Skill Management
Prerequisites: Before reading this guide, we recommend familiarizing yourself with the Skill configuration section in Bot Management.
A Skill is a reusable Markdown instruction package that is injected into a Bot’s system prompt, giving the Bot specialized capabilities for a specific domain.
Overview
Skills address the following problem: the system prompt carries a Bot’s general behavioral instructions, but different scenarios require different specialized capabilities. Packaging specialized instructions as Skills allows you to enable them on demand, preventing the system prompt from becoming bloated.
The essence of a Skill is simple — a piece of Markdown text injected at the end of the Bot’s system prompt. But through a standardized format and management mechanism, Skills can be reused across multiple Bots.
Skill Format
Each Skill is a .zip archive containing a SKILL.md file and optional attachments.
SKILL.md Structure
---
name: code-reviewer
display_name: Code Review Expert
version: 1.0.0
allowed_tools:
- github_*
- web_search
---
## Role Definition
You are a code review expert, specializing in...
## Review Process
1. Fetch the PR's changes
2. Analyze code quality
3. ...
## Notes
- Always watch for security vulnerabilities
- ...YAML Frontmatter Fields
| Field | Description | Required |
|---|---|---|
name | Skill identifier (globally unique) | Yes |
display_name | Display name | Yes |
version | Version number | No |
allowed_tools | List of tools required by this Skill (glob patterns) | No |
Body
The Markdown body below the YAML frontmatter is the instruction content injected into the Bot’s system prompt. Write it the same way you would write a system prompt — tell the Bot how to act in specific scenarios.
Attachments
A Skill package can include attachment files (such as reference documents, template files, etc.). Attachments are stored on the platform and can be referenced within the Skill instructions.
Uploading Skills
- Click Skills in the left navigation bar
- Click Upload Skill
- Select a
.zipfile (containingSKILL.md) - Once uploaded successfully, the Skill appears in the Skill list
Skill List
The Skill list displays the following for each Skill:
- Name and display name
- Version
- Type (instructions only / includes tools)
- Tool count (number of tools in
allowed_tools) - Attachment count
Viewing Skill Details
Click a Skill name to view its details, including:
- Instruction content
- Allowed tools list
- Attachment list
- Metadata (version, license, etc.)
Enabling Skills for a Bot
- Navigate to the Bot detail page and select the Settings tab
- In the Agent Capabilities section, find the Skills configuration
- Select the Skills you want to enable from the dropdown
- You can enable multiple Skills simultaneously
- Save
Once enabled, the Skill’s instruction content is automatically appended to the end of the Bot’s system prompt, presented under the ## Skills section.
Skill Scopes
Skills are divided into two scopes:
| Scope | Description |
|---|---|
| System | Built-in Skills provided by the platform, available to all workspaces |
| Workspace | Skills uploaded to the current workspace, available only within this workspace |
Common Use Cases
Code Review Skill
Contains instructions for the code review process, focus areas, and output format. Bind it to a code review Bot so the Bot follows a standardized review process.
Operations Inspection Skill
Contains a checklist of inspection items, anomaly detection criteria, and report format. Bind it to an operations Bot to ensure standardized inspection procedures.
Report Generation Skill
Contains report templates, data source specifications, and formatting requirements. Bind it to a reporting Bot to generate consistently formatted reports.
FAQ
Skill upload failed
- Confirm that the
.zipfile contains aSKILL.mdfile - Check that the YAML frontmatter format is correct
- Confirm that the
namefield is globally unique
Bot behavior hasn’t changed after enabling a Skill
- Confirm that the Skill has been saved in the Bot settings
- Skill instructions are injected via the system prompt; effectiveness depends on how well the LLM follows the instructions
- Check whether the Skill instructions are sufficiently clear and specific