Request step that validates and truncates LLMRequest.history after earlier
request steps have prepended context. It repairs history rather than skipping:
unknown messages are dropped, prior citation markup is stripped, and
conversational turns are truncated to complete user/llm pairs.
Entry point group: brainforge.inference.request_steps
| Step id | Class |
|---|---|
brainforge.history |
RequestHistoryStep |
pipeline:
request:
- id: brainforge.historyRun this step last on the request pipeline. It writes
transaction.request_step_context["brainforge.history"].