SKILL
A SKILL is a reusable capability unit. LanSphere comes with official skills built in, and also supports importing or creating custom skills — consolidating proven business capabilities into standard components that you attach to Agents for on-demand calling.
This article covers the skill concept, using the SKILL Management page, the Skill Store, the full import and creation flows, the three skill types, and how to attach skills to an Agent. It is for developers and admins.
What Is a Skill
In an Agent, the LLM understands intent and plans the steps, while skills make sure each specific kind of task gets done right and reliably:
- Reusable: Create once, share across many Agents — no reconfiguring the same capability in every app.
- Standardized: Inputs and outputs are explicitly defined with JSON Schema, with clear calling boundaries.
- Diverse sources: Use the built-in official skills, import external skill packages, or create your own based on code, Workflows, or MCP Services.
The difference from plugins and MCP Services: plugins and MCP Services answer “where capabilities come from”; skills answer “how capabilities get reused in a standardized way.” The three compose — a skill can use a plugin or an MCP tool as its execution vehicle.
The SKILL Management Page
Entry: left menu → “SKILL Management.” The page has two tabs: “System Skills” and “Custom Skills.”
System Skills
Shows the list of installed system skills, with fields:
- Name
- Type
- Version
- Description
- Installed at
You can uninstall system skills you no longer need; click the “Skill Store” entry to browse and install more skills.
Custom Skills
Shows all custom skills you imported or created, with edit, delete, and export actions in the list. The page offers two entries:
- Import Skill File: Import a ready-made skill package from your local machine.
- Create Skill: Orchestrate a new skill from scratch.
Skill Store
In the Skill Store, you can:
- Browse official and community-contributed skills.
- Search for a target skill by name.
- Click install — installed skills appear in the “System Skills” list, ready to attach and use.
You can also publish your own custom skills to the Skill Store for the whole platform — publishing requires Admin review before listing, and Admins also manage skill delisting.
Tip: Before publishing, make sure the skill description is clear and debugging has passed, so reviewers and later users understand what it is for.
Import Skill File
For when you have a ready-made skill package from elsewhere:
- In the “Custom Skills” tab, click “Import Skill File.”
- Pick a local file: only
.mdand.zipare supported, at most 1 file per import. - For a zip package, its root directory must contain a
SKILL.mdfile, or the import fails. - The skill name and description prefer the
nameanddescriptionfields inSKILL.md— check the file content before importing.
Tip: Only 1 file per import; for batch imports, do them one by one. A common cause of failure is a missing
SKILL.mdat the zip root — check the package structure and retry.
Create Skill
In the “Custom Skills” tab, click “Create Skill” and complete five steps:
- Basic info: Name required (up to 30 characters); description optional (up to 200 characters); icon can be set. Name it after the capability so it is easy to recognize and check in an Agent.
- Choose a type: Code Skill, Workflow Skill, or MCP Skill.
- Configure input / output parameters: Define the skill’s inputs and outputs with JSON Schema.
- Configure execution logic: Write a prompt, or pick a built-in tool (plugin, MCP, Custom Tool) as the execution vehicle.
- Debug and save: Debug a run to verify; at least one successful debug is required before saving; debug timeout is 30 seconds.
Note: Skill names are unique within a workspace — duplicates cannot be saved. The skill type cannot be changed after creation; create a new skill to switch types.
The Three Skill Types
| Type | Definition | Best for |
|---|---|---|
| Code Skill | Implements logic in a Python / JavaScript snippet with input/output schemas defined | Logic that needs precise, deterministic processing, like data computation and format conversion |
| Workflow Skill | Packages an existing Workflow as a skill | Reusing an already-orchestrated multi-step flow as a whole capability |
| MCP Skill | Wraps an MCP Service tool as a skill, inheriting its authentication and configuration | Reusing capabilities of already-connected external systems |
Choosing:
- Deterministic logic expressible in code → Code Skill first.
- A proven Workflow already exists → package it directly as a Workflow Skill — no duplicated building.
- The capability lives in an external system → connect the MCP Service first, then wrap it as an MCP Skill.
Attaching to an Agent
Skills take effect through Agents. Attach like this:
- Go to the target Agent’s “App Development → Tool Configuration.”
- Check the installed skills in the “Skills” category.
- After saving, the LLM decides when to call them based on user intent — no hardcoded call logic in the prompt.
You can also allow or forbid specific skills per Agent: the same skill can be attached to some Agents while disabled for others — fine-grained capability control.
Permissions
| Action | Viewer | Developer | Admin |
|---|---|---|---|
| Install / use system skills | — | ✓ | ✓ |
| Create / import / edit / delete custom skills | — | ✓ | ✓ |
| Uninstall system skills | — | ✓ | ✓ |
| Skill publish review, delisting | — | — | ✓ |
Tip: Insufficient permissions prompts “You don’t have permission to perform this action. Please contact an Admin.” For the full definitions and capability matrix of the four roles, see “Members & Permissions.”
Related Reading
- “Agent”: The full picture of Tool Configuration and Skill Orchestration.
- “Plugins & MCP”: How to connect an MCP Service as a skill’s execution vehicle.
- “Members & Permissions”: The role-permission system at large.

