Feature: Don't count 'how to use Claude' questions against usage limits

Feature Request

Problem:
When users ask Claude Code questions about how to use Claude Code — things like "what does --system-prompt-file do?", "how do I configure hooks?", "can I preload files into context?" — those questions cost tokens and count against usage limits just like any other interaction.

This discourages users from learning about Claude Code's features, which means they use it less effectively, which means they get less value from it.

Proposed solution:
Detect when a conversation is primarily about learning how to use Claude Code itself (features, flags, configuration, workflows, best practices) and exempt those interactions from billing / usage limits.

Possible approaches:

• A dedicated /help-me or /learn slash command that enters a free-tier mode for questions about Claude Code itself.
• Classify "meta" questions about Claude Code usage and credit back the tokens used.
• An enhanced built-in help system (beyond /help) that can answer usage questions without LLM calls, e.g., a local documentation search.

Why this matters:
Users who understand Claude Code's features get dramatically more value from it. Making it free to learn encourages exploration and adoption, which benefits both users and Anthropic. The current model penalizes curiosity — a user who spends 10 minutes learning about advanced features like hooks, MCP servers, or subagents is paying for knowledge that will make them a better (and more loyal) customer.