information_gain: A New Measure of Investigative Progress in Luduan
In one full Luduan digital-forensics agent run, all 12 Actions completed successfully and the terminal kept streaming execution status — yet the run ended with ...
Buno.dev
Things I build, study, and keep.
Recent
Why I Built DFTK Instead of Accumulating More Scripts and Tools
Traditional digital forensics tools are mature but rigid. They struggle to adapt to emerging AI-driven forensics scenarios involving large language models, agen...
RAG Knowledge Base Poisoning: The Model Isn't Wrong — The Material It Reads Is.
RAG can make the model’s answers appear more evidence-based, but once the knowledge base is contaminated, the evidence itself becomes an attack surface....
Prompt Injection: Do Not Trust the Context Read by the Model
Prompt injection is no simple prompt prank. It emerges when input, permission and execution boundaries are intertwined in LLM applications, posing genuine secur...
TRC20-USDT On-Chain Evidence Preservation Experiment
TRC20-USDT transaction records are publicly queryable, have relatively stable field structures, and can be cross-checked through multiple on-chain data interfac...