# Skill Management

> Manage Kian's installed Skills, including installing a skill from a GitHub repository, listing installed skills, enabling or disabling visibility, and uninstalling non-builtin skills. Use this when the task is to change which Skills an agent can use.

- Skill: `sandai-org/skill-management` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add sandai-org/skill-management`
- Raw SKILL.md: https://api.skillmd.com/api/skills/sandai-org/skill-management/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: sandai-org (https://skillmd.com/u/sandai-org)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/sandai-org/skill-management

---


# Skill Management

Use this skill when the task is to manage Skills for Kian itself rather than to change project files.

## When to Use

- The user wants to install a Skill from a GitHub repository.
- The user wants to see which Skills are already installed.
- The user wants to enable, disable, or uninstall a Skill.
- The task changes the global Skill set available to agents.

## Read First

- Read [references/skill-files.md](references/skill-files.md) before making changes so the on-disk layout and install rules match the current implementation.
- Skills are global under `<GlobalWorkspaceRoot>/.kian`, not scoped to a single project.
- Builtin skills use `builtin://kian`; they can be enabled or disabled, but they must not be uninstalled.

## Workflow

1. Inspect the installed Skills first. Do not guess from memory.
2. For GitHub installs, normalize the repository URL to `https://github.com/<owner>/<repo>`.
3. Fetch the repository into a temporary location by cloning it or downloading the GitHub archive, then inspect that temporary copy.
4. Find candidate skill directories by recursively locating `SKILL.md`.
5. If the repository contains exactly one Skill and the user did not specify a path, install that Skill directly.
6. If the repository contains multiple Skills and the target is ambiguous, list the candidates and ask the user which one to install.
7. When creating or repairing a Skill directly on disk, always write the full installed `.skill.json` metadata from the reference schema. Do not write `config.json`, and do not use a minimal manifest with only `name`, `description`, or `version`.
8. Preserve existing `installedAt`, `mainAgentVisible`, and `projectAgentVisible` values when reinstalling the same Skill.
9. If the user asks to enable or disable a Skill only for the current agent, change only the flag for the current scope and preserve the other one.
10. If the user asks to enable or disable a Skill for all agents, update both visibility flags explicitly.
11. Never uninstall a builtin Skill.

## Scope Mapping

- In the main agent, the current-scope flag is `mainAgentVisible`.
- In a project agent, the current-scope flag is `projectAgentVisible`.
- If the user says only "enable" or "disable" without a scope, prefer changing the current-scope flag only.

## After Changes

- Skill changes are picked up by new Agent sessions.
- The current conversation keeps the Skills that were loaded when the session started.
- If the user wants the updated Skill set immediately, use the `NewSession` tool after the change.

