Foresight

A Claude Code skill for making specific, falsifiable, short-to-medium term predictions about how AI will reshape industries, jobs, and society.

What it does

Most predictions about AI are vague ("AI will change everything") or hedged ("it depends on many factors"). Foresight forces specificity:

• Concrete claims with numbers and timeframes you can bet on
• Confidence levels (50-95%) with calibration guidance
• Falsification conditions — what would prove the prediction wrong
• Leading signals — what to watch for first

How it works

Current facts (not opinions)
      +
Premises (high-confidence trend extrapolations)
      ↓
Logically necessary or highly probable outcomes

The skill ships with a default set of premises about AI development trends (see references/premises.md). Users can override any premise to see how conclusions change.

For complex systems ("what will X look like?"), the skill can perform structural decomposition — breaking the system into functional modules, analyzing each independently, then examining their interactions. See references/structural_decomposition.md for a worked example.

Usage

/predict autonomous driving 6 months
/predict entry-level employment market
/predict                                  # interactive exploration

Install

# Clone to your Claude Code skills directory
git clone https://github.com/Fidingks/foresight.git ~/.claude/skills/foresight

Or copy manually:

~/.claude/skills/foresight/
├── SKILL.md
├── references/
│   ├── premises.md
│   └── structural_decomposition.md
└── README.md

Customization

• Premises: Edit references/premises.md to adjust the baseline assumptions. The skill works with any set of premises — swap in your own domain-specific ones.
• Prediction log: After each prediction, the skill can optionally save it to predictions/ in your working directory for later backtesting.

Example output

Core prediction: Within 12 months, at least 3 of China's top 10 law firms will cut entry-level hiring by 30%+.

Confidence: 75%. This follows directly from P3 (cost pressure) + P2 (AI tools becoming accessible to non-technical professionals). The leading signal to watch: job posting volumes on major legal recruitment platforms dropping quarter-over-quarter starting Q3 2026. This prediction fails if legal industry headcount stays flat or grows through 2027 — that would mean regulatory or cultural barriers to AI adoption in law are stronger than expected.

License

MIT