Stricks Learn · one small thing

One Move: Turn a Repeating Prompt Into a Skill

July 22, 2026 · today's lever: Agent Skills

TL;DR When you catch yourself pasting the same multi-step instructions into chat for the third time, stop pasting and start packaging. A skill is that set of instructions written once, given a name and a trigger description — Claude picks it up on its own the next time something matches. The move is trusting the trigger description enough to stop repeating yourself.

↺ From previous weeks

Nothing yet — first page. From week two on, this ribbon holds 2–3 one-line refreshers pulled from earlier pages.

The move

  1. Notice the repeat: you're about to type nearly the same checklist you typed last week, for a different task.
  2. Write it once as a skill: a short description of when to use it, plus the actual steps. In Claude Code that's a file at .claude/skills/<name>/SKILL.md; other Claude surfaces have their own skill locations. The description is what gets matched against your future asks — write it like you're explaining the trigger to someone else, not summarizing the topic.
  3. Let Claude find it, not you: the skill only loads into context when something matches its description, so it can sit unused for months at effectively zero ongoing cost.
▶ Try it now — package something you keep retyping ~cheap 
I keep giving you almost the same set of steps for [the recurring
thing]. Here's what I usually ask: [paste your usual instructions
or checklist]. Turn this into a skill, and write a trigger
description specific enough that you'll reach for it on the right
future asks but not the wrong ones.
Where it breaks — the three edges worth knowing

1. Vague triggers are the real failure mode. "Helps with docs" won't fire when it should, and might fire when it shouldn't. Specificity in the trigger description is the whole mechanism — there's no other dial to turn.

2. Unused skills aren't free. A skill's description sits in context on every ask that looks even a little relevant. Stale, forgotten skills are common enough that Claude Code's own /checkup command has a dedicated check for them — that's next week's page.

3. Skills are instructions, not code. If you write down the wrong process, Claude will confidently follow the wrong process. Reviewing your own skills occasionally matters as much as writing them.

Under the hood

What actually runs when a skill fires — for when you want to read the machinery, not just use it.

Open the hood: the format, what loads, and the primary sources
The lever
Agent Skills — packaged judgment in a markdown file, not a new model capability. It doesn't teach Claude anything it couldn't already do; it tells Claude when and how to do a specific thing.
What it actually calls
Claude Code reads a directory at .claude/skills/<name>/SKILL.md. Frontmatter controls whether you invoke it directly (/name) or Claude triggers it automatically; the description field is the literal text compared against your request to decide relevance.
The standard underneath
The open Agent Skills standard — the same file shape is meant to work across more than one AI tool, not just Claude.
Primary sources
agentskills.io — the open standard itself. · code.claude.com/docs/en/skills — Claude Code's implementation (invocation control, subagent execution). · github.com/anthropics/skills — Anthropic's own public skill library, including the skill that builds other skills.

Watch & listen

⚡ What did this spark?

A question this page raised, a thing that didn't work, or a topic you want next. It lands in Learn Sparks — and the weekly run turns it into a future page or a quick answer in the digest. You choose which on the form.

⚡ Send a spark

Opens the Learn Sparks form in a new tab — about ten seconds, one required field.