Skills
Reuse task instructions and supporting resources with your chosen model or Agent.
A skill packages instructions for a repeatable task, such as reviewing a document or preparing a release summary. You can inspect it, add it to Yours, and activate it in a text conversation without changing the selected model or Agent.
Choose your task
Use skills in chat
Inspect, select, activate, and stop using a skill.
Upload and manage skills
Create packages, edit revisions, share, copy, export, and delete.
Skill package reference
SKILL.md format, supporting files, metadata, and limits.
Understand the lifecycle
- Discover or upload: inspect a shared package or upload your own. New uploads are private.
- Select: add it to Yours so it is available to chat. Selection alone does not load its instructions.
- Activate: pick it with
/in the composer, or let the assistant load it when the skill allows automatic activation. - Verify: check the Loaded row and revision in chat, then inspect the answer against the task.
- Stop or revise: stop using it in the conversation, or publish a new package revision. Existing conversations retain their pinned revision.
For example, upload the release-summary example, then send /skill:release-summary Summarize these changes: added CSV export; fixed duplicate notifications. Check that the skill loaded and that the response follows its instructions.
Skills, Agents, and tools
| Capability | What it supplies |
|---|---|
| Agent | A saved model, system prompt, and optional knowledge and tool configuration. |
| Skill | Reusable task instructions and supporting resources for the selected model or Agent. |
| MCP Server | Callable tools with their own permissions and credentials. |
A skill does not grant tool access, run bundled scripts, or install plugins. It can guide use of tools that are already available. Inspect packages before selecting them, especially shared instructions and supporting files.
Skills are supported in Portal text chat, not Deep Research, live Voicebot, or the OpenAI Responses and Anthropic-compatible API paths. A package-management API is not the same as skill activation in a model request.