This is an AI agent designed to communicate with the VAST parser dialect converter.
The VAST parser dialect converter can make use of externally-sourced information in order to make its conversions more accurate -- in a pre-AI world, this information would have come from a human operator receiving prompts in a VSCode session. This agent replaces the VSCode session and its human operator with Claude 3.7 Sonnet.
This project uses uv to run and manage its dependencies.
First, set the ANTHROPIC_API_KEY environment variable to a valid Claude API key.
Next, run vast-detect-parsers to listen on a Unix socket:
vast-detect-parsers -vast-hl-to-parser=socket=/tmp/vast.sock -o parser.mlir input.mlir
Then, run the agent:
uv run vspells-ai-agent /tmp/vast.sock
The AI overlord static analysis expert is now doing its thing.
The VAST Parser detection AI agent is licensed according to the Apache 2.0 license.
This research was developed with funding from the Defense Advanced Research Projects Agency (DARPA). The views, opinions and/or findings expressed are those of the author and should not be interpreted as representing the official views or policies of the Department of Defense or the U.S. Government.
Distribution Statement A – Approved for Public Release, Distribution Unlimited