Most AI rollouts fail the same way: a tool gets bought, a memo goes out, and three months later two people use it while everyone else went back to old habits. When you roll out AI skills, you have an advantage — there is almost nothing to learn. The skills install once and fire automatically. But you still need a plan, because the failure mode isn't confusion. It's drift back to private prompts.

Here is the rollout that sticks, in five steps.

Step by step: the two-week-per-role rollout

1. Name an owner. One person holds the skill files, decides which packs are installed, and fields the "can we tweak this?" questions. Without an owner, five people quietly fork five versions and the standard is gone in a month.
2. Install once, in the right place. Solo users install into their own Claude.ai settings or Claude Code directory. Teams should share through a Claude Project so everyone works from the same files (see the install paths below).
3. Pilot one role for two weeks. Pick the role with the most recurring deliverables — support replies, client reports, listings. Have that person produce their normal work through the skills only. Collect what worked and what needed editing.
4. Expand pack by pack. Add the next role only after the first one holds. Each pack behaves like the last — same architecture, same standard — so each rollout gets faster.
5. Make the skills the official path. The rule that makes it stick: for deliverables a skill covers, the skill is how the work gets done. Private prompt stashes retire. Review output against the skill's own guardrails, not against taste.

The three install paths

Same files, three homes. Pick by how your team works:

New to the format entirely? Start with Claude Agent Skills, explained for business owners — it covers what's inside a SKILL.md file and how the automatic loading works.

What changes for each person (almost nothing)

This is the part to say out loud in the kickoff, because it kills the resistance: nobody has to learn prompting. The before/after looks like this:

The person's job shifts from "operate the AI well" to "review the output" — a much easier thing to hire for, train, and hold people to.

The mistakes that sink rollouts

• Rolling out everything to everyone on day one. Coverage without adoption is shelf-ware. One role, two weeks, then expand.
• No owner. Skills are plain-text files — editable by design. That's a feature with an owner and chaos without one.
• Letting private prompts survive. If the old prompt stash stays "allowed," the standard is optional, and optional standards lose.
• Buying packs that don't map to real work. A pack nobody's role matches will never be adopted. Run the fit math first — one pack or all nine? walks the break-even honestly.
• Reviewing against taste instead of the standard. The skill defines what good means — evidence-tied, assumptions flagged, nothing invented. Review against that, and reviews stop being arguments.

Why teams buy the whole library for rollouts

You can absolutely roll out one pack at a time — and small teams should. But teams covering several roles at once tend to buy the Complete Skills Library for the rollout itself, for one practical reason: every pack shares the same architecture and honesty rules, so the second, third, and ninth rollouts are repeats of the first. One purchase ($479 for all nine packs — $781 separately), one standard, one playbook, run as many times as you have roles. The full case for a business-wide skills setup is in the pillar: AI skills library: one standard for your whole business.