Nd Scaffold Asset

Scaffold a new asset in an nd source

Create a new coding agent asset inside a registered nd source with the correct structure so nd's scanner can discover and deploy it.

When to use

• Adding a new skill, agent, command, rule, context file, output-style, or hook to an existing nd source directory.
• Unsure what directory layout or files an asset type requires.
• Want to confirm the asset will pass nd's scanner validation before deploying.

Prerequisites

• nd is installed and at least one source is registered. Run nd source list to confirm.
• You know which source directory you want to add the asset to (its path on disk).

Workflow

Step 1: identify the target source

If the user has not specified a source, list registered sources and ask which one to use.

nd source list

Record the source path from the output. All new files will be created under this path.

Step 2: determine the asset type

Ask the user what type of asset to create. The supported types and their directory names are:

Step 3: collect name and description

Ask the user for:

1. Name: a kebab-case identifier (e.g., code-review, no-emojis, go-project-rules). This becomes the filename or directory name.
2. Description: a one-line summary of what the asset does.

Step 4: create the asset

Use the templates below. Replace <SOURCE> with the source path, <NAME> with the asset name, and <DESCRIPTION> with the description.

Skill

Create a directory with a SKILL.md file inside.

<SOURCE>/skills/<NAME>/
└── SKILL.md

SKILL.md content:

---
name: <NAME>
description: <DESCRIPTION>
---

# <Title derived from NAME>

<DESCRIPTION>

## When to use

- (describe the situations where this skill applies)

## Workflow

1. (step-by-step instructions for the agent)

Agent

Create a single .md file.

<SOURCE>/agents/<NAME>.md

Content:

---
name: <NAME>
description: <DESCRIPTION>
---

# <Title derived from NAME>

<DESCRIPTION>

## Role

(describe the agent's role and responsibilities)

## Instructions

(provide the agent's behavioral instructions)

Command

Create a single .md file.

<SOURCE>/commands/<NAME>.md

Content:

---
name: <NAME>
description: <DESCRIPTION>
---

# /<NAME>

<DESCRIPTION>

## Usage

`/<NAME> [arguments]`

## Behavior

(describe what happens when the command is invoked)

Output style

Create a single .md file.

<SOURCE>/output-styles/<NAME>.md

Content:

---
name: <NAME>
description: <DESCRIPTION>
---

# <Title derived from NAME>

<DESCRIPTION>

## Formatting rules

(describe the output formatting rules)

Note: output styles require manual registration in the agent's settings.json after deployment.

Rule

Create a single .md file (or a directory if the rule spans multiple files).

<SOURCE>/rules/<NAME>.md

Content:

# <Title derived from NAME>

<DESCRIPTION>

(rule content: instructions the agent must follow)

Context

Create a folder with a CLAUDE.md file and an optional _meta.yaml.

<SOURCE>/context/<NAME>/
├── CLAUDE.md
└── _meta.yaml   (optional)

CLAUDE.md content:

(context content: project rules, conventions, or instructions for the agent)

_meta.yaml content (if the user provides metadata):

description: "<DESCRIPTION>"
tags: []

The _meta.yaml file supports these optional fields: description, tags, target_language, target_project, target_agent.

Hook

Create a directory with a hooks.json file.

<SOURCE>/hooks/<NAME>/
└── hooks.json

hooks.json content:

{
  "hooks": []
}

Note: hooks require manual registration in the agent's settings.json after deployment.

Step 5: create parent directories

If the asset type directory does not exist yet in the source (e.g., there is no skills/ folder), create it. The scanner only looks for directories that exist.

mkdir -p "<SOURCE>/<TYPE_DIR>/<NAME>"

Step 6: validate the asset

After creating the files, verify the scanner will recognize the asset by running:

nd list --source <SOURCE_ID>

The new asset should appear in the output. If it does not, check:

• Skill: the directory contains a file named exactly SKILL.md (case-sensitive).
• Agent/command/output-style: the file ends in .md and is not inside a subdirectory deeper than one grouping level.
• Context: the folder contains at least one .md file that does not start with _.
• Hook: the directory contains a file named exactly hooks.json.
• General: the entry name does not start with . (dot-prefixed entries are skipped). The entry is not a symlink (symlinks are skipped during scanning). The entry is not inside .git or node_modules.

Step 7: show deployment instructions

Tell the user how to deploy the new asset:

# Deploy globally (default)
nd deploy <NAME>

# Deploy to a specific project
nd deploy <NAME> --scope project

# Deploy with explicit type qualifier
nd deploy <TYPE>/<NAME>

Where <TYPE> is the asset type directory name (skills, agents, commands, output-styles, rules, context, or hooks).

If the asset type is hooks or output-styles, remind the user that they must also register the asset in the agent's settings.json after deploying.

Grouping directories

nd supports one level of grouping inside asset type directories. For example:

skills/
├── coding/              # grouping directory
│   ├── code-review/     # skill (has SKILL.md)
│   └── refactoring/     # skill (has SKILL.md)
└── writing/             # grouping directory
    └── tone-check/      # skill (has SKILL.md)

If the user wants to organize their asset inside a group, create it one level deeper. The scanner recurses one level into non-matching directories to find assets inside grouping folders.

Manifest-based sources

If the source contains an nd-source.yaml manifest, convention-based scanning is disabled. The manifest must explicitly list the directory paths for each asset type. After creating the asset, verify the manifest includes a path entry that covers the new asset's location. If not, update nd-source.yaml:

paths:
  skills:
    - skills
    - custom/path/to/more-skills