The Prompt That Teaches Any Skill You Install
When we finished module 5 of the Freedom Ag training, we had five skills to ship: humanizer, spin-selling, influence, storybrand, and pdf. Five skills means five distinct behaviors. It also meant a choice.
We could write five tutorials explaining what each one does. Or we could write one prompt that makes Claude write the tutorial itself, live, for whichever skill the user just installed.
We went with the second option, and it composes better than we expected.
What the Prompt Does
The first prompt in module 5 is labeled "Start here, make Claude teach you the skill you just installed." The user types the skill name in place of a bracket, pastes it into Claude, and sends it.
Claude's job at that point is specific. Not to describe the skill from memory or guess from the name. To actually open it, read it, and then walk the user through four things: what it does and why the defaults would not have gotten there, what changes about Claude's answers when the skill is running, the exact words or phrases that activate it, and three jobs from a real week at Freedom Ag where it pays off.
The fifth step is a before-and-after: a short sample in, a short sample out, so the user can see the difference rather than take it on faith.
The Gate That Matters
There's one rule in the prompt that runs before anything else:
Before you explain anything, actually open that skill and read it. Do not describe it from memory, and do not guess from the name. If you cannot find it or cannot read it, say so plainly and stop there.
That instruction exists because a skill that didn't install correctly looks exactly like a skill that installed correctly from the outside. The file just isn't there. Without the gate, Claude would describe spin-selling from its training data, sound completely accurate, and the user would believe they had a working skill when they didn't.
The gate produces one of two outcomes: Claude reads the file and teaches from it, or Claude says it can't find the file and stops. Either way the user knows the actual state of their install.
Why One Prompt Scales
Freedom Ag will add more skills. They'll find things on GitHub they want to try. Six months from now they'll install something we didn't build and won't have a tutorial for it.
The same prompt works on any of those. Type the skill name, paste, send. Claude opens whatever is there and explains it the same way it explained the five we shipped. The prompt doesn't know about spin-selling specifically. It knows how to read a skill file and surface what matters about it for someone selling ag inputs.
We've used this pattern before in other contexts: one prompt that works generically over a known structure rather than five prompts hard-coded to five known objects. It's easier to maintain and it covers things that don't exist yet.
What This Changes for the Client
The README we included with the skills download explains the whole setup in about 90 seconds to read. At the bottom, there's a paragraph pointing to that first prompt:
Don't take my word for what any of these do. The first prompt on the training page makes Claude open the skill you just uploaded, read it, and walk you through it.
The client doesn't need to understand skills to get value from skills. They install one, run the teaching prompt, and Claude explains what they just installed in terms of their own week. That's a low enough bar that the install rate on the first skill actually converts to using it.
The prompt is also the diagnostic. If Claude says it can't find the skill, you know the install failed. You're not debugging six weeks later when the output seems off.