Anti-Apophenia Realist Skill

A Claude Skill that makes an agent reason like a hard-nosed realist: evidence first, null model first, falsification before confidence.

It is designed for puzzles, anomaly analysis, hidden-message claims, investigations, visual/data pattern searches, debunking, and any work where the agent must not hallucinate meaning into noise.

What It Does

• Treats attractive patterns as hypotheses, not findings.
• Forces a claim -> evidence -> null model -> alternatives -> prediction -> disconfirmation loop.
• Penalizes post-hoc fitting, multiple attempts, cherry-picking, and vague resemblance.
• Uses evidence grades A/B/C/D/X and verdicts such as strong, weak, rejected, and unknown.
• Keeps tempting branches in a parked ledger so the agent does not jump between tasks.
• Allows creative exploration only inside a clearly labeled speculative sandbox.

Repository Layout

anti-apophenia-realist-skill/
├── anti-apophenia-realist/
│   ├── SKILL.md
│   └── references/
│       └── evidence-principles.md
├── .gitignore
├── LICENSE
└── README.md

The installable skill folder is:

anti-apophenia-realist/

Install In Claude Code

Project-Local Install

Copy the skill folder into a project's .claude/skills/ directory:

New-Item -ItemType Directory -Force .claude\skills | Out-Null
Copy-Item -Recurse -Force .\anti-apophenia-realist .claude\skills\

Then invoke it in Claude Code:

/anti-apophenia-realist

Personal Install On Windows

New-Item -ItemType Directory -Force "$env:USERPROFILE\.claude\skills" | Out-Null
Copy-Item -Recurse -Force .\anti-apophenia-realist "$env:USERPROFILE\.claude\skills\"

Personal Install On macOS/Linux

mkdir -p ~/.claude/skills
cp -R ./anti-apophenia-realist ~/.claude/skills/

Example Use

/anti-apophenia-realist
Analyze this visual puzzle, but do not treat resemblance as evidence. Give me only claims that survive controls.

Expected response shape:

Verdict: weak/plausible/strong/rejected/unknown
Why:
- direct evidence
- counterevidence or uncertainty
Next:
- smallest test that changes the verdict

Notes

This skill is not medical advice and does not diagnose users. It translates well-supported reasoning and debiasing principles into an agent workflow: controls, falsification, base rates, disconfirmation, and explicit uncertainty.

Donations BTC - bc1q38wzs4vwgx84grry7z8mwfxv002weedm0puwtm ETH - 0xC7C0669726dfcAe130da615ACB94c8923229823F SOL - BaHamqokZt9UJTvMm31yPRXfAhTqvcMNQdqcbRNUGpEY