Skip to main content

Skill Development

Overview​

Skill Development lets you create new tasks and specialties (skills) for the AI assistant. You develop them while chatting with the AI, and deploy them to production workers when they are ready.

Go to the skill development screen from the gear icon in the header → "Workers & Skills" → the "Skills" tab at the top.

Note: The skill development screen is desktop-only (mobile devices are not supported). Depending on your environment's configuration, the skill feature itself may also be disabled.

Skill List​

Skills are managed in packages. Each card shows the following information.

  • Package name and description
  • "Deploys to": the target worker for production deployment
  • Creation date and owner
  • "Shared" / "Read-only" badges (shared packages)

Setting Up a Dev Worker​

Skill development requires a dev worker isolated from production workers. If you click "Create New" before setting one up, you are guided to the setup page.

  1. On the "Dev Worker Setup" page, enter the worker's slug, display name, and description
  2. Click "Register"

Creating a Skill Package​

  1. Click "Create New" in the skill list
  2. Enter a name (alphanumeric characters and hyphens) and a description
  3. Choose a creation method:
    • "From Template"
    • "Duplicate Existing": select the package to duplicate
    • "Generate with AI": enter your goal and a draft is generated automatically
  4. Optionally specify the target worker (the default for production deployment) and a Git repository, then click "Create"

Development Screen (IDE) Layout​

AreaDescription
Top left "Skill Development"List of files in the package, grouped by category: Skills / Commands / System Prompt / Other Files
Bottom left "Env Vars"Environment variables for test runs (see below)
Center "Skill-building chat"Develop skills by conversing with the dev worker
Right (when a file is open)Code editor

The top of the screen shows the package name, the "Deploys to" worker, and the dev worker's status ("Dev worker: ◯◯ running").

Developing via Chat​

Use the center chat to build skills and commands by giving instructions to the dev worker.

  • Testing commands: from the "+" menu in the input area ("Test a command"), select and run a command from this package
  • File attachments: supports "Upload file", drag and drop, and pasting images from the clipboard
  • Questions to the user: a question modal appears when the worker asks for a decision
  • Restarting the session: "Start new session" resets the conversation history and the AI's context (files, skills, commands, and environment variables you have already created are not affected)

Editing Files​

  • Click a file to open it in an editor tab. Tabs with unsaved changes show a "●"
  • Press Ctrl+S / Cmd+S to save the file being edited. To save multiple files at once, use "Save All" in the header
  • Attempting to close an unsaved tab shows a confirmation dialog ("Save & Close / Discard Changes / Cancel")
  • For Markdown / HTML files, switch the view with "Code" / "Preview" at the right of the tab bar
  • If a file is updated on the worker side, a "Load Latest" bar appears

Environment Variables (Test-Only)​

Use the "Env Vars" panel at the bottom left to manage environment variables that are used only for test runs on this screen.

  • Register a key from "Add" and choose the value source:
    • "Use the value set in the test environment": uses the value entered on this screen
    • "Use the user's personal setting": uses the value each user registered in their personal settings (only the variable name is shared)
  • Turn on "Secret" to encrypt the value

Note: Values set here are not carried over when you deploy to production workers. Set production values on each worker.

Sharing​

The package owner can share the package with groups from "Share".

  • For each group, use the toggles to set "Edit" (Editable / Read-only) and "Approve" (whether the group can approve production deployment requests)

Deploying to Production​

Deploying a developed skill to production follows a request → approval flow.

  1. From "Create Request" in the header, select the files to deploy and submit a request
  2. A member with approval permission reviews it in "Requests"; once approved, it is deployed to the production workers
  3. If something goes wrong, an approver can revert to a previous state with "Rollback"

System administrators can also deploy directly with the "Deploy" button, but only for packages that are not shared (for emergencies).