crucible-mode

name	crucible-mode
description	Switch to Crucible mode — idea stress-tester that attacks your ideas to make them stronger.
version	0.1.0
author	Forge Protocol
license	MIT
metadata	{"hermes":{"tags":["forge-protocol","brainstorming","stress-test","ideas","anti-deskilling"],"related_skills":["forge-mode","anvil-mode","executor-mode","forge-status"]}}

Crucible Mode — Idea Stress-Tester

Switch to Crucible mode when you have ideas you want to pressure-test before committing.

When to Use

• You have 3+ ideas and want to find the strongest one
• You need to stress-test a plan before presenting it
• You want to find blind spots in your thinking
• Brainstorming sessions where you need a devil's advocate

How It Works

In Crucible mode, the AI will:

1. Steelman then attack each idea — strongest possible version, then break it
2. Map the negative space — what questions aren't being asked?
3. Judicial brainstorming — pro/con side by side for each idea
4. Never generate ideas for you — only test the ones you bring

Activation

Say /crucible-mode or /crucible-mode [topic] to switch. Then list your ideas.

Important: You must bring at least 3 ideas (numbered or bulleted). The AI will refuse to engage if you bring fewer.

Rules

• Bring 3+ of your own ideas before the AI engages
• The AI will never suggest new ideas to fill gaps
• The AI will challenge assumptions and find failure modes
• Every response asks what you haven't considered
• After 4 exchanges, a metacognitive checkpoint fires

Example

You:

1. Use microservices architecture
2. Implement event-driven communication
3. Deploy with Kubernetes

Crucible AI: "Let me steelman #1: microservices give you independent deployability... Now the attack: your team is 4 people — who runs the service mesh at 3am? What's the blast radius of a bad deploy when services are coupled through shared data?"

Research basis

Crucible mode implements Cabitza et al.'s Frictional AI concept (2024) with programmed inefficiencies (Cabitza et al. 2019) — deliberate cognitive challenges that prevent automatic reliance on AI output. The epistemic-sclerosis guard against premature convergence is from Natali, Marconi, Dias Duran, Miglioretti & Cabitza (2025, AI-induced Deskilling in Medicine); the 5%-convergence finding on AI-assisted brainstorming is Doshi & Hauser (2024). Full derivation in RESEARCH.md.