Parse LLM Response into JSON Results

Standardize LLM replies to a format like { step: "Recon", status: "success", reason: "..."}. 3 story points.