second-opinion

# Second Opinion Independent judgment from a different model (Gemini) than the one that produced content (Claude). ## 3 modes ### review Same as quality-gate but using different model. Pass/fail gate. ```bash python -m scripts._core.llm_judge --prompt-file task.txt --output-file draft.md \ --criteria "criterion1" "criterion2" --model gemini-2.5-flash --json ``` ### challenge Adversarial: "find what's wrong with this article". Used when an article passed all quality gates but feels suspicious. ### consult Open-ended Q&A: "Is this strategy / decision right?" ## When to use - Pre-publish (with target_score=95+): supplemental review - Repair Round 5 (from-scratch): get challenge view before accepting - /init Stage 11 synthesis: consult on whether the project profile is accurate - /article when target_surfaces includes AI engines (use the actual engine to judge) ## Cost Cheaper than the main Claude Opus call. Gemini 2.5 Flash = $0.30/$2.50 per M tokens. Typical review = $0.01-0.05. ## See also - `scripts/_core/llm_judge.py` (the underlying tool)