Activate prompt caching for Claude and xAI agents

[gitcommit90]

Summary

• add Anthropic cache_control breakpoints only for direct Claude models
• retain one stable conversation breakpoint plus a rolling three-tool-result frontier so growing agent loops reuse the preceding prefix
• add a stable, thread-scoped prompt_cache_key only for direct xAI models
• leave Bedrock, custom, OpenRouter, NVIDIA, and all other providers untouched

Verification

• exact request-shaping tests pass through the pinned ReRouted Claude and xAI adapters
• standalone live-provider probe using the exact candidate helper, outside live 1Helm:
  • Claude: create 15,687; next round read 15,687 + create 1,593; third round read 17,280 + create 1,591
  • xAI: 128 cached on first call, then 6,784 cached on calls 2 and 3 of 6,803 prompt tokens
• typecheck passed
• build passed
• focused provider/routing/architecture tests passed
• autonomy platform 25/25 passed
• full local suite reached browser tests; browser-only tests are unavailable in this resident container because Chromium lacks libglib-2.0.so.0, so hosted Source CI is the authoritative broad proof

Live 1Helm was not modified or restarted.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Team

Run ID: 16598cd3-66cc-4ca6-9890-22d68cf33428

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}