{{skill.key}}. Any update to a Skill propagates automatically to every reference.
Use Cases
Modular agent capabilities
Modular agent capabilities
Build agents that invoke specialized Skills on demand: a support agent that draws on a “summarize thread” Skill only when the conversation warrants it, or a research agent that invokes a “extract key claims” Skill selectively.
Shared tone and brand voice
Shared tone and brand voice
Reusable domain expertise
Reusable domain expertise
Encode compliance rules, classification criteria, or structured output schemas as Skills. Any team building on top of the same agents or prompts shares the same versioned definition without duplication.
Using Skills with Agents
Open the Skills section in the agent configuration, click Skills, and search for the Skill to attach it. Once attached, the agent can invoke it when relevant, without the Skill being present in the instructions at all times. Use this for capabilities the agent should draw on selectively depending on the conversation, for example a “summarize thread” Skill invoked only when the user asks for a summary.
Creating a Skill
AI Studio
MCP
Choose a Project and Folder, select the button, and choose Skill.Fill in a key that will be used to reference the Skill and configure the content.

Managing Skills
List
MCP
list_skills and returns all skills with their keys and descriptions.Get
MCP
get_skill with the skill key and returns the full skill configuration.Update
AI Studio
MCP
Open the Skill in AI Studio, edit the content or settings, and save to publish the update. All references update automatically.
Delete
AI Studio
MCP
Open the Skill in AI Studio, open the context menu, and select Delete.
Static references with {{skill.key}}
Prompt Snippets are now Skills. The
{{snippet.key}} syntax is replaced by {{skill.key}}.{{skill.key}}, where key is the key you set when creating the Skill. The Skill content is injected at that position every time it is rendered, and any update to the Skill is automatically reflected in every reference.
This works in:
- Agent instructions: inject content that should always be present on every run, such as a company policy block or a standard output format. See Agent Studio for more on writing instructions.
- Prompts: compose prompts from reusable blocks instead of duplicating text.
- Jinja templates: combine
{{skill.key}}references with Jinja expressions to compose dynamic, structured prompts from reusable blocks deterministically.