Skill System
What It Is
A Skill is a dynamic context management and capability extension mechanism for Agents: it packages a domain's methodology, rules, and tool combinations into a reusable capability package that the Agent loads on demand — avoiding a bloated context from stuffing everything into the System Prompt, and improving response quality and execution efficiency.
In one sentence: the System Prompt is the Agent's personality; Skills are the professional skills the Agent can pick up whenever needed.
Three Ways to Create
| Method | When to Use |
|---|---|
| Create and edit manually (create an empty Skill) | Build your own methodology from scratch |
| Upload a Skill archive | Import your team's existing Skill assets |
| Search GitHub Skills online and import | Reuse mature community Skills |
The creation flow has two steps: ① upload/search/create empty → ② name and describe → publish.
Step 1: upload is the default, supporting .zip or .skill files (must contain a SKILL.md); the "Other creation methods" section below lets you switch to Search Skill or Create Empty Skill:
Choosing "Search Skill" lets you search community Skills on GitHub online, retrieving by keyword (e.g. the superpowers series); click to import:
Step 2: name and describe — the system automatically extracts the name and description from the uploaded/imported files, which you can edit manually; confirm and click "Publish":
- Naming rules: lowercase letters, digits, and hyphens only, up to 64 characters
- Description: describe the functionality and applicable scenarios precisely (so the model can judge when to invoke it); line breaks are not supported
- Skills imported from GitHub automatically record their source URL; you can also mount tools to the Skill at this step
Skill Details & Editing
After publishing you land on the Skill detail page:
- User-invocable: a switch; when on, users can trigger the Skill directly by typing
/in a conversation - Tools: view/add the tools mounted to the Skill; tools already added are preserved when the Skill is updated
- File browser: view the list of files inside the Skill, with support for creating files/folders and adding files to supplement assets
- Content preview: select a file (such as
SKILL.md) to preview the rendered result; click "Edit" to enter edit mode
Edit mode provides Editor/Preview tabs; the top of SKILL.md is frontmatter (name, description), and the body is the Skill's methodology content — click "Save File" to apply changes:
Three Ways to Use
Fixed Skills (Agentic Chat / Chatflow / Workflow)
Mount the Skill to the Agent, which then automatically decides whether to invoke it based on the task context and executes the relevant capability at the right stage.
- Typical scenario: earnings report analysis — the Agent automatically loads the earnings analysis Skill when it receives an earnings-related question
Two mounting points:
① Agentic Chat: in the "Skills" section of the Agent configuration page (below the knowledge base), click "+ Add Skill" to mount:
② Agent node in Chatflow / Workflow: in the "Skills" section of the node editing panel, click + to add (alongside the tool list, with the same support for variable injection):
Both places share the same selection dialog: it lists your custom Skills and official built-in Skills with an "Official" tag, and you can also create a Skill directly from the dialog:
Dynamic Skills (Agent node in Chatflow)
Instead of being hard-coded at configuration time, the Skill is passed in dynamically at runtime by the frontend via the Start node.
- Typical scenario: strategy analysis — when the user manually selects a strategy, the frontend passes in the corresponding Skill, and the Agent invokes that Skill during execution to complete the task
Manual Trigger in Conversation (User-Invocable)
With "User-invocable" enabled on the Skill detail page, the user can type / in a conversation with the Agent to bring up the Skill selection popover, which shows the Skill name, description, and Skill tags; selecting one triggers that Skill directly in the current turn:
- Difference from the first two methods: those let the Agent/flow decide when to use a Skill, while this one is explicitly specified by the user — suited to cases where "the user knows exactly which specialized task to run"
- How clearly the Skill's description is written directly affects how efficiently users can choose in the popover; see naming and description guidelines in Step 2 of the creation flow above
After selecting a Skill, you can append specific instructions and send them together (e.g. /superpowers-brainstorm help me list possible analysis angles for Google): the Skill appears as a tag in the message bubble, and the Agent executes and outputs following the methodology/template structure defined by that Skill — in the screenshot below, the answer strictly follows the Goal / Constraints / Known context / Risks framework defined in the Skill:
Official Built-in Skills
The platform provides high-quality official Skills, available on Pro and above plans.
Notes
- Quantity scales with plan: custom Skill limits are Starter 3 / Pro 20 / Premium 100; delete or upgrade when over the limit — see Workspace & Plans
- Skill content follows the same principles as prompts: conclusions first, actionable rules, positive and negative examples attached; see Prompt Writing Tips
- Before using community Skills imported from GitHub in customer-facing Agents, review their content yourself against the Compliance Requirements
Related Reading
- Agent Node Documentation — where Skills are mounted