Skip to content
Get packs

Guide · Agent skill

How to use the Create Paperclip Bundled Skill agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

create-paperclip-bundled-skill is an agent skill, a folder containing a SKILL.md file that tells an AI agent how to do a specific job. This one is written for work on the Paperclip project: it tells the agent how to turn a tweet, task or blog post into a skill in Paperclip's skills catalog and deliver it as a reviewed pull request. This guide explains what the skill does, how to install it, how to trigger it, what it needs and where to be careful.

Create Paperclip Bundled Skillpaperclipai/paperclip · MIT

Quick start

  1. 1Download the ZIP. Download the skill ZIP from this page. Keep SKILL.md at the top level of the folder.
  2. 2Unzip it into a skills folder. Unzip into ~/.claude/skills/ for personal use, or into .claude/skills/ inside a project if only that project should use it.
  3. 3Or upload it in the Claude apps. In the Claude apps, open Settings → Capabilities → Skills and upload the ZIP.
  4. 4Check the Paperclip prerequisites. Confirm you have a Paperclip checkout with packages/skills-catalog, an authenticated GitHub CLI and PR permissions for the target repository.
  5. 5Ask with a clear source and goal. Make a request that names the source, such as a tweet URL or a task, and the outcome you want, such as an external reference or a new local skill.
  6. 6Review the agent's search and choices. Check that the agent searched for an existing skill first and that the kind, category and slug match the rules in the SKILL.md.
  7. 7Review the pull request. Read the generated changes and the PR description, including attribution and the validation run, before merging.

What does this skill make the agent do?

The skill walks an agent through turning source material into a catalog skill in packages/skills-catalog/. The procedure has a core rule, FIND before MAKE: the agent first searches the catalog, the repo and GitHub for an existing skill, and reuses or adapts one if it is good enough. Only when nothing usable exists does it write a new local skill.

The guide covers capturing the source material, choosing a kind, category and slug, either referencing an external skill pinned to a specific commit or authoring a local one, adding worked examples, regenerating the manifest and opening a pull request. The shipped SKILL.md is cut off mid-example, and its two example files cover the two paths, so later steps are not fully verified.

How do you install the skill?

Option one is to download the ZIP from this page and unzip it into your personal skills folder, ~/.claude/skills/, or into a project's .claude/skills/ folder so that only that project uses it. The skill's folder should contain SKILL.md at its root along with the examples folder.

Option two is to upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

  • Personal: ~/.claude/skills/
  • Project only: .claude/skills/ inside the project
  • Claude apps: Settings → Capabilities → Skills

How do you trigger the skill with a natural request?

You do not need a special command. The skill's description tells the agent to use it when asked to FIND or MAKE a skill and publish it as a bundled or optional catalog skill. Ask in plain language and name the source, the goal and the delivery format.

Requests that fit: a link to a tweet or post with a request to turn it into a Paperclip skill; a task asking to bundle an existing repo skill into the catalog; a task asking to add a published external skill to the catalog as a pinned reference.

  • Turn this tweet into a Paperclip catalog skill and open a PR.
  • Find a published SKILL.md on GitHub for code review checklists and add it to the catalog as an external reference.
  • Make a skill that has agents write user-facing text in Simplified English.

What do you need before using it?

The steps assume a Paperclip repository checkout that contains packages/skills-catalog. The agent also needs the GitHub CLI (gh), authenticated, for code and repo search and for pinning commits. Git and pull request permissions on the target repository are needed to open the PR.

The xc CLI for X/Twitter links is preinstalled and authenticated only in Paperclip engineering agent environments. Elsewhere the skill says to delegate the fetch to a teammate with X access. python3 appears in the example manifest for an external skill, but the SKILL.md does not require it directly.

  • Paperclip repository checkout with packages/skills-catalog
  • GitHub CLI (gh), authenticated
  • Git and PR permissions for the target repository
  • xc CLI only where it is preinstalled and authenticated

What does the workflow look like?

Step 0 captures the source: the post, thread, linked articles or READMEs, with the source URL recorded for attribution. Step 1 checks the catalog, the repo's .agents/skills/ and skills/ folders, and GitHub for existing work. Step 2 chooses kind (default optional), category and a lowercase kebab-case slug that is unique across the catalog.

For an external reference, the skill pins the exact commit and writes a catalog-ref.json with no SKILL.md in that folder. For a local skill, it creates SKILL.md with frontmatter, optional examples, references and assets, and a body with When to use and When not to use sections.

gh search code --filename SKILL.md "<topic>" --limit 20 grep -i '<topic>' packages/skills-catalog/generated/catalog.json gh api repos/<owner>/<repo>/commits/<ref> --jq .sha

What safety points should you check?

This skill runs no scripts of its own, but it uses the network and directs the agent to run shell tools such as gh, grep and xc, and to fetch web pages and READMEs. Any scripts/ file in a catalog skill makes that skill audit-gated, so the SKILL.md advises avoiding scripts unless they are needed.

Do not use this on a fork or in an environment with live X or GitHub credentials without review. The agent will run commands and can open pull requests under those accounts. Keep company-private skills out of the shipped catalog; the SKILL.md says they belong in the company's own library through the Skills UI or API.

  • Review every command before the agent runs it, especially network calls.
  • Do not let the agent use X or GitHub credentials you have not reviewed.
  • Check the upstream license is compatible with redistribution before referencing an external skill.

What tips help you get good results?

Give the agent the exact source URL and say whether the procedure is already in the source or you want the agent to author it. The SKILL.md says that if the source is only an aspiration, the agent is authoring the procedure itself and should say so in the PR.

Default to the optional kind. Use bundled only when every company should get the skill by default, which the SKILL.md says needs explicit human or board direction.

  • Start with a FIND search so you do not duplicate an existing slug.
  • Read the example files, examples/external-reference.md and examples/new-local-skill.md, to see the two paths in practice.
  • Expect the later steps to need checking by a maintainer, because the shipped SKILL.md is incomplete.

What people use it for

  • Packaging a useful public SKILL.md from GitHub into the Paperclip catalog as a pinned external reference with attribution.
  • Turning a tweet or a shared idea into a new optional local skill with a worked example and a pull request.
  • Checking whether a topic already has a skill in the catalog, the repo or GitHub before anyone writes a duplicate.
  • Adding a procedure your team wants to review as a PR, rather than letting it live in someone's private prompt.

Frequently asked questions

Is the create-paperclip-bundled-skill skill free to use, and what is its licence?

The item's licence is MIT. The source does not state pricing, so check the repository before assuming anything about cost beyond the licence.

Do I need the Paperclip repository to use this skill?

Yes, for the full workflow. The steps assume a Paperclip checkout with packages/skills-catalog, and the skill is not meant for general-purpose use outside that layout and its build tooling.

How do I install create-paperclip-bundled-skill in Claude?

Download the ZIP and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. Alternatively, upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

Why does the agent say xc is not available?

The xc CLI is preinstalled and authenticated only in Paperclip engineering agent environments. Elsewhere, the skill says to delegate the X/Twitter fetch to a teammate with access, and you should not install or create credentials for it yourself.

Can I use this skill for company-private skills?

The SKILL.md says no. Company-private skills belong in that company's own library through the Skills UI or API, not in the shipped catalog.

How is this different from a general-purpose skill for writing agent skills?

This one assumes Paperclip's catalog layout, its manifest builder and its validation rules. A general-purpose skill-writing guide will not follow those conventions, so pick it only if you are contributing to Paperclip.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you maintain or contribute to the Paperclip skills catalog and want an agent to find, pin or author a skill and deliver it as a reviewed pull request. Pick something else if you want a general-purpose skill, need company-private skills, or cannot review commands run under live X or GitHub accounts, and note that the shipped SKILL.md is cut off, so the later steps are not fully verified.

Download the skill