Adith Manikonda Project / 2026 / Security / AI / ML / Research
AGENTLINT
Framework for evaluating check-then-act binding risks in LLM agent tool workflows, with tool contracts, a workflow IR and a pilot benchmark.
Source ↗Purpose
Research prototype for finding the TOCTOU gap between what an agent verified and what it then acted on.
Built With
- Python
- Static Analysis
- Benchmarks
More from Adith Manikonda
- 2026MORPHReproduces the machine a bug happened on: capture a user's CPU, memory, locale and network conditions as a profile, replay them somewhere else, and keep the cause as a regression test.
- 2026BF-SERVERHTTP server whose request parsing, routing and response generation are written in Brainfuck, fed by a plain TCP listener.
- 2026LOBOTOMLCompiles a trained neural network into a Brainfuck program: weights are quantised to fixed point, lowered through an IR and run as inference on a Brainfuck VM.