Claude, Expansion Kickoff.
This video corrects the assumption that installing a skill teaches Claude a new capability, using the expansion-kickoff skill to show it is only a followed checklist.
Someone might assume that when Claude picks up a skill called expansion-kickoff, it actually learned how to run a business expansion. It didn't. This video uses that skill as a case study to explain what a Claude skill really is: a specification Claude follows, not a capability the model acquires.
A skill is one file, read before acting
A skill is a folder holding a single file, SKILL.md, written in plain language. Claude reads that file before it starts and works through it top to bottom, step by step, in the order it's written, with no branching unless the file itself calls for it. Expansion-kickoff, specifically, is described as a folder holding instructions for starting a structured checklist when a team or business expands into somewhere new.
Nothing in the model changes
The video makes a direct test of the "did Claude learn something" question: delete the skills folder and see what happens. Claude doesn't forget how to plan an expansion, because there was nothing to learn in the first place. It just stops following that one routine. Nothing in the underlying model changes when a skill is added or removed, only whether Claude has a specific set of written steps in front of it to follow.
Specification, not power
This is the video's core distinction: a skill is a specification, not a new power. The payoff is consistency, the same checklist, the same starting steps, every single kickoff. The limit is exactly as narrow as the file itself: anything outside those written steps, and Claude is off the map, with no fallback knowledge the skill was supposed to provide. Expansion-kickoff never taught Claude how to run a real expansion. It only guarantees that every time it runs, Claude reads that same file and starts the same structured checklist.
Why a tidy output proves nothing either way
The video makes a point of not over-reading Claude's output as evidence. Watching Claude produce a tidy, well-organized kickoff plan doesn't prove it understood the business, a file can be followed to the letter in a situation the model never really grasped. Equally, watching it produce a rough or awkward plan doesn't prove the skill is broken. It may simply be a case the checklist doesn't fit well. Neither outcome, by itself, is proof of understanding or failure.
The prompt to try
The video's suggested exercise is to pick one process a person repeats every time they start something new, then ask Claude to write a SKILL.md for it in plain language with ordered steps, and have Claude read it back and walk through exactly what it will do before doing it. The point of running this against a real repeated process is to see the same specification-versus-power distinction play out with something concrete.
Key takeaways
- A Claude skill is one file, SKILL.md, that Claude reads before acting and follows step by step in order.
- Deleting a skill's folder does not make Claude forget anything, because nothing was learned to begin with.
- A skill is a specification, not new model capability: it guarantees the same checklist every run, nothing more.
- Anything outside the file's written steps is outside what the skill covers.
- A tidy or rough output from a skill run does not, on its own, prove understanding or a broken skill.
Who this is for
This is for anyone trying to understand what installing a Claude skill actually changes, especially people confusing "the model learned this" with "Claude has a checklist to follow."
Chapters
- 0:00The naive framing: "Claude learned expansion planning"
- 0:11Sounds like a strategy upgrade
- 0:21Broken, with a case — delete the folder, nothing is forgotten
- 0:31The anchor: one file — expansion-kickoff's single SKILL.md
- 0:46Read it, then follow it in order
- 0:55Spec, not power — the payoff and the limit
- 1:06The anchor returns — same file, every run
- 1:19Both directions — neither one is proof
- 1:34Carry-out
- 1:43Your turn
- 1:56Outro
Full transcript(auto-generated, with timestamps)
The naive framing: "Claude learned expansion planning"
[0:00]You guess Claude learned expansion planning from a skill called expansion kickoff. It didn't. It was handed a file of instructions. Expansion kickoff just names one such file. Let's see what's actually inside it. Here Claude has an
Sounds like a strategy upgrade
[0:12]Expansion kickoff skill, and it sounds like a strategy upgrade, like it studied business expansions and came back knowing something it didn't before.
Broken, with a case — delete the folder, nothing is forgotten
[0:21]But nothing in the model changes. Delete the skills folder and Claude doesn't forget how to plan an expansion. There was nothing to learn. It just stops following that one routine.
The anchor: one file — expansion-kickoff's single SKILL.md
[0:31]Here's what a skill actually is, one file. Expansion kickoff, for instance, is a folder holding a single skill. MD instructions for starting a structured checklist when a team or business expands somewhere new, written in plain language Claude reads before it starts.
Read it, then follow it in order
[0:46]Claude reads that file top to bottom and works through it step-by-step in the order it's written, no branching unless the file itself says branch.
Spec, not power — the payoff and the limit
[0:55]That makes a skill a specification, not a new power. The payoff, the same checklist every kickoff, the same starting steps every time. The limit, anything outside those written steps and Claude is off the map. So expansion
The anchor returns — same file, every run
[1:07]Kickoff never taught Claude how to run a real expansion. It just guarantees that every time expansion kickoff runs, Claude reads that same file and starts the same structured checklist. That's the whole trick.
Both directions — neither one is proof
[1:19]Watching Claude produce a tidy kickoff plan doesn't prove it understood the business. A file can be followed to the letter on a situation it never really grasped. And watching it produce a rough plan doesn't prove the skill is broken. It may just be a case the file's checklist doesn't fit. A skill doesn't
Carry-out
[1:35]Teach Claude how to run an expansion. It's a file of steps Claude reads before it starts. So every kickoff comes out with the same checklist every time.
Your turn
[1:43]Your turn. Paste this into Claude. Pick one process I repeat every time I start something new. Write me a skill. MD for it, plain language ordered steps, then read it back to me and walk me through exactly what you'll do before you do it.
Outro
[1:56]Claude kickoff, Liam in for Bear.
