Skill

Desktop only

Skills require the Dessix desktop app to be running and are only available on desktop.

The Actions and Scenes you create in Dessix are more than prompts — they can serve as skills that AI clients (like Cursor, Claude Code, and OpenClaw) call directly.

A single MCP tool performs one operation. A skill chains multiple operations into a complete workflow, letting AI drive a task within the boundaries you define.

How skills relate to MCP

Layer	What it does	Example
MCP tool	Single read or write operation	dessix_search_blocks, dessix_create_block
Skill	Multi-step composition toward a clear goal	Distill a discussion into a reusable Action

The MCP page covers individual tools. This page focuses on combining them to achieve more.

Enable skills on different platforms

Dessix provides a skill connector (@dessix/skill) that lets AI clients call the Dessix desktop MCP bridge directly via a local socket — no extra JSON-RPC middleware required.

Choose the installation method for your platform.

Prerequisites

• Node.js v18+
• Dessix desktop app running (the bridge is served by the Electron process)
• (Optional) Set the DESSIX_MCP_BRIDGE_ENDPOINT environment variable to override the default socket path

Cursor / Claude Code

Both Cursor and Claude Code load Dessix skills from a skill folder.

npx (recommended)

npx skills add dessix-skill

git clone

git clone https://github.com/DessixIO/skill.git
cd skill

Manual copy

# Copy SKILL.md, scripts/, and references/ into your project directory

After installation, verify the connection:

node scripts/dessix-bridge.mjs health

OpenClaw

OpenClaw manages and distributes skills through the ClawHub skill registry.

Install OpenClaw

macOS / Linux

curl -fsSL https://openclaw.ai/install.sh | bash

Windows (PowerShell)

iwr -useb https://openclaw.ai/install.ps1 | iex

npm

npm install -g openclaw@latest
openclaw onboard --install-daemon

Install the Dessix skill

ClawHub (recommended)

npm i -g clawhub
clawhub install dessix-skill

Manual install to workspace

git clone https://github.com/DessixIO/skill.git ./skills/dessix-skill

Manual install globally

git clone https://github.com/DessixIO/skill.git ~/.openclaw/skills/dessix-skill

OpenClaw skill loading precedence: ./skills (workspace) > ~/.openclaw/skills (global) > bundled skills. Start a new session after installation for the skill to take effect.

Three typical workflows

Skill discovery

Goal: See what reusable Actions and Scenes exist in your current workspace.

Step	What you say	What AI does
1	Show me the current workspace	Retrieves workspace info
2	List all Actions and Scenes	Searches blocks by type
3	Summarize each skill for me	Reads each skill's content and outputs a catalog

Useful when onboarding to a new project or reviewing existing knowledge.

Distill a discussion into a skill

Goal: Turn insights from a Topic discussion into a new, reusable Action or Scene.

Step	What you say	What AI does
1	Extract the context from my "Growth Strategy" topic	Retrieves the full conversation history
2	Find related notes to supplement it	Searches by meaning for supporting content
3	Create a "Growth Analysis" action based on all of this	Drafts and creates the Action

Insights no longer stay buried in conversations — they become repeatable AI behaviors.

Skill maintenance

Goal: Review and refresh existing Actions and Scenes to keep them accurate.

Step	What you say	What AI does
1	Find all Actions related to "weekly report"	Searches for matching Action blocks
2	Check if this Action's content is still valid	Reads and analyzes the current content
3	Change the output format to a Markdown table	Updates the Action content

Regular maintenance reduces drift between what a skill says and what you actually need.

Things to keep in mind

• Start with read-only workflows (skill discovery) to confirm everything works before writing.
• Before creating or editing an Action or Scene, confirm the target workspace and content scope.
• The skill connector connects to the Dessix desktop app directly — no separate MCP configuration required.