Blockchain Weakness Classification (BWC) & Bounty Hunter Benchmark

This repository serves as a comprehensive framework for Ethereum security research, combining a novel vulnerability taxonomy with an automated testing harness for LLM agents.

Mission

1. Taxonomy (BWC): Maintain the Blockchain Weakness Classification, an open-source standard for categorizing EVM smart contract vulnerabilities, moving beyond simple bug lists to comprehensive failure modes.
2. Incident Benchmark: Curate a database of real-world security incidents, analyzing root causes and impact.
3. Bounty Hunter Agent: Develop and refine an LLM-driven "Bounty Hunter" skill capable of autonomously identifying vulnerabilities and generating working Proof-of-Concept (PoC) exploits using Foundry.
4. Reproduction Harness: Build a standardized benchmark of historical hacks to rigorously test the agent's ability to "one-shot" reproduce known exploits against local forks.

Repository Structure

• docs/: Contains the BWC taxonomy, incident logs (2016-2026), and the Bounty Hunter skill definition.
• src/ & test/: Foundry environment for the Exploit Benchmark and agent testing harnesses.

Installation & Setup

Prerequisites

• Foundry (Forge, Cast, Anvil)

1. Build Project

forge build

2. Install LLM Bounty Hunter Skill

To use the Bounty Hunter skill with your LLM CLI agent (e.g., Gemini CLI), create a symbolic link from this repository's skill definition to your agent's skills directory.

For Gemini CLI:

# Ensure your local skills directory exists
mkdir -p ~/.gemini/skills

# Link the bounty-hunting skill directory
ln -s "$(pwd)/docs/src/skills/bounty-hunting" ~/.gemini/skills/bounty-hunting

Once linked, you can activate the skill in your LLM session using: activate_skill bounty-hunting

Usage

Run Tests (Exploit Benchmark)

Execute the reproduction harnesses to verify historical hacks:

forge test

Documentation

Access the BWC and Incident Logs locally:

cd docs
mdbook serve