WHAT A HARNESS IS

ARTIFICIAL INTELLIGENCE ASSISTANT PROMPTS [START OF PROMPT] Review this harness's control-plane code and tell me exactly which lines belong to "the model" (prompt and response handling) versus "the harness" (everything else): [PASTE LAB main_loop.py OR YOUR HARNESS'S LOOP FILE HERE]. Flag any place where harness logic and model-prompted behavior are tangled together in a way that would make the two hard to test or swap independently. Output format: an annotated line-range list (Lines X-Y: Model or Harness) followed by a short list of entanglement risks and suggested refactors (cleanups that reorganize code without changing what it does). [END OF PROMPT] [START OF PROMPT] Given this incident transcript, determine whether the failure originated in the model's reasoning or in the harness's control-plane logic (routing, retries, tool dispatch): [PASTE TRANSCRIPT OR LAB RUN LOG HERE]. Produce a short root-cause note suitable for a postmortem, citing the specific lines or events that support your conclusion, and recommend one control-plane change that would prevent recurrence. Output format: Root Cause (one sentence), Evidence (bullets), Recommended Fix (one paragraph). [END OF PROMPT]