools or AI-assisted coding
The skill searches recent conference listings and returns:
-
AI Dev Summit (2 months out, CFP open, focus: AI-powered development)
-
DevTools Con (3 months out, CFP open, focus: developer productivity)
-
CodeGen Conference (2.5 months out, CFP closed — skip)
You pick AI Dev Summit. The domain layer gave you what you needed: conference options filtered by relevance and timeline.
Step 4: Write the Abstract (Domain Layer)
Time to write. Use the abstract-writer skill you built in Module 3:
/conference-toolkit:abstract-writer Write a full-session abstract about "Agent Skills: Teaching AI How to Work With You" for AI Dev Summit
The skill generates a complete abstract:
-
Hook: "AI coding assistants are powerful, but they don’t know how you work."
-
Promise: "Learn how Agent Skills let you teach AI your process, compose domain capabilities, and track work across sessions."
-
Outline: Introduction to skills, layered ecosystem, live workflow demo, Q&A
-
Takeaways: Practical understanding of skill architecture, examples to start from, confidence to build custom skills
Perfect. Copy it, polish if needed, and paste into the conference submission form.
Step 5: Close the Work Item (Workflow Layer)
The work is done. Close the beads issue:
bd close 1
Replace 1 with the ID from Step 1. The issue moves to closed status. If you run bd ready, it no longer appears in the ready list — the workflow layer knows this work is complete.
Step 6 (Stretch): Generate a Trip Report
After attending the conference, use the trip-report skill you built in Module 4:
/conference-toolkit:trip-report Generate a trip report for AI Dev Summit. Sessions attended: keynote on LLM evals, workshop on agent architecture, panel on AI safety. Key themes: evaluation rigor, production-readiness, human-in-the-loop design.
The skill produces a structured report: summary of sessions, key takeaways, people met, and follow-up actions. Share it with your team or post it internally.
How Skills Compose
Look at what just happened:
-
Three skill sources: superpowers (Anthropic marketplace), conference-toolkit (your marketplace), beads (third-party plugin)
-
Three purposes: process (brainstorming), domain (conference tools), workflow (issue tracking)
-
Zero conflicts: each operated at its own layer
Claude resolved skill activation correctly. When you said /brainstorming, it knew you meant the superpowers harness skill. When you said /conference-toolkit:conf-finder, it knew you meant the domain skill from your marketplace. When you ran bd close, it knew you meant the workflow CLI.
This is the "harness within a harness" concept in action. Superpowers is itself a harness — it provides process scaffolding — running inside Claude Code, which is also a harness. Skills at different layers compose cleanly because they don’t compete for the same role.
The layered ecosystem isn’t theoretical. You just used it to plan a talk, write an abstract, and track the work — all with skills operating at their appropriate levels.
Course Summary
Across five modules, you’ve built a complete mental model of the Agent Skills ecosystem:
| Module | What You Built |
|---|---|
Module 1 |
Understood what skills are and used a pre-built skill (simple-greeter) |
Module 2 |
Hand-wrote talk-outliner from scratch — learned skill anatomy with a single SKILL.md file |
Module 3 |
Used skill-creator to generate abstract-writer — refined it with templates and supporting files |
Module 4 |
Built conf-finder and trip-report, bundled all three into conference-toolkit marketplace |
Module 5 |
Used three skill layers together (harness + domain + workflow) in a real workflow |
You now understand:
-
How skills are structured (prompts, arguments, context)
-
How to write them by hand and generate them with AI
-
How to compose them into marketplaces
-
How skills at different layers work together without conflict
The Agent Skills ecosystem is young. There are opportunities to contribute: build domain-specific skills for your field, share them on public marketplaces, improve tooling like skill-creator, or write harness skills that encode your team’s process.
Module 6 (bonus) covers programmatic skill use — how to invoke skills from code and build automation on top of the skills layer.
You’re ready to build.