Skip to content
OpenAgentSkill
Browsing is open · Catalog captured 2026-11-21 · Automatic updates and account services are paused.
Productivity

create-skill

Create one new skill end to end - asks about purpose, trigger phrases, tools and context, then writes .claude/skills/<name>/SKILL.md with valid frontmatter and registers it in CLAUDE.md. Use when user says "create a skill", "make a slash command for this", "turn this workflow into a skill", "new skill for X". Do NOT use for hooks, rules or MCP integrations (use /extend), and do NOT use to change an existing skill - edit that SKILL.md directly.

onurpolat05 · Static checks on file

Installation

Read the source and dependencies before running a command. Copying does not install or execute anything.

npx skills add onurpolat05/ALBA --skill create-skill

About this skill


name: create-skill description: Create one new skill end to end - asks about purpose, trigger phrases, tools and context, then writes .claude/skills//SKILL.md with valid frontmatter and registers it in CLAUDE.md. Use when user says "create a skill", "make a slash command for this", "turn this workflow into a skill", "new skill for X". Do NOT use for hooks, rules or MCP integrations (use /extend), and do NOT use to change an existing skill - edit that SKILL.md directly. context: inline effort: medium argument-hint: "[skill-name]" allowed-tools: [Read, Write, Edit, AskUserQuestion]

ALBA Skill Creator

Interactively create a new skill with proper frontmatter and structure.

Process

Step 1: Understand the Skill

Ask these questions (one at a time):

  1. "What should this skill do?" - Core purpose in one sentence
  2. "What exact phrases should trigger it?" - Push for the words the user actually types, not abstract categories. "research", "look into X", "find info on" beats "information gathering".
  3. "When should it NOT fire?" - The nearest existing skill it could be confused with. This answer becomes the Do NOT use for ... clause and is what stops two skills fighting over the same prompt.
  4. "What inputs does it need?" - What the user provides
  5. "What output should it produce?" - Format, structure
  6. "Does it need web access or just local files?" - Determines tools and context

Step 2: Determine Configuration

Based on answers, decide:

Context:

  • inline - Quick operations that need conversation context (< 1 min), or anything that asks the user a question
  • fork - Heavy operations that should run as a subagent (research, log analysis, wide file sweeps). A fork keeps its own reading out of the main context window.

If you choose fork, you must also decide background. Forked skills default to background: true — they detach and report back later. Set background: false when the user is waiting for the answer in the same turn, which is the normal case for an interactive skill. Leave the default only for genuinely fire-and-forget work. (background needs Claude Code v2.1.218+.)

Tools: Pick from:

  • Read, Write, Edit, Glob, Grep (file operations)
  • WebSearch, WebFetch (web access)
  • Bash (system commands)
  • AskUserQuestion (user interaction - only works with inline; a fork has no one to ask)

Optional fields worth considering: argument-hint if the skill takes arguments, disable-model-invocation: true if firing it by accident would damage something, model if the work is mechanical enough to run on a cheaper model. Full reference with every valid field: templates/skills/SKILL-TEMPLATE.md.

Step 3: Generate SKILL.md

Create .claude/skills/[name]/SKILL.md:

---
name: [skill-name]
description: [what it does]. Use when user says "[phrase]", "[phrase]", "[phrase]". Do NOT use for [nearest neighbour / wrong trigger].
context: [inline or fork]
background: false        # fork only - drop this line for inline skills
effort: [low|medium|high]
allowed-tools: [Read, Write, ...]
---

# [Skill Name]

[Purpose description]

## When to Use

- [Trigger 1]
- [Trigger 2]

## Process

1. [Step 1]
2. [Step 2]
3. [Step 3]

## Output Format

[Expected output structure]

## Tools

- [Tool]: [what it's used for]

The description is the whole ballgame. It is the only thing Claude reads when deciding whether to invoke the skill — the body is loaded after the decision is made. A description like "Create weekly reports" will not fire; one that names the phrases and the boundary will. description and when_to_use together are capped at 1,536 characters.

Step 4: Update CLAUDE.md

Add entry to the Skills table in CLAUDE.md:

| /[name] | [Description] | [inline/fork] |

Step 5: Test

Run the new skill with a test input and verify it works.

Guidelines

  • Keep skill names short and descriptive (1-2 words)
  • Each skill does ONE thing well
  • Use context: fork for anything that reads lots of files or does web research — and pair it with background: false unless the user really is meant to walk away
  • Write the Do NOT use for clause even when nothing seems to conflict; the next skill added will
  • Always include a clear output format section
  • Reference templates/skills/SKILL-TEMPLATE.md for the full field reference