Skip to content

Repository files navigation

FailFast.ai — AI Agent Crash Test

Don't ask an AI coding agent if it can code. Try to break its code.

A benchmark and adversarial testing platform for AI coding agents. Connect Cursor, Claude Code, Codex, or any agent in your terminal — the platform feeds requirements level by level, runs chaos tests, and generates a performance report.

Built by Shubhransh Gupta

Analyse Mode (hosted): https://shubhransh-gupta.github.io/failfast-ai/


Two Modes

Analyse Mode Live Mode
Where GitHub Pages or local UI without API Local install (npm run dev:all)
Purpose Explore UI, watch simulated demos, sample data Benchmark your real agent via CLI
Connect your agent
Live monitor + report

Analyse Mode is static — mock crash tests and sample leaderboard data, no setup required.

Live Mode runs on your machine — CLI bridge, chaos tests on your code, real performance reports.


What It Does

  1. Gives your AI agent a coding challenge (e.g. "Build a stopwatch")
  2. Progressively adds harder requirements across 7 levels
  3. Runs chaos tests after each level (kill, corruption, race conditions…)
  4. Scores your agent and produces a shareable performance report
Your Terminal (Cursor/Codex)  ←→  CLI Bridge  ←→  FailFast API  ←→  Web Dashboard

Quick Start (Live Mode)

git clone https://github.com/shubhransh-gupta/failfast-ai.git
cd failfast-ai
npm install
cd backend && npm install && cd ..
npm run cli:build
npm run dev:all

Then open http://localhost:3000/run/ and follow the Launch Pad wizard.

Full setup instructions: setup.md · Web app: /setup/ when running locally

Try Analyse Mode without setup: open the hosted site or run npm run dev and click Watch Demo.


Supported Agents

Cursor · Claude Code · Codex · GitHub Copilot · Gemini · Custom


Architecture

Web Dashboard → FailFast API → CLI Bridge → Your Agent
                     │
              Chaos Engine → Evaluation Engine → Report

Documentation


Tech Stack

Next.js 15 · Express + WebSocket · CLI bridge · TypeScript · Tailwind CSS


Author

Shubhransh GuptaGitHub


License

MIT

About

Benchmark and adversarial testing platform for AI coding agents. Built by Shubhransh Gupta.

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages