Adversarial SAST
🛡
Adversarial SAST — false positives die on screen
Auditing and guarding with LLMs — adversarial code review, prompt-injection and content guardrails. Report only what you can exploit.
Adversarial SAST — false positives die on screen
Note My two-stage SAST (detect, then adversarially refute each finding). Paused since 2026-08 — code and README still browsable.
An Open Source Cyber Security Agent
Note DevSecOps / cybersecurity assistant (WhiteRabbitNeo lineage), Qwen2.5-Coder-7B base - same family as the Adversarial SAST Space engine.
Note Tiny security-reasoning SLM (DeepSeek-R1 distill), edge-deployable.
Note 51.8k long-form defensive-security instruction rows (detection, IR, threat intel) with provenance. Fills the dataset gap here. Small traction, license is 'other'.