When this skill is invoked, PLAN your entire response before writing a single word. This implements LeCun's core criticism of current AI: that language models generate responses one token at a time without planning ahead, while real intelligence requires thinking multiple steps forward before acting.
Planning Process
Before producing any output, work through:
- Goal decomposition — Break the request into 5-10 sub-goals that must be accomplished in sequence.
- Dependency mapping — Which sub-goals must be completed before others can start? Identify the critical path.
- Resource identification — What knowledge, data, frameworks, and reasoning tools are needed for each sub-goal?
- Obstacle anticipation — What could go wrong at each step? How to handle it if it does?
- Alternative paths — If the primary plan hits a dead end, what's the backup approach?
- Quality criteria — What does "excellent" look like for each sub-goal? Define the standard before executing.
- Execution sequence — The exact order to tackle each sub-goal for maximum coherence.
- Integration plan — How all sub-goals connect into one unified, consistent final response.
- Self-evaluation checkpoints — After completing each sub-goal, verify it meets the quality criteria before moving on.
Output Format
Show the plan first, then execute it step by step. Make the reasoning visible at each stage — not a stream-of-consciousness answer. Mark each sub-goal as you complete it.
When to Use
Use this skill for complex requests: multi-file refactors, architecture designs, debugging sessions, feature implementations, or any task where diving in without a plan leads to backtracking.
If the user provides a task, plan and execute it. If they invoke this without context, ask:
What task should I plan before executing?
Source
Adapted from prompt #2 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.