Mental Models Skill for Claude Code

A portable Claude Code skill for applying core mental models to audits, stress tests, brainstorming, strategy, product, code, and decision reviews.

It combines two workflows:

• Audit mode — find blind spots, hidden assumptions, second-order effects, incentive problems, and downside risk in an existing plan or artifact.
• Brainstorm mode — generate reframes, option sets, and fast tests for open-ended problems.

Install

Clone this repo and copy the skill into your global Claude Code skills directory:

git clone https://github.com/ogiberstein/mental-models-claude-code-skill.git
mkdir -p ~/.claude/skills
cp -R mental-models-claude-code-skill/skills/mental-models ~/.claude/skills/

For project-scoped installation, copy it into a repository's .claude/skills/ directory instead:

mkdir -p .claude/skills
cp -R /path/to/mental-models-claude-code-skill/skills/mental-models .claude/skills/

Or install globally with the helper script:

git clone https://github.com/ogiberstein/mental-models-claude-code-skill.git
cd mental-models-claude-code-skill
./scripts/install.sh

Use

In Claude Code, ask naturally:

Use the mental-models skill to stress test this launch plan.

Apply mental models to brainstorm non-obvious growth ideas for this product.

What are we missing in this architecture decision? Use first principles, inversion, incentives, and second-order thinking.

What it covers

The skill uses a compact set of 19 mental models across five groups:

• Decision-making: first principles, inversion, second-order thinking, opportunity cost
• Systems: feedback loops, emergence, leverage points, map is not the territory
• Probability/statistics: Bayesian updating, regression to the mean, base rates, fat tails
• Human psychology: Hanlon's Razor, incentive-caused bias, availability heuristic, identity lock-in
• Strategy/competition: comparative advantage, margin of safety, network effects

The skill tells Claude not to force every model into the answer. It should select the models that create signal, then collapse findings into concrete corrections or tests.

Output modes

Audit mode

Best for reviewing an existing plan, strategy, implementation, proposal, memo, or decision.

Typical output:

• Object summary
• Highest-signal findings
• Additional checks
• Cross-model themes
• Missing evidence or blind spots
• Recommended corrections

Brainstorm mode

Best for creating options and reframes.

Typical output:

• Problem frame
• Best reframes
• Idea families
• Weak ideas to discard
• Ranked shortlist
• Fastest next tests

Repository layout

skills/mental-models/SKILL.md   # Claude Code skill
examples/                       # Example prompts
scripts/install.sh              # Local installer

Compatibility

This is packaged as a Claude Code skill directory with a SKILL.md file. It is intentionally dependency-free and can also be adapted for other agents by copying the markdown instructions.

License

MIT