Arbitrage Skill — Development skill for Claude Code

arbitrage

[**English**](#english) · [**Français**](#français)

A Claude Code skill that flags when an unvalidated hypothesis is being treated as an absolute rule — in code, schemas, prompts, or architecture decisions.

Un skill Claude Code qui detecte les hypotheses non validees traitees comme des regles absolues — dans le code, les schemas, les prompts ou les decisions d'architecture.

---

English

What it does

In natural language as in code, we often treat a frequent observation as a universal law. That slippage produces fragile systems. This skill spots it.

It distinguishes:

• Heuristic — empirical rule, often true, based on observation or intuition.
• Algorithmic — closed contract, deterministic, precise expectation.
• Gray zone — presented as algorithmic but resting on an unvetted heuristic premise.

When you write `if language == "fr": apply_formal_tone()`, that's a bet (Quebec, francophone Africa, French startups have different registers). The skill calls it out so you decide on purpose, not by accident.

Usage

Three depth levels:

Trigger	Level	Output
Background flag, inline detection	QUICK	2–3 line callout inside the normal answer
/arbitrage or /arbitrage	STANDARD	Full classification block + post-analysis workflow
/arbitrage deep	DEEP	Standard block + 3–6 month impact projection + validation tests

Trigger phrases the skill recognises:

• /arbitrage
• "check my assumptions"
• "is this always true"
• "am I oversimplifying"
• "review this spec/prompt/schema for logic"
• "is this assumption safe"

Install

Claude Code looks for skills in `~/.claude/skills//SKILL.md`.

git clone https://github.com/Moodtuner997/arbitrage-skill.git ~/.claude/skills/arbitrage

Or copy the folder manually into `~/.claude/skills/arbitrage/`. Restart your Claude Code session.

Optional: background mode

Background mode injects a brief QU