AI Skill Library Register

To use this register, copy it with the "Copy page" button above or open the Google Doc version. Read the companion guide to running a company skill library for the reasoning behind each rule.

AI skill library register

A free template for AI Ops leads who run a shared library of AI skills. Use one copy for the whole organisation. Log every skill other people rely on, review each one before it's published, and retire what nobody uses. It works with any AI tool that supports shared skills, and maps to Claude's Plugins & skills settings on Team and Enterprise plans.

0. Library details

• ☐ Organisation:
• ☐ Library owner (AI Ops lead):
• ☐ Reviewers (at least two, not the authors):
• ☐ Publishing setting (Open / Requires review / Off):
• ☐ Security scanning on (Enterprise)? Yes / No / Not available
• ☐ Monthly review date:

1. Library rules

• Every published skill has one named owner: a team or a person.
• Authors never review their own skills.
• Test before publishing: three to five test prompts, covering when the skill should trigger, when it shouldn't and one ambiguous case.
• Merge duplicates rather than publishing both. Credit both authors.
• Default to "available to install". Use "installed by default" or "required" only when a whole group needs it.
• Scope specialist skills to the group that uses them.
• Retire skills with no uses in 90 days, no owner, or a process that has changed.

2. Skill register

One row per skill that other people rely on. Status: Proposed, In review, Changes requested, Published or Retired.

3. Other accelerators

Claude's library holds the skills you've provisioned. List everything else people could reuse here: plugins, prompts, apps, agents, guides and training, including things that live outside Claude or that some people can't reach yet. Type: Plugin, Prompt, App, Agent, Guide or Training.

Registering entries with the Kowalah plugin. In Claude, run /my-ai-tools from the Kowalah plugin to record the skills, plugins, connectors and projects your session uses in your organisation's AI register. You see the list and choose what's shared; it records names and IDs, not contents. To find what colleagues have already built, ask Claude "have we done this before?" and it searches your organisation's accelerators and Kowalah's library. Recording is free.

4. Intake questions

Ask the submitter to answer these in the submission note.

1. What work does this skill do, in one sentence?
2. Who is it for: everyone, or one team?
3. What does it touch: connectors, files, scripts or websites?
4. How did you test it? Paste your test prompts and what happened.
5. Who will own it once it's published?
6. Does it replace or overlap an existing skill?

5. Review checklist

Security

• ☐ Every file read: the SKILL.md, any referenced files and any scripts.
• ☐ Scripts do only what the description says.
• ☐ No network calls, or URLs to unexpected domains.
• ☐ No hard-coded passwords, API keys or tokens.
• ☐ Connectors and other tools it calls are listed and expected.
• ☐ No instructions to hide actions, skip safeguards or send data elsewhere.
• ☐ Scan result checked (pass, warn or fail), where scanning is available.

Quality

• ☐ Triggers for the right requests and stays quiet for unrelated ones.
• ☐ Doesn't clash with, or steal triggers from, existing skills.
• ☐ Follows its own steps.
• ☐ Output is good enough for the process owner to sign off.

Fit

• ☐ Not a duplicate, or merged with the existing skill.
• ☐ Owner named and has agreed.
• ☐ Audience and "offered as" setting agreed.

Decision: ☐ Approve ☐ Request changes ☐ Decline. Reviewer: ______ Date: ______ Note: ______

6. Duplicate and merge log

7. Retirement and offboarding

• Monthly: list skills with no uses in 90 days. Ask each owner: update, keep or retire?
• When an owner leaves or changes role: reassign the skill within two weeks, or retire it.
• When a process changes: the process owner flags every skill it affects.
• Keep the last version and the reason in this register, even after you remove a skill from the library.

8. Monthly review

In Claude, usage analytics show uses and cost per use for each skill on Team and Enterprise plans.

9. Outcome log

Record a skill here when it changes how a process runs.

10. Next skills to build (ICE)

Score each opportunity for impact, confidence and ease, each from 1 to 10, and multiply them for a score out of 1,000. Build the highest first.

11. Steering group lines

• Published skills: [number] (up [number] this month), each with a named owner.
• Merged or retired this month: [number].
• Most used skill: [name], [number] uses. Outcome: [result].
• Waiting for review: [number], oldest [number] days.
• Decision needed: [for example, make a skill required for all staff].

12. Implementation notes

• Start with the skills people already share. Don't commission new ones until the register exists.
• Keep skill source files in one place with version history, such as a Git repository, so you can roll back a bad change.
• Name reviewers from outside the authoring team.
• This template isn't security or legal advice. For skills that touch personal or client data, involve your security team and DPO.