cc-add

name	cc-add
allowed-tools	Read, Write, Glob, AskUserQuestion, Bash(ls:*), Bash(mkdir:*), Bash(pwd:*)
argument-hint	<rule|skill|agent> [local] args...
description	Create a Claude Code rule, skill, or agent — global (~/.claude) or project-local (.claude)

Arguments

$ARGUMENTS = {type} [local] {type-specific args...}

• Type (required, first token): rule, skill, or agent
• Scope (optional, second token): local targets the project's .claude/ directory; omitted means global ~/.claude/
• Remaining args by type:
  • rule: [{path-pattern}] {rule content...} — pattern is the first token only if it starts with * or contains / or {
  • skill: {name} {description...} — name becomes the directory name
  • agent: {name} {description...} — name becomes the filename (kebab-case)

Target Directory

Type	Global	Local
rule	~/.claude/rules/{category}/{name}.md	{project-root}/.claude/rules/{category}/{name}.md
skill	~/.claude/skills/{name}/SKILL.md	{project-root}/.claude/skills/{name}/SKILL.md
agent	~/.claude/agents/{name}.md	{project-root}/.claude/agents/{name}.md

For local, locate the project root first (look for .git, package.json, Cargo.toml, deno.json, etc.).

Workflow

1. Parse - Extract type, scope, and type-specific args
2. Analyze - Determine the target path and shape the content:
   • rule: pick a category (existing global ones: tools/, git/, rust/, ts/, i18n/, claude/, code/; create a new one when none fit — for local, choose a category fitting THIS project) and a kebab-case filename. If a path-pattern was given, plan paths: frontmatter.
   • skill: core conventions, patterns, anti-patterns; frontmatter name, description (specific enough for auto-discovery; add disable-model-invocation: true when the skill mutates state and must only run on explicit user invocation)
   • agent: core purpose, model tier per rules/claude/model-selection.md, color; frontmatter name, description, model, color
3. Draft - Write the full proposed file content
4. Confirm - Show the proposed file path and content
5. STOP - Wait for user approval (use AskUserQuestion)
6. Create - Only after approval, create the file (and parent directories)

Formats

rule (with optional path-pattern)

---
paths: "{ path-pattern }"   # only when a pattern was given
---

# Rule Title

Rule content here.

skill

---
name: { name }
description: { Brief description ending with period }
---

## Conventions

- **Key Area**: Guidance

## Anti-patterns

- Avoid X because Y

agent

---
name: { name }
description: { Brief description ending with period }
model: { tier per model-selection rule }
color: { blue | green | yellow | red | purple | cyan | magenta | orange | white }
---

{One-line persona/expertise statement}

## Principles

1. **Key Principle**: Explanation

## Workflow

1. **Phase**: Actions

## When Uncertain

{How to handle ambiguity}

Examples

• /cc-add rule *.py Always use type annotations → ~/.claude/rules/python/type-annotations.md with paths: "**/*.py"
• /cc-add rule local src/**/*.tsx Use functional components only → {project-root}/.claude/rules/react/functional-components.md
• /cc-add skill python Python idioms and best practices → ~/.claude/skills/python/SKILL.md
• /cc-add agent local api-designer Design RESTful APIs following best practices → {project-root}/.claude/agents/api-designer.md

Content Guidelines

• In English, optimized for AI comprehension
• Use clear, imperative language
• Focus on actionable instructions rather than verbose explanations

Begin

Parse $ARGUMENTS and draft the appropriate file. Ask for user approval before writing.