Pinned Loading

1. Tool-Agent-Shift-Benchmark Tool-Agent-Shift-Benchmark Public

   Deterministic benchmark for tool-using agent safety under synthetic distribution shift, fault injection, monitor gating, and evaluator-boundary redaction.

   Python

2. streaming-agent-safety-evals streaming-agent-safety-evals Public

   No-training benchmark for evaluating agentic systems under distribution shift, uncertainty, and unsafe overconfidence.

   Python

3. android-trust-lab android-trust-lab Public

   Framework for analyzing Android device trust state across stock, rooted, and modified environments.

   Python