Skills (Skill Library)

14. Skills (Skill Library)

A Skill is a reusable package of methodology and expertise: it tells the Agent how to think about a class of work, what to check, and how to deliver. It is not a tool grant, and it does not expand which tools the Agent may call or bypass product governance.

Lantide uses the common Agent Skills package format (including a standard SKILL.md). The built-in Agent and external Agents connected over MCP can share the same skill definition; what they can actually do is still decided by Agent state, tool activation, MCP access mode, Approve, and audit.

14.1 What a Skill is (and is not)

Is	Is not
Analysis / delivery methodology, checklists, quality rules, and reference resources	Extra API keys, database connections, or system privileges
On-demand guidance (progressive disclosure)	A way to obtain new tools or bypass Plan / Approve
A definition shareable with the built-in Agent and external Agents	A public marketplace or remote skill store

Common misconception: importing a skill ≠ authorizing the Agent to do more. A skill only changes guidance on how to work; whether work is allowed remains under the existing governance layers.

14.2 Open Skill Library

Click the Skill Library icon in the top-right of the main window.
The dialog lists your imported personal (User) skills.
For each skill you can toggle:
- Enabled: whether it is available to load in built-in chat
- Share with external agents: whether an external Agent connected over MCP may read the same definition
Other actions: Import Skill…, Open Skills Folder, Validate again, and remove (deletes only Lantide’s managed copy; your original import folder is unchanged).

Built-in skills do not appear in this list. Product-built Plan / statistics / HTML methodology is managed by the app; Skill Library only manages packages you import. Executable scripts inside skills are not supported yet; packages with unsupported content show Scripts unsupported.

14.3 Import a skill

Prepare a folder that follows the common format and contains a standard SKILL.md.
In Skill Library, click Import Skill… and select that folder.
After a successful import, use it in chat as /skill-name (see §14.4).

If a managed skill with the same name already exists, import fails with a collision; remove or rename first, then retry. Use Validate again to re-check whether the package is still valid.

14.4 Explicit load with `/` in chat

In the right-hand AI chat input, type / to open the slash picker listing skills that are currently Enabled and slash-visible:

Keep typing a name fragment to filter the list.
Click with the mouse, or use Up/Down then Enter.
When you send the message, the selected skill loads with that turn for the built-in Agent.

This is explicit load: you choose when to apply which methodology, instead of parking every skill in the system prompt.

Consumer	How the skill is used
Built-in Agent	Skills Enabled in Skill Library; load explicitly with `/` in chat
External Agent (MCP)	The same managed definition; turn on Share with external agents, and the external connection remains bound by §13 access mode / scope

The shared piece is the definition (how to work), not a raise of the external Agent’s privileges. Observe / Execute / Admin, Approve & Execute, Activity, and audit rules are unchanged.

14.6 Boundaries and limits

Cannot bypass governance: skill instructions cannot bypass MCP access mode, Plan approval (Approve & Execute), audit, or artifact contracts.
No marketplace: there is no public store, ratings, paid catalog, or remote registry; imports come from packages you pick on your machine.
Third-party skills are not guaranteed drop-in: skills from other people or tools may need front-matter, path, or content adjustments before they validate and run in Lantide.
Scripts not supported yet: packages with executable scripts do not gain execution capability; treat them as methodology docs and resources.
Remove affects the managed copy only: the original import source folder is not deleted or modified.