chore: C2 DELETE 204+body, C4 cached_tokens, C5 utcnow + LICENSE (T011)

agent_platform/llm.py

@@ -72,10 +72,20 @@ async def chat(
 
     usage = {}
     if hasattr(response, "usage") and response.usage:
+        # LiteLLM exposes cached prompt tokens via prompt_tokens_details.cached_tokens
+        # (see litellm.types.utils.Usage). Providers like Anthropic/OpenAI surface this
+        # when prompt caching is active; surfacing it here flows through LLMResponse.usage
+        # to any audit log that serialises the usage dict.
+        cached_tokens = 0
+        details = getattr(response.usage, "prompt_tokens_details", None)
+        if details is not None:
+            cached_tokens = getattr(details, "cached_tokens", None) or 0
+
         usage = {
             "prompt_tokens": response.usage.prompt_tokens or 0,
             "completion_tokens": response.usage.completion_tokens or 0,
             "total_tokens": response.usage.total_tokens or 0,
+            "cached_tokens": cached_tokens,
             "finish_reason": response.choices[0].finish_reason or "stop",
         }