LLM Fuzz CI: agentic fuzzing finds bugs via existing tests

LLM Fuzz CI is a new tool where an AI agent reads your code and tests, generates adversarial inputs, and your normal pytest or vitest suite replays them. Only real assertion failures are reported, avoiding static-analysis false positives. It runs as a GitHub Action with a spend cap of roughly $5–10 per run.

• Agent generates inputs; standard pytest or vitest runs the tests
• Only real assertion failures are reported, no CVE noise
• budget_usd parameter caps LLM API spending
• No state persists between runs, so bugs may be rediscovered