New Skill

/new-skill

Scaffold a new skill following the Claude Code SKILL.md spec.

Usage

/new-skill /

E.g.: /new-skill general/find-duplicates or /new-skill research-pkg/new-experiment

Pre-flight

• Verify skills/// does not already exist
• Verify ` is one of: general/, research-pkg/, static-site/, productivity/, personal/` — or ask the user to add a new bucket if needed

> ⚠️ Claude Code plugin discovery is TOP-LEVEL only. CC auto-discovers skills//SKILL.md one level > deep — it does NOT scan skills///. So a skill you want Claude to auto-discover MUST live at > the top level skills// (like memory-flywheel, tui-installer, figma-design-fetch). The > skills/general/* copies are reached by Codex (skills/ glob) and opencode (deploy-skills), NOT by the CC > plugin. If the skill should be Claude-discoverable, create it at skills// (bucket is then just > an organizational tag in INVENTORY, not a path level).

Steps

1. Create directory:

``bash mkdir -p skills// ``

1. Ask the user for:

• One-line description (becomes description: frontmatter — this is what Claude reads to decide auto-invocation)
• Trigger: auto (Claude invokes when matched) / slash-only (user must type /) / both
• Skill body content — what does the skill do?

1. Write skills///SKILL.md:

```markdown --- name: description: disable-model-invocation: true ---

# /

## Master TOC

• [Pre-flight](#pre-flight)
• [Steps](#steps)
• [Notes for the agent](#notes-for-the-agent)
• [Companion](#companion)

## Pre-flight

## Steps

...

## Notes for the agent

## Companion

```

1. Update skills/README.md — add a row to the skill index table.

1. Update INVENTORY.md AND INVENTORY.zh.md — add a row to the Skills section.

1. Show the user the diff and ask for confirmation before committing.

Frontmatter conventions

• name — kebab-case, matches the directory name
• description — be specific about WHEN the skill fires. Claude uses this to decide auto-invocation. Vague descriptions cause both false positives and false negatives.
• disable-model-invocation: true — set for skills with side effects (deploy, force-push, etc.) that should only fire on explicit user request

Companion

• docs/CONTRIBUTING.md §"Adding a skill" — formal spec
• skills/README.md — index of all skills + bucket conventions
• Anthropic's skill-creator plugin — for more sophisticated skill authoring with evals

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: jajupmochi
• Source: jajupmochi/agent-harness
• License: MIT

Install and usage instructions live in the source repository linked above.