/create-skill

Create a New Skill or Command

Guide the user through authoring a Claude Code extension.

Key Concepts

Commands (like this one): Explicit /slash invocations. Single .md file in .claude/commands/. The user types /command-name to run it.

Skills: Passive capabilities that activate automatically when relevant. Live in .claude/skills/<name>/SKILL.md. Triggered by hooks or pattern matching on user prompts — no explicit invocation needed.

Core Principles

1. Conciseness — The skill/command text is injected into the context window alongside user code and conversation. Every line competes for space. Cut ruthlessly. Target: commands < 150 lines, skills < 100 lines.
2. Degrees of freedom — Specify what to do and constraints, not how. Let the model choose the implementation. Bad: "Use os.path.join to construct the path." Good: "Construct the full path to the config file."
3. Progressive disclosure — Start with the minimal instruction set. Add detail only in sections the model reads on-demand (e.g., "If the user asks about X, then...").

---

Step 1 — Define the Extension

Ask the user:

1. What should it do? Get a one-paragraph description.
2. Skill or command?
   • Does the user invoke it explicitly? -> Command
   • Should it activate automatically in certain contexts? -> Skill
3. What tools does it need? (Read, Write, Edit, Bash, Glob, Grep, WebFetch, WebSearch, etc.)
4. What arguments does it accept? (for commands only)

Step 2 — Generate the File(s)

For a Command

Create a single .md file:

.claude/commands/<name>.md

Structure:

---
name: <kebab-case-name>
description: <one-line summary>
argument-hint: "<what the user provides>"
allowed-tools: <comma-separated tool list>
---

# <Title>

**Input:** $ARGUMENTS

## Step 1 — <First action>
...

## Step 2 — <Second action>
...

For a Skill

Create a directory with a SKILL.md:

.claude/skills/<name>/SKILL.md

Structure:

---
name: <kebab-case-name>
description: <one-line summary>
triggers:
  - <pattern or context that activates this skill>
---

# <Title>

## When to Activate
<Conditions under which this skill should engage>

## Behavior
<What to do when activated>

## Constraints
<Boundaries — what NOT to do>

Step 3 — Install Locally

Copy the generated file(s) to the appropriate location:

# Command
cp <generated>.md .claude/commands/

# Skill
mkdir -p .claude/skills/<name>
cp SKILL.md .claude/skills/<name>/

Verify the file is in place:

ls -la .claude/commands/<name>.md   # or
ls -la .claude/skills/<name>/SKILL.md

Step 4 — Test

Tell the user:

1. Start a new Claude Code session (skills and commands are loaded at session start)
2. For commands: type /<name> and verify it appears in autocomplete
3. For skills: describe a scenario that should trigger it and verify activation
4. Iterate — edit the .md file and restart to test changes

Step 5 — Share Back (Optional)

Offer to contribute the skill to the claude-eureka collection:

Want to share this with other researchers? Run /contribute-skill to package it and open a PR to the claude-eureka repo.

Checklist Before Finishing

• File is under 150 lines (commands) or 100 lines (skills)
• Frontmatter has all required fields
• Instructions use what/constraints style, not how style
• No redundant examples — one clear example per concept max
• Tool list in allowed-tools is minimal (don't grant tools you don't need)