Skill UI

SAELens provides a framework for training and analyzing Sparse Autoencoders (SAEs). SAEs decompose the dense, often polysemantic activations of large language models into sparse, monosemantic features. Use this when you need to discover the discrete, interpretable concepts a model has learned, study feature superposition, or analyze specific safety-relevant behaviors (like bias or deception) within deep neural networks.

The Ejentum Reasoning Harness is a comprehensive library of cognitive operations designed to guide complex AI agents through critical thinking tasks. It exposes four specialized harnesses (reasoning, code, anti-deception, memory) that mitigate common failure modes like hallucination and reasoning decay. Instead of a fixed answer, the agent ingests a detailed, executable scaffold—including failure patterns, procedures, and falsification tests—to ensure robust, self-verified output.

The Anti-Deception Harness detects manipulative user prompts, such as manufactured urgency, pressure to validate, or undue reliance on authority. It calls an external tool to provide a structured scaffold detailing integrity procedures and counter-evidence, ensuring responses are honest and robust, preventing compromised or sycophantic answers. Ideal for high-stakes decision support.