How Skills Work
Skills are reusable workflows for AI. They hold instructions, scripts, and assets that should be applied consistently.
Use them for recurring work such as pull request reviews, support replies, weekly reports, candidate screening, pipeline reviews, or policy checks.
What a skill contains
Section titled “What a skill contains”| File or field | Purpose |
|---|---|
SKILL.md | The main Markdown instructions the agent reads |
| Scripts | Optional executable files that support the workflow |
| Assets | Reference files, examples, or templates |
| Handle | The slash-command-style identifier people can invoke |
| Description | Explains when the skill should be used |
Browse the public skills library
Section titled “Browse the public skills library”The Type Skills Library is a public catalog of reviewed skills. Search by workflow, filter by role or industry, and open a listing to inspect the included instructions, scripts, and examples.
Download a complete package for Claude Code or Codex and follow the included setup instructions. Skills with requirements and skillsets include the linked skill folders, with shared requirements included once. Each ordinary skill keeps its own SKILL.md, scripts, and references. Or choose Add to Type to add the skills to your current workspace. If you need to sign in or finish workspace setup, installation resumes automatically afterward. If that workspace already has a skill with the same name, Type opens the existing skill instead of creating a duplicate.
The public catalog is separate from your Space’s Library, where your team manages its own skills. It shows skills published by the Type team through the internal library tool. Drafts stay private, and publishing changes does not require a website rebuild. Imported and downloaded copies do not automatically update.
Each listing includes a readable title and short summary, a cover image or category illustration, and optional author details, benefits, and requirements. Sample output is stored with the listing and displayed inline on the public page. The Type team can write it in Markdown and preview it in the internal library tool, or ask an agent in the Type workspace to update it with update_library_skill_sample_output. New publications require sample output, an example report link, or a screenshot; drafts may omit them. The internal tool also previews the public card and page. Listing edits do not change the instructions inside the skill package.
In the internal tool, Choose files accepts individual Markdown (.md or .markdown), Word (.docx), and Pages (.pages) documents. A single document becomes SKILL.md, with a name added from the filename when needed. Choose folder imports a complete package with SKILL.md at its root. Word and Pages references are converted to .md locally before saving. Open a filename in the upload preview to check the text and conversion notes; complex layouts and embedded images may need manual review. Text files are limited to 256 KB each; document uploads can be up to 20 MiB before conversion. Add cover and example images through the listing’s image fields.
Uploaded images can be repositioned by dragging or with the position sliders. Choose Fit image to show the whole image, or Fill frame and adjust the zoom to crop it. The saved framing appears in both the preview and the public listing.
Add skills to a Space
Section titled “Add skills to a Space”Skills live under Library → Skills in a Space.
You can:
- Create a skill from the Library.
- Ask AI in the Space to create one.
- Import one from GitHub.
- Add an existing organization skill.
- Confirm a skill proposed by a Space template.
Templates only propose skills. Review the setup plan before creating or adding them, and reuse a good organization skill instead of making a duplicate.
Invoke a skill
Section titled “Invoke a skill”Select a skill from the composer or use its slash-command-style handle. The selected skill guides the AI response.
AI can also use skill descriptions to recognize when a workflow applies.
Skillsets
Section titled “Skillsets”A skillset is a curated collection of ordinary skills. For example, a revenue skillset could include Revenue Definitions, Revenue Reporting, Revenue Reconciliation, and Daily Revenue Check. Adding or downloading the collection includes its skills and their requirements once. Adding it does not execute those skills.
The public library shows Required skills, Included skills, and Used by relationships. These help you discover and package related skills; they do not create a new execution format or automatically invoke one skill from another. The instructions inside each SKILL.md must explain what to read and in what order. Public downloads work as ordinary skill folders outside Type.
In the internal library editor, choose Skill or Skillset under Listing type. Select Required skills for shared prerequisites and Included skills for a collection’s members. Linked listings must be published before you publish the dependent listing. A skillset’s uploaded SKILL.md describes the collection; it is not installed as another skill. The collection installs its selected ordinary skills instead.
Shared business definitions
Section titled “Shared business definitions”A definition skill is still an ordinary skill. Revenue Definitions, for example, explains how to establish and maintain revenue rules. The business’s actual answers belong in a shared reference file, such as finance/revenue_definitions.md in that business’s project.
Reporting skills should explicitly read that file before calculating results. If required rules are missing, they should use the definition skill to establish them with the appropriate owner. Reuse the business’s configured file instead of creating a separate copy for every report; keep different clients’ definitions separate.
Library links package the prerequisite skills. They do not configure business rules, grant data access, or guarantee that an agent loaded a reference file. Keep those requirements explicit in the portable skill instructions. Installed and downloaded copies do not automatically update.
Editing and reviewing changes
Section titled “Editing and reviewing changes”Creators can edit their own skills, admins can edit any skill, and other workspace members can propose changes.
Open Suggestions from the skill to find current and past suggestions. This stays available after review. Open a suggestion to read the author’s note, reviewer feedback, and outcome together, then review its changed files. The Changes view shows additions, removed passages, and replacements. Select a file in the file list to inspect its proposed content. Other live files are available for context and are labeled separately.
Markdown files stay formatted while you edit. Source shows the underlying Markdown when needed. Scripts open in a code editor. Switching files or views keeps your edits.
The review is read-only. The formatted comparison shows what will change; View source diff is available when you need the exact Markdown.
If the suggestion changes while you review it, reload and review the newer version before approving. Reviewers accept or deny the author’s suggestion; further edits belong in a new suggestion.
Edit several files, then preview and send them as one suggestion. A note for the reviewer is optional. After sending, View suggestion opens that exact suggestion.
If a file changes while you edit, your unsent draft stays available. Update my draft combines independent changes. When edits overlap, compare the latest version, resolve your draft, and confirm it before sending. If the file was removed, copy your draft or discard it. Older suggestions that lack an original comparison must be recreated before approval.
Reviewers read the changes, select another file when needed, then Accept suggestion or Deny suggestion. Denial can include an optional note. Both decisions are final. To propose further changes, open the current skill and create a new suggestion. Reviewing scripts does not execute them.
Approval updates open skill views automatically. If updates were missed while disconnected, Type refreshes when the connection returns or the page regains focus. Unsent edits are preserved for comparison.
Built-in skills
Section titled “Built-in skills”Type and its Claude or Codex runtimes provide built-in skills for common workflows, including presentations, documents, spreadsheets, apps, images, data visualization, and video.
These skills help AI choose the right tools, create native files such as .pptx, .docx, and .xlsx, validate the result, and attach finished work to the thread.
Image generation uses Type’s generate_image tool with gpt-image-2.5-flare. The Remotion skill creates React video compositions as Type Video artifacts and queues MP4 export through Type’s self-hosted Render Workflow.
Video artifacts use Type’s build configuration and installed dependencies. Include composition source and assets; do not include remotion.config.*, other executable build configuration, or node_modules. Compile-time macros and imports of host files are also blocked. Set video properties in the composition and export settings.
Skill hygiene
Section titled “Skill hygiene”Keep each skill narrow, explicit, and easy to reuse. A support reply skill, a code review skill, and a weekly marketing ideas skill should usually stay separate.
Never put secrets in a skill. Put credentials in connections and workflow guidance in skills.
Managed Claude runs disable skill lifecycle hooks and inline skill shell execution. Skills that need to run a script should instruct the agent to use Type’s managed execution tools.