Create Skill

Run this skill when asked to create a new skill from a conversation or description of a task.

Step 1 — Gather requirements

Ask the user only these two questions:

1. What specific task or problem do you want this skill to solve?
2. What are the specific instructions and guidelines for this task?

Wait for the user's answers before proceeding.

Step 2 — Determine skill name and location

Derive a short, lowercase, hyphenated skill name from the task description (e.g. deploy-lambda, check-postgres-health).

Ask the user which AI agent they are using, then use the corresponding path:

| Agent | Skill path | |---|---| | Claude Code | .claude/skills/<skill-name>/SKILL.md | | GitHub Copilot | .github/skills/<skill-name>/SKILL.md | | Other | skills/<skill-name>/SKILL.md |

Create the directory if it does not exist:

bash
mkdir -p <path-to-skill-dir>

Step 3 — Write the SKILL.md file

Use the template below, filling in all <...> placeholders based on the user's answers.

markdown
---
name: <skill-name>
description: <One-liner description of what this skill does.>
---

# <Skill Title>

Run this skill when asked to <task statement>.

## Step 1 — <First step title>

<Instructions for the first step.>

## Step 2 — <Second step title>

<Instructions for the second step.>

...

## Rules

- <Rule 1>
- <Rule 2>

Guidelines for writing the skill

• Each step should have a clear title and concrete instructions or commands.
• Include bash commands in fenced code blocks where relevant.
• Rules section must cover safety constraints, error handling, and edge cases.
• Keep the skill focused — one skill, one workflow.
• Do not include secrets, credentials, or hardcoded environment-specific values.

Step 4 — Confirm with the user

After writing the file, show the user the generated SKILL.md content and the path where it was saved. Ask if any adjustments are needed.

Rules

• Always place the skill in the correct path for the user's AI agent (see Step 2 table).
• The skill name must be lowercase and hyphenated.
• Never include credentials, API tokens, or secrets in the generated skill.
• If the user's description is too vague to produce actionable steps, ask clarifying questions before writing the file.