wtf is skynet?

Skynet Audit is a Python tool that uses a large language model to find security bugs in a codebase, but it tries to do this more carefully than just pasting code into a chatbot. The problem it is solving is that plain rule based security scanners understand syntax but not the meaning of the business logic, while an LLM reading raw code tends to make things up, miss connections that span multiple files, and get expensive fast if you are not careful. Skynet's answer is to first turn the codebase into a map before asking the model to look at anything. It parses the repository into a graph of functions and classes and how they call each other, storing this in a local database. Instead of handing the model one giant pile of code, it hands it small, well connected chunks, prioritizing the pieces most likely to matter for security first, so token spending is not wasted on irrelevant code. On top of that graph, it traces how untrusted data could flow from an entry point to a dangerous operation, and when the graph has gaps it cannot resolve on its own, it uses a small agent with access to a language server to fill in the missing links rather than guessing. It also layers in known vulnerability patterns, framework specific security knowledge, and a memory of past false positives from the same project, all fed into the model's prompt. For deeper investigations, there is an eight stage pipeline that goes from initial reconnaissance through hunting, validating, filling gaps, removing duplicates, and finally producing a report, with its own tool agent that can read files, search, and list directories. Progress and cost are tracked so a run can be paused and resumed. It works with any OpenAI compatible LLM endpoint, including DeepSeek. The project is explicitly labeled Alpha stage, meant to support security research and human reviewed audits rather than replace professional testing tools, and it is released under the MIT license.