Skip to Main Content

Dependency scanners check versions, not code. If you install a vulnerable version of a package, you get an alert, even if your code never calls the function with the bug. A small Django shop gets 94 alerts, and nobody can tell which ones matter. vexproof answers that with evidence, for Python projects, without running them. It reads each advisory and the commit that fixed it, maps the fix to the functions it changed, and searches for call paths from the project's code to those functions. Meta's Muse Spark does the reading that static analysis can't: it picks which changed function the advisory is about, quoting the advisory to justify the pick, and judges call-path steps whose target is only known at run time. A deterministic verifier checks every model answer before vexproof uses it. The AI proposes; the verifier proves. For the shop: 14 alerts proven reachable with a verified call path, 14 proven unreachable, and 66 left open with a reason, because vexproof never marks an alert harmless without proof. Results are ranked (fix now, fix soon, review), with an upgrade plan and an OpenVEX export, in an accessible web app (WCAG 2.2 AA) or the CLI. I checked it against reality by running two Django projects under scripted workloads: 0 of 30 clears were contradicted, 0 of 1,246 loaded modules were missed, 25 of 29 reachable paths really ran, and Muse Spark's picks matched 22 of 22 hand-checked answers.

Team

henrymbuguak

More from Meta Global AI Developer Hackathon

Cardo turns the world around you into Kurdish flashcards. Kurdish is spoken by tens of millions of people, but it has very few learning tools, and most…

Kurdolingo project preview

oqed is an LLM-guided superoptimizer for x86-64. You point it at a C or C++ project and it does the rest: - builds the project and runs its workload…

oqed project preview

Playtest Agent plays a game from screenshots, looks for bugs, then replays each report and has a judge check it. I make games alone. Solo developers and…

Adrian project preview

Open-source npm malware detection on Meta's models: a live monitor that flags new malware about a day before the public lists, a Muse agent that investigates…

Gatekeeper project preview

Video2URDF turns one short phone video of a real articulated object (a laptop lid, a drawer, a desk lamp, a microwave door) into a measured URDF/MJCF a robot…

Video2URDF project preview

Glimmer Scope opens up Muse Glimmer 30B so you can see and steer the concepts it thinks with. I trained sparse autoencoders on nine layers of Glimmer's…

Kroonen AI project preview