Optimize AI prompts for provider caching

[ryokun6]

Summary

• Add shared prompt-cache helpers for Anthropic cache-control metadata, long-content cache marks, per-step cache marks, and provider-option merging.
• Mark reusable system prompts in chat, Telegram, memory extraction/processing, applet AI, IE generation, room replies, parse-title, and song title AI parsing.
• Add per-step Anthropic cache control for Ryo chat and Telegram tool loops while preserving OpenAI reasoning provider options.
• Expand provider-options unit coverage for prompt caching helpers.

Testing

• bun test tests/test-ai-model-provider-options.test.ts tests/test-telegram-status.test.ts
• bun run test:unit
• bun run build

  

[ryos-deploy]

The preview deployment for ryos-dev failed. 🔴

Open Build Logs | Open Application Logs

Last updated at: 2026-04-08 12:25:13 CET