LLM Fuzz CI

LLM Fuzz CI is a GitHub Action that lets a coding agent write adversarial inputs for tests, reporting only failures.

LLM Fuzz CI is a GitHub Action that lets a coding agent read your code and write adversarial inputs for tests you mark. The tool operates in two phases on separate machines to prevent reward hacking. First, an agent reads the code, tries to break marked tests, and exports the inputs. Second, a fresh runner replays those inputs through your normal test run. It currently supports Python with pytest and JavaScript with vitest. The agent can use Codex with an OpenAI or OpenRouter key, or Claude Code. Every run writes a summary, an artifact with inputs and traces, can create a GitHub issue, and can fail the build.

LLM Fuzz CI at a glance

Best for: Software engineer
Runs on: GitHub Action, API, CLI
Pricing: Free
Released: October 2026
Category: AI coding tool
Website: tiime-software.github.io

Key features

Runs in GitHub Actions for continuous integration
Uses coding agents to generate adversarial test inputs
Supports Python with pytest and JavaScript with vitest
Runs generation and test replay on separate machines
Configures spend limits with budget_usd
Exports inputs and agent traces as build artifacts
Creates GitHub issues on test failure

Use it for

Attack tests with adversarial inputs generated by LLM agents
Find logic bugs and advanced vulnerabilities using existing tests
Test accounting agents or prompt injection on LLM apps
Gate releases and schedule regular checks in CI

Alternatives to LLM Fuzz CI

jevgrep — Reduces coding agent costs by providing targeted, relevant repository context.
Hazzel — Brings inspectable safety mechanisms, local model support, and zero subscriptions.
DeepTeam — It automates penetration testing using adversarial LLM simulations.
Agentfork — It replays steps from cache and runs sweeps to provide pass rates instead of anecdotes.
FuXi — Runs as a single binary with automatic model routing and failover.
OpenCode — Works with 75+ model providers, local models, and existing subscriptions
uithub — Provides immediate full repository context for LLMs by changing one URL letter.
Squeeze — Cuts token usage for coding agents by shrinking verbose tool output while retaining key data.
GitAgent — Manage AI agents with familiar git workflows like branching, diffing, and reverting.
CodeScan Artex — It uses multiple parallel AI agents to read source code and verify vulnerabilities.
Vibium — Provides browser automation and evidence capture to check AI-built software.
Mantis — Automates vulnerability discovery and patching using AI agents.

All LLM Fuzz CI alternatives