Claude Code skills, slash commands, and hooks tuned for quant research, trading-strategy analysis, and crypto backtesting workflows.
By @shakeebshaan · MIT license · contributions welcome
A collection of Claude Code skills you can drop into ~/.claude/skills/ to supercharge quantitative research and trading workflows.
Each skill is a focused prompt + instruction set that teaches Claude how to approach a specific quant task — backtesting, indicator design, risk analysis, strategy review — with the rigor the domain demands.
Walks Claude through a structured audit of a backtest:
- look-ahead bias
- survivorship bias
- overfitting signals (train/test split, parameter count vs. sample size)
- transaction-cost realism
- slippage model
- regime-dependence
Adversarial review of a trading strategy. Claude acts as a skeptical prop-desk risk manager and tries to find why the strategy will blow up.
Interactive skill for designing custom technical indicators. Claude asks about the hypothesis, helps you operationalize it, and generates vectorized pandas implementations.
Generates a one-page risk summary from a returns series: Sharpe, Sortino, max DD, CVaR, tail ratio, rolling correlation to BTC/SPY.
Explores hedging constructs for a position: delta hedging, basis trading, funding-rate capture, cross-exchange arbitrage. Claude walks through pros/cons and risk of each.
Audits a market-data CSV for common issues: missing bars, timezone drift, weekend handling, stale ticks, duplicate timestamps, survivorship in historical symbol lists.
# clone this repo
git clone https://github.com/shakeebshaan/claude-code-quant-skills.git
# symlink the skills dir into your Claude config
ln -s "$(pwd)/claude-code-quant-skills/skills" ~/.claude/skills/quantThen in Claude Code: /backtest-review — autocompletes.
skills/
├── backtest-review/SKILL.md
├── strategy-critique/SKILL.md
├── indicator-design/SKILL.md
├── risk-report/SKILL.md
├── hedge-lab/SKILL.md
└── data-scrub/SKILL.md
hooks/
└── pre-commit-backtest-check.sh
Quant workflows are high-stakes and boring to do manually. LLMs without guardrails hallucinate numbers, miss look-ahead bias, and produce confident nonsense. These skills encode domain judgment so Claude pushes back, asks the right questions, and produces output you can actually ship.
Add a skill as a new directory under skills/<name>/SKILL.md. Keep skills narrow — one skill, one job. Include a frontmatter block describing triggers.
MIT — see LICENSE.
- shakeebshaan/awesome-crypto-trading-agents — curated list of OSS trading bots + agent frameworks
- shakeebshaan/btc-hedge-lab — hedging strategies research playground