~/docs/skill-creator.md▼
Frontmatter
{
"title": "Agent Skills & Skill Creator",
"created": "1970-01-01T00:00:00Z",
"updated": "1970-01-01T00:00:00Z",
"slug": "skill-creator"
}Agent Skills & Skill Creator
Welcome to the ShadowClaw Agent Skills guide! Skills are workspace-local instruction packages and automated tool pipelines that give your AI assistant specialized capabilities.
Whether you want to teach the assistant a specialized coding workflow, build silent background utility tools, or trigger automated tasks with slash commands, Agent Skills make it simple and declarative.
🌟 Quick Start: What is an Agent Skill?
An Agent Skill is a directory containing a SKILL.md file located inside your workspace's .agents/skills/ folder:
.agents/skills/
└── main/
├── skill-creator/
│ ├── SKILL.md
│ └── references/
│ ├── frontmatter-spec.md
│ ├── declarative-pipelines.md
│ └── declarative-tools.md
└── toast-random-number/
└── SKILL.md
Each skill includes:
SKILL.md: Standard frontmatter metadata (name, description, slash command triggers) and step-by-step instructions.references/Directory: Detailed reference guides and technical documentation co-located with the skill.- Declarative Pipelines: Optional deterministic tool workflows (
execution.type: "tools") that execute instantly without calling LLM prompts.
🚀 Creating Skills with /skill-creator
ShadowClaw comes with an official bundled Skill Creator. You can trigger it anytime by typing /skill-creator in the chat thread or asking the AI assistant to create a skill for you.
Recommended Skill Authoring Workflow
- Define the Scope & Trigger: Decide on a clear skill name (e.g.
toast-random-number) and write a concise description of what it does. - Set Up Workspace Folders:
- Skill files belong in
.agents/skills/main/<skill-name>/SKILL.md. - Custom declarative tools belong in
.agents/tools/main/<tool-name>.json.
- Skill files belong in
- Write
SKILL.md: Configure standard frontmatter attributes and instructions. - Add Technical References: Store detailed reference materials under the skill's
references/directory.
📚 Skill Reference Documentation
For deep technical details, explore the co-located reference guides bundled with the official skill-creator skill:
📋 Frontmatter Specification
Complete schema rules forname,description,user-invocable,disable-model-invocation,argument-hint, andmetadata.allowed-tools.⚡ Declarative Execution Pipelines
How to write zero-LLM tool chains usingexecution.type: "tools",$pipeoutput piping,suppressToast, andsuppressOutput.🛠️ Declarative Tools Reference
How to author custom JSON tool definitions in.agents/tools/with sandboxed JavaScript evaluation and WebMCP model context integration.
💡 Example: A Two-Step Declarative Skill
Here is a complete, real-world example of a declarative skill that generates a random number and displays it in a toast notification silently:
---
name: toast-random-number
description: Generate a random integer from 1 to 1000000 and show it in a ShadowClaw toast notification.
user-invocable: true
metadata:
allowed-tools: javascript show_toast generate_random_number
execution:
type: tools
suppressToast: true
suppressOutput: true
tools:
- name: javascript
input:
code: Math.floor(Math.random() * 1000000) + 1
- name: show_toast
input:
title: Random Number
message:
$pipe: prev
---
# Toast Random Number
This skill generates a random integer and displays it instantly in a toast notification.
When you type /toast-random-number, ShadowClaw executes the two steps directly without delay!