Meta Mental Models Agent Skill

Mental Models Toolkit

Framework

IRON LAW: Use Multiple Models, Not Just Your Favorite

"To a man with a hammer, everything looks like a nail." (Munger)
A single mental model creates blind spots. Apply 2-3 models from
DIFFERENT disciplines to any important decision. Where models agree,
confidence is high. Where they disagree, the disagreement reveals
the most important dimension of the decision.

Core Mental Models (Cross-Disciplinary)

From Physics/Engineering

| Model | Principle | Application |

|-------|-----------|------------|

| Inversion | Instead of "how do I succeed?", ask "how would I fail?" Then avoid that. | Risk management, pre-mortem |

| Second-order effects | Every action has consequences, which have consequences. Think two steps ahead. | Policy design, strategy |

| Entropy | Systems tend toward disorder without energy input. Things decay by default. | Maintenance, quality, relationships |

From Biology

| Model | Principle | Application |

|-------|-----------|------------|

| Evolution/natural selection | What survives is what's adapted, not what's "best" in absolute terms. | Market competition, product-market fit |

| Red Queen effect | You must keep improving just to stay in the same place (because competitors improve too). | Competitive strategy |

| Niche specialization | Generalists and specialists coexist because they serve different niches. | Market positioning, career strategy |

From Mathematics/Statistics

| Model | Principle | Application |

|-------|-----------|------------|

| Pareto principle (80/20) | ~80% of effects come from ~20% of causes. | Prioritization, resource allocation |

| Regression to the mean | Extreme results tend to be followed by more average ones. | Performance evaluation, forecasting |

| Bayes' theorem | Update beliefs based on new evidence, weighted by prior probability. | Decision-making under uncertainty |

From Psychology

| Model | Principle | Application |

|-------|-----------|------------|

| Incentive-caused bias | People do what they're incentivized to do, not what you ask them to do. | Compensation design, policy design |

| Circle of competence | Know what you know and what you don't. Stay within your expertise for high-stakes decisions. | Self-awareness, delegation |

| Hanlon's razor | Never attribute to malice what is adequately explained by ignorance or incompetence. | Conflict resolution, workplace dynamics |

Application Method

• State the decision or problem
• Select 2-3 relevant models from different disciplines
• Apply each model to the situation — what does it suggest?
• Compare conclusions — where do they agree? Where do they disagree?
• Synthesize — the disagreement reveals the key trade-off to resolve

Output Format

# Multi-Model Analysis: {Decision}

## Models Applied
| Model | Discipline | Insight |
|-------|-----------|---------|
| {model 1} | {field} | {what this model says about the situation} |
| {model 2} | {field} | {what this model says} |
| {model 3} | {field} | {what this model says} |

## Convergence
{Where models agree — high confidence}

## Divergence
{Where models disagree — key trade-off to resolve}

## Synthesis
{Recommended decision based on multi-model analysis}

Gotchas

• Models are simplifications: Every model omits something. The map is not the territory. Use models as lenses, not as truth.
• Model inventory grows over time: Start with 10-15 core models. Add new ones as you encounter new domains. Quality of application matters more than quantity of models.
• Some models conflict by design: Inversion says "avoid failure." Evolution says "failure is how you learn." The conflict is resolved by context: avoid catastrophic failure, embrace recoverable failure.
• Don't force-fit: Not every model applies to every situation. If a model doesn't naturally illuminate the problem, skip it — don't stretch it to fit.

References

• For expanded mental models catalog (50+), see references/mental-models-catalog.md