agent-generator

Generate agents programmatically following Claude Code best practices. This skill contains templates and validation procedures for creating new agents.

Quick Process

1. Collect requirements: agent name, description, role, tools needed
2. Determine configuration:

- Model: sonnet (default), opus (complex tasks), haiku (simple) - PermissionMode: default (read-only), acceptEdits (implementation) - Context: fork (high output), inherit (needs context) - Hooks: Required for dangerous operations (Bash, Write)

1. Identify required skills (if any)
2. Create .claude/agents/{name}.md
3. Generate frontmatter with all required fields
4. Write concise instructions (20-40 lines)
5. Reference skills for detailed procedures
6. Validate: length, frontmatter completeness, no code examples

Agent Structure

.claude/agents/
└── agent-name.md     # 20-40 lines, instructions only

Frontmatter Template

---
name: agent-name
description: Single sentence when to use (< 100 chars)
tools: [Read, Write, Edit, Bash]
disallowedTools: [Write(/data/credentials/*)]
model: sonnet|opus|haiku
permissionMode: default|acceptEdits|dontAsk
context: fork|inherit
hooks:
  PreToolUse:
    - matcher: "Bash"
      hooks:
        - type: command
          command: "./scripts/validate-command.sh"
skills:
  - skill-name-1
  - skill-name-2
---

Field Rules:

• name: kebab-case, lowercase
• description: Single sentence, < 100 chars, action-oriented
• model: sonnet (default), opus (complex), haiku (simple)
• permissionMode: default (read-only), acceptEdits (implementation), dontAsk (dangerous)
• context: fork (high-volume output), inherit (needs parent context)
• hooks: Required for dangerous operations (Bash, Write)
• skills: List of skills the agent can invoke

Agent Templates

Read-Only Agent:

tools: Read, Grep, FindByName
disallowedTools: Write, Edit
permissionMode: default
context: inherit

Implementation Agent:

tools: Read, Write, Edit, Bash
permissionMode: acceptEdits
context: inherit
hooks:
  PreToolUse:
    - matcher: "Bash"
      hooks:
        - type: command
          command: "./scripts/validate-command.sh"
skills: [testing]

High-Volume Agent:

tools: Read, Bash
permissionMode: default
context: fork
skills: [testing]

Content Guidelines

Keep in Agent File (20-40 lines):

• Role and responsibilities
• Process/workflow steps
• When to use this agent
• References to skills for detailed procedures

Delegate to Skills:

• Code examples
• Detailed procedures
• Tutorials and best practices
• Command references

Validation Rules

• Agent file: 20-40 lines, instructions only
• Frontmatter: Complete (name, description, tools, model, permissionMode, context)
• Description: Single sentence, < 100 chars
• No code examples: Delegate to skills
• No tutorials: Delegate to skills
• Hooks: Configure for dangerous operations

Output

Creates agent file with validated frontmatter and concise instructions following Claude Code best practices.