Skip to main content

Skills Overview

Skills are reusable instruction files that agents can invoke as commands. They define how an agent should perform a specific task — from transcribing audio to writing code reviews to generating reports.

What is a skill?​

A skill is a Markdown file with YAML frontmatter that describes:

  1. What the skill does (metadata in frontmatter)
  2. How to do it (instructions in the Markdown body)

When an agent receives a command that matches a skill, it reads the skill file and follows the instructions.

Example skill​

---
name: summarize
description: Summarize a document or conversation into key points
---

# Summarize

Read the provided content carefully and produce a concise summary.

## Output format

- Start with a one-sentence TL;DR
- Follow with 3–5 bullet points covering the key points
- End with any action items or decisions made

## Guidelines

- Be concise — aim for 20% of the original length
- Preserve important numbers, dates, and names
- Use plain language

How agents use skills​

In OpenClaw, skills can be invoked in two ways:

1. Native Commands​

Skills can be invoked as native commands using the / prefix. When an agent receives a message like /summarize, it:

  1. Looks up the summarize skill file
  2. Reads the instructions
  3. Executes the task according to those instructions

2. Natural Language​

Skills can also be invoked via natural language when the frontmatter description includes trigger keyword examples. For example, if a skill's description mentions keywords like "summarize", "create a summary", or "brief overview", the agent can match natural language requests like "Can you summarize this document?" to the appropriate skill.

Writing Effective Descriptions To enable natural language invocation, include common trigger

keywords and phrases in your skill's frontmatter description. See Creating Skills for best practices. :::

Skills can be invoked by users via chat (e.g. Telegram) or by other agents in multi-agent workflows.

Skills in the MosBot Dashboard​

The Skills page in the dashboard provides a visual browser for all skills across your OpenClaw installation. You can:

  • Browse shared and agent-specific skills
  • View skill content with rendered Markdown
  • Create and edit skills directly in the browser
  • Search skills by name or description

Types of skills​

MosBot OS supports two types of skills:

TypeLocationAvailable to
Shared skills/skills/All agents
Agent-specific skills/workspace-<id>/skills/That agent only

See Shared vs Agent-Specific Skills for details.

To enable OpenClaw agents to work effectively with MosBot OS, we highly recommend implementing the following skills:

Essential Task Management Skills​

For Project Manager agents:

  • task_writing — Create well-structured, actionable tasks from descriptions, meetings, or PRD requirements. Essential for breaking down work into manageable tasks that can be tracked in MosBot OS.
  • create_prd — Generate comprehensive Product Requirements Documents that serve as the foundation for task creation and agent orchestration.

For specialized agents (Developers, Designers, QA, etc.):

  • task_pickup — Help agents select the next task based on priority, dependencies, skills, and workload. Enables effective task orchestration and workload management across agents.
  • memory_flush — Clear the agent's working memory and context for a clean slate. Highly recommended to prevent context pollution between unrelated tasks and maintain clean agent state when switching between different projects or conversations.
  • audio_transcribe — Transcribe audio files (voice messages, recordings) into text. Essential for processing voice inputs from communication channels like Telegram and converting them into actionable text that agents can work with.

Why These Skills Matter​

These recommended skills enable:

  • Task Orchestration: Agents can create, assign, and pick up tasks seamlessly within the MosBot OS ecosystem
  • Context Management: Agents can maintain clean working memory, preventing confusion from accumulated context
  • Multi-Modal Input: Agents can process voice messages and convert them to text for further processing
  • Workflow Integration: Skills bridge the gap between OpenClaw agents and MosBot OS task management features

See the Skill Reference for detailed documentation on these and other available skills.

Next steps​