Skill Guide: Turn Repeated Work Into Reusable Skills

You have one Skill; now build the system around it

This chapter walks you through creating a second Skill, then setting up the weekly review that turns this from a one-time exercise into a compounding system.

Choose your second Skill

Go back to your discovery table. You identified repeated tasks. You built the first one into a Skill. Now pick the second one. If none of the remaining candidates feel right, use the pattern table below to spark a choice, then try one of the five scenarios that follow.

Skills cluster into recognizable patterns. Now that you have built one, this table helps you name what kind of Skill your second candidate is and choose the right structure for it.

If none of your discovery candidates feel right, choose one of the five scenarios below. Each is a task most knowledge workers repeat weekly.

Scenario A: Follow-Up Email After a Meeting

You finish a call and draft the same kind of email every time: decisions first, then action items with owners, then open questions. You keep correcting the AI on your format, your tone, and what to leave out.

To create this Skill, start a conversation and say: 'I want to create a skill that drafts follow-up emails after meetings. When I paste a transcript or my notes and say draft the follow-up, it should produce a ready-to-send email with decisions first, then action items with owners, then open questions. It should match my voice samples and never include internal-only commentary in the email.'

Scenario B: Weekly Status Update

Every week you summarize what happened for your manager, your client, or your team. The structure is always the same. The detail level is always the same. You keep restating both even though the AI may already know your basic preferences.

To create this Skill, say: 'I want to create a skill for weekly status updates. It should take my rough notes from the week and produce a three-section update: completed, in progress, and blocked. One sentence per item. Blockers should be flagged, not buried. It should never list meetings as accomplishments.'

Scenario C: Draft Review Before Sending

Before you send an email, a proposal, or a slide deck, you run the same mental checklist: right tone, no banned words, no accidental budget commitments, right audience level. That checklist belongs in a Skill so the AI can run it for you.

To create this Skill, say: 'I want to create a skill that reviews drafts before I send them. When I say review this or is this ready to send, it should check for my banned words, verify audience fit, flag anything that commits budget or schedule, and produce a review table with issue, location, severity, and suggested fix. It should flag issues, not rewrite the draft.'

Scenario D: Notes-to-Structure Converter

You come out of a meeting, a brainstorm, or a research session with a pile of rough notes. You need them in a table or a brief with a consistent structure, and you keep telling the AI the same column names and the same rules about handling ambiguity.

To create this Skill, say: 'I want to create a skill that turns raw notes into structured documents. When I paste unstructured notes and say organize this or structure these notes, it should produce a table with the columns I specify (usually item, owner, deadline, status). It should preserve ambiguity as tentative and never invent structure that was not in the source.'

Scenario E: Meeting Preparation Brief

You walk into meetings cold. You forget what was decided last time, who cares about which topic, and what the open threads are. A meeting-prep Skill pulls context from your recent history and produces a one-page brief you can scan in five minutes before the call.

To create this Skill, say: 'I want to create a skill for meeting preparation. When I say "prep for my call with [name]" or "meeting prep," it should produce a one-page brief with meeting context, open threads from prior meetings, stakeholder positions, three hard questions that surface real disagreement, and a suggested agenda. It should never invent stakeholder positions, and if it does not have calendar or email access, it should ask me to paste the relevant information.'

Build the second Skill end to end

Where each Skill should live

Global Skills live in the user-level Skill location for the tool you are using. Project Skills live inside a project folder or repository location that the tool reads only for that work. Claude Code and Codex expose these scopes differently, and the exact paths can change, so treat the product documentation and the tool's visible Skill list as the source of truth before installing. Tool-specific Skills live where that tool expects them and carry a note in the Skill file about the dependency.^{[1] [2]}

The weekly review that keeps everything current

A workflow library that never gets reviewed decays quietly. A 15-minute weekly review catches drift before it causes problems. The format is simple: scan recent AI sessions for repeated corrections, new Skill candidates, and foundation file updates.

The weekly review is where you exercise judgment. You look at the evidence, decide what matters, and choose which corrections deserve a permanent home. This chapter teaches you to recognize the maintenance patterns: repeated corrections, stale gotchas, trigger misfires, and foundation drift. A later chapter automates the observation layer so those patterns surface overnight, but the judgment stays with you. Start manual so you know what a productive review produces before asking AI to replicate it.

Your library is built; now learn what to trust from outside

Your workflow library is small right now: two Skills, three foundation files, a few gotchas, and a weekly review habit. That is enough. Before you automate any of this, the next chapter covers an important question: what happens when you encounter Skills that someone else wrote?

• I created a second Skill through conversation.
• I tested the trigger with should-fire and should-not-fire scenarios.
• I ran it three times with real input and asked AI to absorb my corrections.
• I asked AI to create one evaluation scenario.
• I set a recurring 15-minute weekly review on my calendar.
• I have two tested Skills and three foundation files.