See llms.txt for all machine-readable content.
Quick overview
Get an honest second opinion on a plan or decision before you act. Gemini plays devil's advocate and proposes a Plan B, Claude weighs both and recommends A, B or a mix. Works as a form or as a sub-workflow for AI agents.
How it works
- Input comes from a form (logged-in n8n users only) or from another workflow through Execute Workflow, with the fields plan, goal and context.
- A Code node wraps each piece of user text in quoted blocks and neutralises block markers inside it, so instructions in the text are treated as content, not commands.
- Gemini reviews the plan as devil's advocate: it proposes a different Plan B, names what Plan A overlooks and its biggest risk, in a fixed JSON schema.
- Claude weighs Plan A against Plan B and returns a recommendation (A, B or mixed) with reasoning, next steps, strengths, weaknesses and open questions, also in a fixed schema.
- If Gemini fails, Claude takes over both roles. If Claude fails, the report says INCOMPLETE and gives no recommendation.
- Form users see the formatted report. Calling workflows receive the result as JSON.
Setup
- Add a Google Gemini API credential to Gemini Devil's Advocate Model.
- Add an Anthropic API credential to Claude Synthesis Model.
- Set your own error workflow in the workflow settings.
- Activate the workflow and open the form URL of When Plan Submitted, or call it from another workflow with plan, goal and context. One run takes about a minute and uses both APIs.
Requirements
- Google Gemini API key
- Anthropic API key
Customization
- Swap either model sub-node for another provider. Keep the two different, so they don't share the same blind spots.
- Adjust the review instructions in Normalize and Quote Input and Create Synthesis Prompt.
- Use When Triggered by Workflow to add the review as a step before an AI agent acts.
Additional info
Your input is sent to both model providers. Don't submit personal or confidential data unless your agreements with them allow it.