Turn every finished build into a reusable Claude skill

When you finish a client build, the temptation is to file away the code and move on. But the thing with real, compounding value isn't that one folder of code — it's the repeatable PROCESS you just figured out. This is the shift Albert Olgaard teaches for running a one-person business on Claude: you're not selling hours, you're compounding skills. Capture the process as a reusable Claude skill and your next client who needs the same thing is one slash command away — not a rebuild from scratch.

The moment a build actually works, you've discovered a repeatable process. Don't let it evaporate. Have Claude write up the exact steps you just took as a skill: a folder with a SKILL.md describing what it does, when to use it, and the procedure.

A raw capture bakes in THIS client. That's worthless on the next job — worse, it quietly runs the first client's settings on the second client's project. The fix: make the skill ASK for what's client-specific at run time instead of hardcoding it.

If the skill references paths back into the original project, it can't travel. A reusable skill has to run on its own folder, in any repo.

Put the finished skill where every project can see it, and your next client becomes a single command.

This is the SKILLIFY meta-prompt. Run it in the project where a build just succeeded; it converts that work into a clean, generic, standalone skill in one pass.

You just finished a working build in this project. Convert the exact
process you followed into a REUSABLE Claude Code skill, following these
rules strictly:

1. SUMMARISE the end-to-end procedure you actually used (commands,
   order, key decisions).
2. CREATE a skill folder named after the TASK (not this client) with a
   SKILL.md inside. SKILL.md = a short description (so you know WHEN to
   use it) + the step-by-step instructions.
3. MAKE IT GENERIC: find every value specific to THIS client (names,
   domains, IDs, keys, URLs) and replace each with a placeholder the
   skill ASKS the user for at run time. Hardcode nothing client-specific.
   List the inputs it will ask for at the top.
4. MAKE IT STANDALONE: remove every path reference back to this project.
   Use relative paths or ${CLAUDE_SKILL_DIR}. Bundle any files the skill
   needs inside its own folder.
5. ADD a 30-second self-test: how to confirm the skill runs correctly
   for a NEW client with no edits.
6. OUTPUT the full SKILL.md and the folder layout, ready to drop into
   ~/.claude/skills/.

Before you finish, re-read your SKILL.md and confirm: would this run for
a totally different client, in a different repo, with zero edits?
If not, fix it.