skill-creator

name	skill-creator
description	Create, update, or delete skills. Use when the user asks to create a new skill, modify an existing skill, package scripts/assets into a skill, or discusses skill design and structure.

Skill Creator

Skill Structure

skills/{skill-name}/
├── SKILL.md              # Required: frontmatter + instructions
├── scripts/              # Optional: executable scripts
│   └── setup.sh
├── references/           # Optional: docs loaded on demand
└── assets/               # Optional: templates, config files

Create skills under skills/ in the working directory. The Skill tool auto-discovers them by name.

Lifecycle

1. Discovery — Every message, all skill names + descriptions appear in the system prompt
2. Loading — LLM calls Skill(name=..., action=load) to read SKILL.md
3. Tool loading — LLM immediately calls load_tools for all tools listed in the skill's "Required Tools" section
4. File listing — Skill(name=..., action=list_files) returns full paths of all files in the skill
5. Execution — LLM runs scripts via Shell tool using the paths from list_files

Creating a Skill

1. Discover relevant tools

Before writing SKILL.md, use search_tools to find tools the skill will need:

search_tools(query="send feishu message")  → finds feishu_send_message, etc.
search_tools(query="github pull request")  → finds mcp_github_create_pr, etc.

Include the discovered tool names in the skill body so the LLM knows which tools to load_tools after activating the skill.

2. Write SKILL.md

---
name: my-skill
description: What this skill does and WHEN to activate it. Be specific — this is the only trigger.
---

# My Skill

## Required Tools
After loading this skill, immediately call `load_tools` for these tools:
- feishu_send_message
- feishu_search_wiki

## Instructions
Step-by-step instructions for the LLM...

Critical: Every skill MUST include a "Required Tools" section listing tools to load. After Skill(action=load) returns, the LLM must immediately call load_tools for all listed tools before doing anything else.

3. Add scripts (optional)

#!/usr/bin/env bash
# scripts/setup.sh
set -euo pipefail
echo "Running setup with args: $@"

Make scripts executable: chmod +x scripts/*.sh

Reference scripts in SKILL.md with relative paths from the skill root:

Run setup:
`Shell` tool: `bash scripts/setup.sh <args>` (working directory: the skill root)

Or use `Skill(name=my-skill, action=list_files)` to get the absolute path,
then call `Shell` with the full path from any working directory.

4. Add references (optional)

Large docs or API specs go in references/. Load them with:

Skill(name=my-skill, action=load, file=references/api-spec.md)

Updating a Skill

1. Skill(name=..., action=load) — read current content
2. Edit tool — modify SKILL.md or other files
3. Skill(name=..., action=list_files) — verify file layout

Writing Guidelines

Frontmatter:

• name: lowercase with hyphens (e.g. pdf-editor)
• description: WHAT it does + WHEN to use it — this is the sole activation trigger

Body:

• Keep under 300 lines (auto-truncated beyond this)
• Imperative form, concise — only include what the LLM doesn't already know
• Relative paths for internal references (scripts/run.sh, not absolute paths)

Scripts:

• Shebangs: #!/usr/bin/env bash or #!/usr/bin/env python3
• Accept arguments via $@ or $1/$2 for flexibility
• Use set -euo pipefail in bash scripts