[sernote]

Orca-Code-Review — push 1

Severity	Count
P0	0
P1	0
P2	0
P3	0

✅ no blocking findings

The native Messages translator dropped supported cache_control markers and TTLs from system, content and tool blocks. A client's cache policy therefore disappeared before LiteLLM could build the provider request.

Preserve those markers, including marked multipart tool results. Translate a top-level automatic policy into an explicit marker on the last cacheable block of the existing translated payload, checking TTL conflicts and available breakpoint slots. Keep unmarked-request behavior and the established block grouping unchanged. No Anthropic-only top-level parameter is sent to other providers.

Validation: 842 tests passed in the offline Python 3.12 suite, with independent review and additional transformation/HTTP controls. Tests exercise the native HTTP route, schema, real Orca adapter and LiteLLM transformations, plus a separate real OpenAI Router/SDK transport path. The prepared environment lacks LiteLLM's optional mcp dependency needed for full dispatch with custom tools; those tool payloads are checked through the actual provider transformers. Ruff and git diff --check passed. No live provider cache hits or savings measured.

Discovered during a repository audit using audit-prompt-caching, then independently reproduced and reviewed.

[orcacode-review]

🐳 OrcaCode Review

✅ No findings — nothing to flag in this PR. Great work!

_{OrcaCode Review — Route Smarter. Ship Safer. Spend Less.
Engine-reported: 158 calls · 7M tokens · 92% cached}

❤️ Share · Install OrcaCode Review

Free on GitHub — the review runs on your own OrcaRouter key. If it helped, a shout-out goes a long way.

Share: X · Reddit · LinkedIn
Follow: X · Discord · LinkedIn · OrcaRouter