[Plugin Idea]: LLM Self-Management Plugin

[xxynet]

Short description

A plugin that exposes plugin/MCP/Skills management as LLM-callable tools, so the AI can enable/disable/configure its own capabilities through conversation.

Problem to solve

Currently all plugin, MCP server, and Skills management requires the WebUI. The LLM has no way to know what's available, toggle capabilities on/off, or adjust configs — users must switch contexts manually.

Proposed solution

Register a set of LLM tools that wrap the existing manager APIs:

• Plugins: list_plugins, set_plugin_enabled, get_plugin_config, update_plugin_config, reload_plugin
• MCP: list_mcp_servers, set_mcp_enabled, add_mcp_server, update_mcp_server, delete_mcp_server
• Skills: list_skills, set_skill_enabled, refresh_skills
• Meta: get_system_status (overview of entire capability stack), submit_plugin_idea (open a GitHub issue with user confirmation)

Key safety measures:

• The plugin cannot disable itself (self-protection).
• Destructive operations (uninstall, issue submission) require user confirmation.
• Config changes return a before/after diff for transparency.

Alternatives considered

• Cron-based auto-tuning: too opaque, user can't see why things changed.
• Prompt-only approach (LLM tells user to go toggle something): defeats the purpose, still requires manual WebUI interaction.

Additional context

All required manager methods (set_plugin_enabled, enable_server, set_skill_scope, etc.) already exist. This plugin only needs to bridge them to the LLM tool layer — no core changes required.