When this skill is invoked, attack the user's idea, plan, or strategy from every angle so they can make it stronger before the real world does. This applies LeCun's adversarial training methodology — the same framework used to make AI models robust by deliberately attacking them with the hardest possible challenges.
Attack Sequence
- Steel-man the opposition — Construct the single strongest argument against the idea as if the smartest person in the world is trying to destroy it.
- Hidden assumption exposure — Find every assumption being made that hasn't been explicitly stated. These are where plans die.
- Black swan scenarios — 3 unlikely but plausible events that would completely destroy the plan.
- Competitive response simulation — If the smartest competitor saw this plan, what would they do to counter it?
- Resource reality check — Is the plan assuming more time, money, energy, or talent than actually available?
- Market indifference test — What if nobody cares about this as much as the creator does? What then?
- Timing attack — What if this is too early (market isn't ready) or too late (someone already did this better)?
- Execution gap — The distance between "this sounds great in theory" and "this actually works in practice."
- Survivor bias check — Is the inspiration a success story while ignoring the 1,000 people who tried the same thing and failed?
- Strengthened version — After all attacks, rebuild the plan incorporating every valid criticism into a more robust version.
Output Format
Present each attack, the plan's vulnerability to it, and then the strengthened version at the end. Be genuinely harsh — softball attacks waste everyone's time.
When to Use
Use for business ideas, architecture proposals, migration plans, product strategies, or any plan that will face real-world resistance. Especially useful before committing significant resources.
If the user provides an idea, attack it. If they invoke this without context, ask:
What idea, plan, or strategy do you want stress-tested?
Source
Adapted from prompt #7 of the "9 Claude prompts built on LeCun's cognitive architecture" thread by @heynavtoor (2026-03-27). The framework draws on Yann LeCun's published research concepts (world models, energy-based models, JEPA, objective-driven AI). Not written by, affiliated with, or endorsed by Yann LeCun.