Skip to Content
Agentic WorkspaceAI Agent4. SkillsAdvanced skill settings

Advanced skill settings

By the end of this page you will know how to shape a skill beyond its tools — with a prompt overlay, role-based access, structured input/output, and per-tool voice fillers.

Prerequisites

  • A skill with a name, description, and (optionally) tools (Attach tools).
  • Admin or Developer role.

System prompt overlay

Every skill has a System Prompt / Goal editor. When the skill is active, this text is layered on top of the agent’s persona system prompt. Use it to give the skill task-specific instructions — its role, its objective, and any constraints.

Write the overlay

Type your instructions in the editor, for example: “You are a helpful assistant that specializes in… Your goal is to…”. The prompt allows up to 16,000 characters — a live counter warns when you exceed the limit.

Generate with AI Assist

Click AI Assist (or press Ctrl/Cmd + Alt + G in the editor) to draft or refine the prompt with help from the LLM.

ℹ️

Persona vs. overlay: the persona sets the agent’s base personality and tone across every skill; the skill overlay adds instructions that apply only while that skill is active.

Allowed roles (RBAC)

If your project defines organization roles, each skill shows an Allowed Roles control. Use it to restrict a skill to specific roles.

Choose which roles may use the skill

Open the roles popover and tick the org roles allowed to use this skill. The trigger shows “Unrestricted (all roles)” or a count of the roles you’ve selected, and selected roles appear as removable chips. Leave it empty to allow all users.

Allowed Roles control on a skill card with several org roles selected
💡

Role restrictions are enforced at run time against the caller’s roles. If a user’s roles don’t match, the skill won’t match for them — and the agent returns the fallback message. You can preview this with role impersonation in Test the agent.

Accepted input and output types

By default a skill accepts and produces plain text. You can declare additional content types for multi-modal or structured work.

Add input types

Under Accepted Input Types, add the MIME types the skill can accept — for example, images (PNG, JPEG, WebP, GIF), audio (WAV, MP3, WebM, M4A), PDF, Excel, CSV, or Word. text/plain is always included as a baseline.

Add a JSON input schema

Selecting application/json reveals an inline schema editor. Provide a JSON Schema to validate structured input.

Add output types

Under Accepted Output Types, choose Plain Text and/or JSON. Selecting JSON reveals an output schema editor.

Accepted Input Types and Output Types controls with a JSON schema editor open

Voice filler phrases

ℹ️

Voice fillers appear on a skill only when the agent’s Voice mode is enabled and the skill has tools attached.

Voice fillers are short phrases the agent speaks while a tool runs — so the caller hears something instead of silence. You set them per tool and per language.

Set filler phrases

For each tool, enter what the agent should say while it executes — optionally in multiple languages (Default/English plus Tamil, Hindi, Telugu, Kannada, Malayalam, Marathi, Bengali, Gujarati, and Punjabi). If no filler is set, the agent falls back to its persona’s default filler.

See Voice configuration for agent-wide voice settings.

Save your changes

Click Save in the header to persist these settings into a draft version.

Common issues

  • The Allowed Roles control isn’t shown — Your project has no organization roles defined yet. Add roles first.
  • A JSON schema is rejected — The schema must be valid JSON Schema. Check for syntax errors in the inline editor.
  • Voice fillers aren’t visible — Enable Voice mode for the agent and attach at least one tool to the skill.