[harness]
Served-model metadata for agent harnesses and prompt cache refreshes.
| Key | Type | Default | Description |
|---|---|---|---|
harness.annotate_chat_assistant_body |
bool | false |
annotate_chat_assistant_body appends a short “[Prism route: provider/model]” line to assistant content for /v1/chat/completions: non-streaming JSON bodies and OpenAI-style SSE streams. A step that only calls tools gets none, nor does an answer that is only a bracketed marker for the client program such as [SILENT], and the lines are removed from the history a client sends back before the model sees it. |
harness.announce_served_model |
bool | false |
announce_served_model adds separate, bounded Prism metadata to successful model responses without modifying choices[].message or output content. |
harness.prompt_cache_keep_warm_minutes |
integer | prompt_cache_keep_warm_minutes is how long after a conversation’s last turn Prism keeps its Anthropic prompt cache warm with one-token refreshes (Claude on anthropic, claude-oauth and OpenRouter). Nil selects the default (180); 0 disables the refreshes. |