add-component

Forge Add Component

Add a new component to an existing Claude Code plugin.

Usage

• /forge:add-component hook <plugin-path> — add a new hook + handler
• /forge:add-component agent <plugin-path> — add a new agent
• /forge:add-component skill <plugin-path> — add a new skill
• /forge:add-component mcp <plugin-path> — add MCP server config
• /forge:add-component bin <plugin-path> — add a bin tool
• /forge:add-component output-style <plugin-path> — add an output style

Parse $ARGUMENTS to extract the component type and optional plugin path (default: cwd).

Process

1. Read the existing plugin: Read plugin.json to understand the plugin's current structure
2. Understand context: Read existing hooks, agents, skills to understand naming conventions and patterns
3. Ask what's needed: Use AskUserQuestion to understand:
   • What the new component should do
   • For hooks: which event, what behavior
   • For agents: role, model tier, tool access
   • For skills: deep vs thin, auto-invoke vs manual
4. Research (quick): Run 1-2 WebSearch queries for best practices specific to this component type
5. Generate: Write the new component file(s) following the plugin's existing conventions
6. Update manifest: If needed, update plugin.json (e.g., add hook file reference)
7. Update install script: Add the new file to the required files list in install.sh
8. Validate: Run a quick validation pass on the new component

Component Templates

Reference ${CLAUDE_PLUGIN_ROOT}/data/component-templates.jsonl for proven patterns.

Use templates from the /forge:generate skill for the specific component type:

• Hook handlers: Python stdlib only, stdin JSON, stdout JSON
• Agents: YAML frontmatter with name, description, model, tools, color
• Skills: YAML frontmatter with description, argument-hint, version
• MCP: .mcp.json format with server configurations
• Bin tools: Python executable with argparse, data dir resolution
• Output styles: YAML frontmatter with name, description, formatting rules