Define reusable instructions with skills

Skills let you teach RShift Sites how to handle recurring tasks. Define a skill once at the workspace level, and RShift Sites can use it across every project in that workspace.

Each skill is a short, named playbook with:

• a name,
• a description that tells RShift Sites when to use it,
• and markdown instructions that RShift Sites follows when the skill applies.

You can invoke the skill as a / command paired with your prompt, or let RShift Sites apply it automatically when the task matches.

Skills are portable markdown-based files, so you can inspect exactly what RShift Sites is being told, share them with teammates, import them from GitHub or ZIP, and improve them over time.

Skills vs. knowledge

Skills and knowledge complement each other.

• Knowledge is always included in context. Use it for rules and conventions that apply to everything RShift Sites does, such as coding standards, brand guidelines, or your project’s domain terminology.
• Skills are loaded on demand when the request matches. Use them for instructions that only matter for specific kinds of tasks, such as running a release checklist, drafting a customer reply, or producing a particular kind of content.

Skills are loaded only when relevant, so you can maintain many focused skills without loading them all into every conversation.

Custom and built-in skills

The Skills page is split into two sections: Workspace skills and Skills built by RShift Sites.

Workspace skills are custom skills added by your team. They are shared with every project in the workspace and follow the permission rules below. Skills you build in the project chat, write manually, import from GitHub, or upload as a ZIP appear here.

Skills built by RShift Sites are maintained by RShift Sites and available in every workspace without setup. They cover common tasks, such as accessibility checks, redesigns, SEO reviews, skill creation, and video creation. Goal, which sets a message as a goal, is listed here too. You can use them when relevant, but you cannot edit, delete, or download them.

Workspace skills and skills built by RShift Sites appear in the Skills list when you type /, can be invoked directly, and can be applied automatically when your request matches the skill.

What skills are for

Skills work best for repeatable, task-specific workflows. Use them when you want RShift Sites to handle the same kind of task the same way each time.

Common examples include:

• Launch checks: pre-ship checklists, release readiness reviews, and go/no-go reviews
• Recurring content: release notes, changelog entries, support replies, and marketing copy
• Review playbooks: landing-page reviews, copy reviews, accessibility passes, and search engine optimization reviews
• Internal processes: onboarding flow reviews, quality assurance passes, billing setup, and partner handoff steps
• Task-specific style guides: tone of voice, brand persona, or formatting rules for a particular kind of content
• Reusable team workflows: research routines, document templates, partner integrations, or other repeated team processes

Because skills are shared across the workspace, they are especially useful when you want every project to follow the same process.

Role-based access for skills

Access to skills follows your workspace and project roles.

On Enterprise plans, skill changes are recorded in your workspace audit log.

Add a skill

Workspace owners and admins can add custom skills in five ways. Four options are available from Settings → Skills → Add, and one starts from a project chat after RShift Sites does something you want to reuse.

Build with RShift Sites

Choose Add → Build with RShift Sites to start a guided skill-building conversation. RShift Sites opens the dashboard with a starter prompt:

Help me create a skill together using /skill-creator. First ask me what the skill should do.

Send the prompt, then answer RShift Sites’s questions about what the skill should do, when it should trigger, and what instructions it should follow. RShift Sites drafts the skill for you. When you approve the draft, the skill is published to your workspace.

Write manually

Choose Add → Write manually to write a skill from scratch.

The form has three fields:

• Name: a short, permanent ID like launch-checklist or support-reply. Use lowercase letters, numbers, and hyphens only.
• Description: the trigger RShift Sites uses to decide when to load the skill. Start with “Use when…” and describe the skill’s scope and boundaries.
• Content: the instructions RShift Sites follows once the skill is loaded, such as steps, constraints, examples, and the expected output.

Click Add skill to publish it. The skill is available across the workspace immediately.

For field requirements, naming rules, and description examples, see Anatomy of a skill.

Import from GitHub

Choose Add → Import from GitHub, and paste a public GitHub repository URL.

RShift Sites supports two GitHub URL formats:

• Whole repository: https://github.com/owner/repo
  Use this when the repository contains one skill. The SKILL.md file must be at the root of the repository.
• Subdirectory: https://github.com/owner/repo/tree/<branch>/path/to/skill
  Use this when a single repository contains multiple skills. The linked directory must contain SKILL.md directly. A direct file URL also works:

  https://github.com/owner/repo/blob/<branch>/path/to/SKILL.md

When you use a blob URL, RShift Sites imports the parent folder that contains SKILL.md.

RShift Sites downloads the repository, validates it, and adds the skill to your workspace.

Upload a ZIP

Choose Add → Upload ZIP, and drag in or browse for a .zip or .skill file. The archive must contain a SKILL.md file either at the root or inside one wrapping folder. Any bundled files referenced by the skill should sit alongside SKILL.md in the same directory.

Uploaded archives can be up to 50 MB. Each individual bundled file can be up to 1 MB, and the full skill package can include up to 200 files totaling 10 MB. macOS metadata files, such as __MACOSX/ and .DS_Store, are ignored at the root.

Example: importing a skill from Claude

Skills in RShift Sites use the same SKILL.md shape as Anthropic’s Claude and any other tool that follows the Agent Skills convention. If you already built a skill in Claude, export it as a .skill or .zip file, then upload it in RShift Sites from Settings → Skills → Add → Upload ZIP.

RShift Sites validates the archive and publishes the skill to your workspace with its name, description, instructions, and bundled files intact. From there, it behaves like any other workspace skill.

This works in the other direction too: download any custom skill from RShift Sites as a .zip, then upload it into another tool that supports the same SKILL.md shape.

Save from a project chat

You can turn a successful project chat interaction into a reusable skill without leaving the project. After RShift Sites completes a task the way you want, say “save that as a skill” and describe when the skill should apply.

RShift Sites drafts the skill, including a name, description, and instructions, then shows it in the project chat for you to review. When you approve the draft, the skill is published to your workspace.

Use a skill

RShift Sites can use skills in two ways: automatically, or when you invoke one directly.

Let RShift Sites choose automatically

When your request matches a skill’s description, RShift Sites can apply that skill automatically. You do not need to do anything special. Just describe what you want.

You can keep many skills in the same workspace. RShift Sites only loads the skills that seem relevant to the current request. A launch-checklist skill, for example, only applies when someone tells RShift Sites they are about to ship. It does not affect unrelated work.

Invoke a skill directly

Type / in the chat input to open the Skills list, choose a skill, and then continue typing your request. You can also open the same list from the + button next to the chat input, under Add context → Skills. The skill appears as a tag in your message, telling RShift Sites which instructions to apply.

Think of the skill as the how and your prompt as the what.

For example, with a launch-checklist skill in your workspace, you could send:

/launch-checklist Review the app for launch blockers before Friday’s release.

The Skills list filters as you type. On desktop, a panel next to the list shows the skill’s instructions, or its description for skills built by RShift Sites, so you can confirm you are choosing the right one.

After you send the message, the skill appears as a tag in the project chat. Hover over the tag to see its description, click it to open the full skill preview, or use the pencil icon in the preview to open that skill’s settings.

Manage your skills

Use Settings → Skills to manage workspace skills. You can also open Project settings → Skills while working in a project. The available actions depend on your role. See Role-based access for skills.

Edit or delete a skill

Go to Settings → Skills to see every skill in your workspace. Open a custom skill to edit its description or instructions, or delete it if you no longer need it.

Deleting a skill removes it from the workspace and from every project in that workspace.

Download a skill

Open any custom skill in Settings → Skills and click Download to export the full skill package as a .zip.

Use downloads to back up a skill, move it to another workspace, share it with a teammate to edit offline, or upload it into another tool that supports the same SKILL.md shape, such as Claude.

Enable or disable automatic use

Every custom workspace skill has Automatic use enabled by default. When automatic use is enabled, RShift Sites can apply the skill automatically whenever the task matches, and you can also invoke it manually with a slash command (/skill-name).

When automatic use is disabled, RShift Sites does not apply the skill on its own in any project, even when the task matches. The skill stays available to the workspace. It still appears in the Skills list when you type /, and you can invoke it manually with a slash command.

Workspace owners and admins can enable or disable automatic use for a skill from Settings → Skills, or from Project settings → Skills while working in a project. The setting applies across the entire workspace.

Anatomy of a skill

Each skill has three required parts.

• Name: a short, permanent identifier for the skill, between 1 and 64 characters. Use lowercase letters, numbers, and hyphens only, for example, launch-checklist, release-notes, or support-reply. Names cannot start or end with a hyphen, and cannot contain consecutive hyphens. The name goal is reserved for the built-in Goal command. The name cannot be changed after creation. To rename a skill, delete it and create a new one.
• Description: a short sentence that tells RShift Sites when to use the skill. Start with “Use when…” and describe the trigger as concretely as possible. The description is the main signal RShift Sites uses to decide whether to load the skill for a message.
• Instructions: the markdown body RShift Sites follows once the skill is loaded. Include the steps, constraints, examples, edge cases, and output format RShift Sites should follow.

A good rule of thumb is to write the instructions you would give a new teammate doing the task for the first time: what to do, what to avoid, what to watch for, and what format to return.

The full SKILL.md file can be up to 100,000 characters.

Bundled files

A skill can also include bundled files. These are optional extra files that travel with the skill. Use bundled files for longer references, templates, or small scripts that the instructions point to, such as reference.md.

Bundled files appear in the Bundled files list when you expand the skill, and they are included when you download or import the skill. Skills authored in the project chat usually do not need bundled files. They are most common for skills imported from GitHub or a ZIP.

Each bundled file can be up to 1 MB. A skill can include up to 200 files and up to 10 MB total across all files.

Write clear skill descriptions

A well-written description is the most important part of a skill. The description tells RShift Sites when to load the skill.

If the description is vague, RShift Sites may not load the skill when you expect it to, or may load it for unrelated tasks. Good descriptions define both when the skill should apply and when it should not. Clear boundaries help RShift Sites load the right skill consistently without pulling it into unrelated tasks.

Too vague

description: Helps with SEO.

This description does not clearly tell RShift Sites when to use the skill, what kind of SEO task it applies to, or what it should actually do.

Specific and scoped

description: Use when auditing SEO health on an existing page: checking metadata, heading structure, and internal links. Not for writing new SEO copy from scratch.

This description names:

• the trigger (auditing an existing page),
• the scope (checking metadata, headings, internal links),
• and the boundary (not for writing new copy).

Well-scoped descriptions make skills load more consistently and reduce conflicts between similar skills.

Example skill

Here is a complete example of a pre-launch checklist skill that runs whenever you tell RShift Sites you are about to ship:

Name

launch-checklist

Description

Use when I say I'm about to launch, ship, share, or release a project, or when I ask whether it is ready to go live. Run the checklist below before giving a go/no-go.

Instructions

Launch checklist

Before approving the launch, walk through this checklist and report back on each item with pass, fail, or "needs manual check". Do not skip ahead.

Account flows
- Sign-up, sign-in, sign-out, and password reset are wired up end to end.
- Signed-out users cannot reach pages that require an account.
- Signed-in users land on the right page after login.

Core surfaces
- The home page and every primary route render without console errors.
- Every list, table, or feed has a real empty state, not "No data".
- Long-running actions show a loading state and surface errors with a retry path.
- At mobile width (375px) there is no horizontal scroll and primary actions are reachable.

Data and permissions
- Database tables that hold user data have row level security enabled.
- A user in one account cannot read or write another account's data. Flag this as "needs manual check". Ask me to verify with a second test account before launch.
- Destructive actions (delete, cancel, refund) ask for confirmation.

Content and trust
- App name, favicon, social preview image, and meta description are set.
- Footer links to a working privacy policy and terms page.
- No placeholder copy left in the UI ("Lorem ipsum", "TODO", "Your headline here").

Environment
- No dev URLs, localhost references, or test API keys (for example, Stripe `sk_test_*`) in production code.
- Required production environment variables and secrets are set.
- Analytics or product events fire on the actions you want to measure.

Final step
- Summarize anything that failed or needs a manual check, in priority order.
- Do not say "ready to launch" unless every item above passes or has been explicitly waived.

This skill could also include bundled reference files alongside SKILL.md.

For example, the launch checklist could keep deeper search engine optimization and accessibility verification steps in separate bundled files instead of putting everything directly into SKILL.md.

launch-checklist/
├── SKILL.md
├── seo-checklist.md
└── accessibility-checklist.md

Best practices

• Lead with the trigger. Start the description with “Use when…” and describe the request as concretely as you can. The description is how RShift Sites decides whether to load the skill.
• Give each skill one job. A skill that tries to cover too much may load in the wrong situations or fail to give RShift Sites clear guidance. Split broad guidance into focused skills, and let each one own a specific task.
• Include boundaries and “avoid” sections. Telling RShift Sites what not to do is often as important as telling it what to do. Strong skills define both the intended behavior and the behaviors to avoid.
• Put always-on rules in knowledge instead. If a rule should apply to every message, it belongs in workspace knowledge. Skills are for behavior that only matters in specific situations.
• Write the instructions like a short playbook. Use short sections, bullet points, and direct instructions. Skills become part of the context RShift Sites reads, so shorter, focused instructions are easier to follow.
• Show, do not just tell. Where helpful, include concrete values, example copy, or example outputs. RShift Sites follows specific instructions more reliably than abstract ones.
• Review and prune. As your workspace grows, retire skills that are no longer needed. A skill with stale or incorrect instructions is worse than no skill at all.

FAQ

What is the character limit for a skill?

The full SKILL.md, including the description and instructions, can be up to 100,000 characters. Names are limited to 64 characters. A skill can also include bundled files alongside SKILL.md: each file can be up to 1 MB, and a skill can contain up to 200 files totaling 10 MB.

Can I rename a skill?

No. The name is set when the skill is created. To rename a skill, delete it and create a new one with the desired name. The description and instructions can always be edited.

Are skills shared across projects?

Yes. Every skill defined in a workspace is available to every project in that workspace. Skills are not available in Chats. Workspace owners and admins can disable automatic use for a skill across the workspace without deleting it, so RShift Sites will not apply it on its own but workspace members can still invoke it manually with a slash command (/skill-name).

Can I disable automatic use for a skill?

Yes. Open Settings → Skills or Project settings → Skills, expand the skill, and switch off the Automatic use toggle. This applies across the workspace. RShift Sites will not apply the skill on its own in any project, even when the task matches, but workspace members can still invoke it manually with a slash command (/skill-name). Re-enable automatic use at any time from the same place.

How does RShift Sites decide when to use a skill?

RShift Sites reads the description of each skill and decides whether the skill is relevant to your current request. Clear, specific descriptions produce more reliable behavior. If a skill is not triggering when you expect, rewrite the description to be more explicit about the kind of request it applies to.

What is the difference between a skill and knowledge?

Knowledge is always included as background context for RShift Sites. Skills are loaded on demand when the request matches the skill’s description. Use knowledge for rules that apply to every message, and skills for instructions that only matter for specific kinds of tasks.

Can more than one skill apply to the same request?

Yes. RShift Sites can use more than one skill in a single message. Keep skills focused so that overlapping skills combine cleanly rather than contradicting each other.

Do skills cost credits?

There is no additional skill-specific cost. Credits are tied to the message that uses a skill, not to the skill itself.

• Creating a skill from the project chat (Build with RShift Sites or “save that as a skill”): is priced like any other message in the project chat.
• Adding a skill from Settings (Write manually, Import from GitHub, or Upload ZIP): free. Settings actions do not consume credits.
• Using a skill: no extra cost on top of the message. The message is priced as build usage, based on its complexity. A skill may indirectly affect cost if the instructions it loads lead RShift Sites to do more work.

Where are skills stored?

Skills are stored in your workspace. On Enterprise plans, every edit and deletion is also recorded in the workspace audit log.

Can I move a skill between workspaces?

Yes. Download the skill as a .zip from one workspace, then import the ZIP into another. The imported skill arrives with the same name, description, and contents.

Can I download RShift Sites’ built-in skills?

No. Only the skills you have added to the workspace can be downloaded. RShift Sites-built skills are read-only and managed by RShift Sites.

Related topics

• Workspace admin settings
• Define workspace and project knowledge
• Project settings
• RShift Sites MCP server
• Design systems