Summary

Would love to see support for Anthropic's prompt caching feature (cache_control) to reduce token usage on repeated system prompts.

Context

• Our system prompt (AGENTS.md, SOUL.md, TOOLS.md, etc.) is ~15-20k tokens
• Anthropic claims 90% token reduction on cached content
• Currently Clawdbot doesn't appear to send cache_control: {"type": "ephemeral"} headers

Questions

1. Is prompt caching on the roadmap?
2. Key question: Does prompt caching reduce Claude Max subscription usage cap, or only API billing? (We're using Claude Max OAuth as primary)

Potential Impact

• 90% reduction on ~15-20k tokens per turn = significant usage savings
• Especially valuable for users on Claude Max who hit daily caps

References

• Anthropic Prompt Caching Docs
• Anthropic Cookbook

Thanks for building Clawdbot! 🦞