How a coding agent like Claude Code works and how to keep it under control, told through one fictitious HbA1c report: rules, context, subagents, skills, privacy and which model to use.
My systems finished second in both the speech recognition and the speaker diarization tracks of the ALBAYZIN 2026 Speech-COSER challenge, on dialectal rural Spanish, trained on nothing but the challenge's own COSER data.
R-LoRA is a variant of LoRA that tempers the size of the adapter's input with one learnable exponent per layer. On four clinical speech corpora it beats LoRA by 12.4 points of Macro-F1, and it is a drop-in change for any Transformer that LoRA can adapt.
The four terminal skills you will use the most: moving around without a GUI, SSH with an alias for you and your coding agent, tmux, and watching the CPU and the GPU.
Where the policy lives, which layers read it, what the router does and why the last block can overturn a verdict: a tour through the internals of four LLM guards, measured on the weights they actually run.
AI safetyLLM safeguardsInterpretabilityTransformersMixture of experts
I audited four candidate guards for an agentic assistant layer by layer. They all decide in one token and late in the network, and the two dedicated guards miss prompt injection hidden in documents. A small generalist catches almost everything if it writes its justification before the verdict.
AI safetyLLM safeguardsPrompt injectionInterpretabilityAgentic AI