← Learn
Playbook · 10 steps · checked 4 October 2026 · 0 adopted

Skills

At most ten steps, in order of how much they help. Each step is from Anthropic's own pages, linked. Re-checked against those pages every week; anything that can't be confirmed is marked unverified rather than removed.

  1. 01

    Make a skill when you keep repeating yourself

    Create one when you keep pasting the same instructions, checklist or multi-step procedure into chat, or when a section of CLAUDE.md has grown into a procedure rather than a fact. A skill loads only when it is needed, so you stop repeating the same guidance in every conversation. One-off instructions belong in the prompt. (Claude Code skills, Skills overview)

  2. 02

    Write the description as the trigger

    Until a skill is used, only its name and description sit in context, and Claude picks a skill by matching your request against the description. So say what the skill does and when to use it, in the third person ("Processes Excel files", not "I can help you"), with the key terms a request would contain. name is at most 64 characters (lowercase letters, numbers, hyphens; no "anthropic" or "claude"); description is at most 1,024 characters. In Claude Code, put the key use case first: the description plus when_to_use text is cut at 1,536 characters in the skill listing. (best practices, Claude Code skills)

  3. 03

    Keep SKILL.md short

    Keep the body under 500 lines and add only what Claude does not already know; ask of each paragraph whether it earns its tokens. In Claude Code a loaded skill stays in the conversation across turns, so every line is a recurring cost. Give one default way of doing things rather than a list of options, and avoid dates that will go stale. (best practices, Claude Code skills)

  4. 04

    Put detail in separate files, one level deep

    Treat SKILL.md as a table of contents that links straight to reference files; Claude reads a linked file only when the task needs it. Links nested inside linked files may be only partly read. Use scripts for fixed, error-prone steps: Claude runs them and only their output enters context. (best practices, Skills overview)

  5. 05

    Test it before you rely on it

    First run Claude on real tasks without the skill and note what goes wrong. Write at least three test scenarios for those gaps, then write just enough to fix them and compare against that baseline. Test with every model you plan to use: what is enough for Opus may be too little for Haiku. There is no built-in way to run these tests yet. (best practices)

  6. 06

    Let Claude write it, then use it in a fresh session

    Finish a task with Claude normally, then ask it to turn what you explained into a skill; no special prompt is needed. Trim any explanation Claude already knows. Use the skill in a fresh session on similar tasks, and take what it missed back to the first session to fix. (best practices)

  7. 07

    Put it where it will run

    Claude Code: a folder with SKILL.md in ~/.claude/skills/ (all your projects) or .claude/skills/ (this repo); set disable-model-invocation: true on anything with side effects, such as a deploy, so it runs only when you type /name. API: upload through the Skills API (/v1/skills) and use it with the code execution tool; it runs with no network access and no package installs. claude.ai: upload a ZIP whose root is the skill folder; code execution must be on. (Claude Code skills, Skills overview, using skills in Claude, creating custom skills)

  8. 08

    Custom skills do not sync across surfaces, so add each one where you need it

    A skill uploaded to claude.ai is not on the API, an API skill is not on claude.ai, and Claude Code's own skill folders are separate from both. One exception: Claude Code signed in with your claude.ai account downloads the skills enabled there (v2.1.273 or later, checked about every 10 minutes). That sync is one-way: it never changes your claude.ai account. Cloud sessions do not read ~/.claude/skills/; they get your claude.ai skills plus the repo's .claude/skills/. (Skills overview, Claude Code skills, using skills in Claude)

  9. 09

    Share at the right scope

    Claude Code: commit .claude/skills/ to the repo, or put the skill in a plugin's skills/ folder. API: uploaded skills are shared across the workspace. claude.ai on Team and Enterprise: share a skill with named people or groups, or submit it to the organisation's library so anyone can install it. (Claude Code skills, Skills overview, using skills in Claude)

  10. 10

    Treat other people's skills like software

    Use skills you wrote or got from Anthropic, and read every file of any other skill before turning it on: a skill can run code and call tools. Never put API keys or passwords in a skill. Anthropic's example skills install in Claude Code with /plugin marketplace add anthropics/skills; they are for demonstration, so test before relying on them. (Skills overview, creating custom skills, anthropics/skills)

Picked from around the web, with a one-line takeaway. The originals are by their creators.

  1. instagram

    There's a new claude skill that can reverse-engineer any app and rebuild it as your own ↗

    Dripthecup (@dripthecup) · still works (checked 8 Oct 2026)

    Introduces Replica, a Claude skill that maps an existing app's screens, reads its reviews and estimates a first slice to build as your own.

Unverified

What Anthropic's own pages disagree on, or could not be confirmed.

  • Description limit. The platform docs say 1,024 characters; the creating custom skills help article says 200. Staying under 200 is safe everywhere.
  • Sharing on claude.ai. The Skills overview still says claude.ai skills are individual only with no organisation-wide distribution; the newer help article describes sharing, an organisation library and admin provisioning on Team and Enterprise.
  • Plans and menu path on claude.ai. The overview says Pro, Max, Team and Enterprise and "Settings > Features"; the help article says Free plans too and "Customize > Skills". Check in the app.