If you've been using Puzzle to map your processes, you already know the feeling — the moment the invisible becomes visible. Your workflows are documented. Your team can finally see how the work actually moves. That's a real win, and most businesses never get there.
But here's what most Puzzle users don't know yet: When you connect Claude to Puzzle, your maps can talk back using Claude skills for Puzzle.
Connect Claude — Anthropic's AI — to Puzzle through MCP, and you can ask plain-language questions about your workflows and get real answers. You can drop in a call transcript and watch Claude build the entire process map for you. You can ask "what breaks if we switch tools?" and get a risk-ranked impact report in seconds.
I built a set of Claude Skills to make all of that repeatable. I've been using this exact combination with my own clients, and it has changed how fast I can deliver real value — so I packaged it up for the Puzzle community.
Here's what they do, how to install them, and where to start.
First — what's a "skill"?
In Claude, a skill is a custom instruction package. It tells Claude how to behave for a specific kind of task: what to look for, what rules to follow, how to structure the answer. When you install a Puzzle skill, Claude knows when to use it based on the phrases you type, what to read from your Puzzle workspace, and how to shape the output.
Think of it like hiring someone who already knows Puzzle, already knows your processes, and is ready to work the second you ask.
Each skill is self-contained. You install only what you need, and they run independently — no shared files, no dependencies, nothing to untangle.
They come in two flavors: skills that read, and skills that build.
The read-and-analysis skills (start here)
These read your Puzzle workspace and answer questions, produce reports, or document your systems. Nothing changes the shape of your process, which makes them the safe place to start. There are seven.
Process Q&A — Any team member, not just the person who built the map, can ask how a process works and get a plain-language walkthrough. Every section, every step, who owns it, what tools are involved — without digging through the map or explaining it by hand. Ask "walk me through how client onboarding works," and Claude answers like a colleague would. It'll even flag things that need your attention that you never asked about.
Change Impact Analysis — This one saves you from expensive surprises. Ask "what breaks if we remove Zapier?" and Claude searches every step, then categorizes the fallout: red for what breaks completely, yellow for what needs updating, green for low-risk. Give it a replacement tool and it hands you a switch sequence — the order to make changes so nothing downstream falls apart. If you've ever pulled one tool and watched your systems collapse like a bad game of Jenga, run this first. Five minutes now versus days of cleanup later.
Tool Evaluation — Before you buy another tool, ask Claude whether you already own something that covers it. "Does our stack already handle automated email follow-up?" It reads your tools, your data models, and your active workflow steps — because a tool sitting unused in your catalog isn't the same as a tool doing real work. You get a verdict (covered, partially covered, or a gap), the evidence behind it, and sometimes a redundancy flag that points straight at money you could stop spending. This is the conversation to have before your team goes shopping for that next AppSumo app.
Onboard a New Hire — When someone new joins, Claude reads the relevant workflows and generates a customized onboarding brief written for that person, based on their role and the steps they own. It's written in plain language for the human, not for a Puzzle user: every process they touch, every tool they'll use, what comes before their step and what depends on it. The manual onboarding load, lightened.
And three more round out the set.
- System to Blueprint documents your tools and data models into Puzzle.
- Tool Architecture maps how a system is actually configured.
- Optimization Ideation reads a process and hands you a prioritized list of improvements. None of them touch the shape of your process — they just make it smarter to work with.
The build engines (when you're ready)
The build skills don't just answer — they build maps inside your Puzzle workspace. Which is exactly why confirm-before-you-build is a non-negotiable rule. Claude always shows you the plan and waits for your approval before it touches your canvas.
There are two, and they work together.
Build Puzzle Process — This is the one I use at the end of every discovery call. Finished a call and have a transcript from Fathom, Otter, wherever? Paste it in, or drop a Google Drive link, and say "vibe map this transcript and build the current-state workflow." Claude extracts the process — sections, steps, roles, tools, pain points, automation candidates — and shows you the full plan first. You review it, you talk it back and forth, you make it exactly right, and then you tell it to build.
Point the same engine at a proposal or scope of work, and it builds your future state from that. Transcript in, current state out. Proposal in, future state out. Every step named and noted. Tools associated. Email steps come with draft email copy. Every automation comes with build instructions. Under thirty minutes, start to finish. That's what a deliverable looks like when you let the robots do robot things.
The Changelog Engine — works quietly behind the first. Every time you build, it records what changed, why it changed, and which steps are affected — captured the moment the work happens. Because things change constantly in your organization, and that's usually where institutional knowledge goes to die. Three months from now, when someone asks "why is it done this way?", the answer is actually written down and linked right to the step.
Installing takes about 30 seconds
In Claude, go to Customize, then Skills. Each skill comes as a file ending in .skill or .md. Upload them by clicking Upload Skill, choose the file(s), and open. The skills are active immediately — no restart. Open a skill and you'll see a short description of what it does and the phrases that trigger it.
Start with the read skills. They're zero-risk — Claude reads your system but never rebuilds it. Bring in the build engines when you're ready.
(You'll need a Puzzle account for any of this to work. New to Puzzle? Start free here.)
The bottom line
It's easier than it's ever been to make your processes visible — and now, to make them answer back, analyze themselves, and get built for you. If you've already done the hard part and mapped your work, this is how you finally get the return on it.
Grab the skills below. And if you want to go deeper on using Puzzle and Claude together to document and automate your operations, book a call and let's talk.
