[Feature Request] Native selective component loading for plugins

Feature Request: Native Plugin Component Selection

Problem: Installing a plugin loads all agents, commands, skills, and MCP servers into context, even if
users only need a subset. For comprehensive plugins (ours has 24 agents, 13 commands, 11 skills), this
creates significant context bloat.

Current Workaround: We built a build-time filtering system:

1. /configure command asks 5 workflow questions (language, tools, etc.)
2. Stores preferences in ~/.claude/compound-engineering.config.json
3. Build script copies only enabled components from _source/ to output directories
4. Auto-rebuild hook reapplies config when plugin updates

Users report 30-50% context savings by disabling irrelevant components (e.g., Ruby-only devs skip
Python/TypeScript reviewers).

Proposal: Native support for selective component loading in plugin.json:

``json
{
"agents": [
{ "name": "rails-reviewer", "groups": ["ruby", "review"] },
{ "name": "python-reviewer", "groups": ["python", "review"] }
],
"installPrompt": {
"questions": [
{ "id": "lang", "ask": "Primary language?", "options": ["ruby", "python", "typescript"] }
],
"enableGroups": { "ruby": ["ruby"], "python": ["python"] }
}
}
``
Or simpler: let users toggle components on/off in /plugins settings.

Why This Matters:

• Context is precious; loading unused components wastes it
• Plugin authors want to build comprehensive toolkits, not minimal packages

Happy to share implementation details if useful.