dsh-ai-teacher

Many learners facing a dense PDF do not want an AI that merely recites what the text states. They want a guide who treats the material as fact, walks through each derivation one step at a time, pinpoints the exact spot where reasoning breaks down, and prompts them to form a hypothesis before revealing the answer. dsh-ai-teacher is a DeepSeek Harness agent preset designed for precisely this role.

Once installed, AI-teacher appears in the new-session preset picker. It reads PDFs through the built-in read_pdf tool of the host, digests the content, and explains it as established truth rather than as something a book once wrote. Every claim is anchored to the source document, a jointly derived intermediate step, or a widely accepted definition. When uncertainty arises, the agent verifies before proceeding, never inventing content. Pacing is deliberately slow: one formula, one equality, or one logical step per message, closed with a confirmation question, then a full stop. Socratic guidance runs throughout the session — guess before prove, reverse-engineer the result, ask meta-cognitive follow-ups. A quiet motivational layer sustains momentum through task chunking, immediate feedback, and visible progress tracking, with no gamified interface in sight. Teaching methods are decomposed into four on-demand skills (new-concept teaching, step-by-step derivation, confusion triage, and a dedicated Frankel geometry-physics drill), which keeps the always-on persona lean while ensuring reliable trigger matching. The preset also ships with the full standard Harness capability set: shell, filesystem, search, plan mode, goals, subagents, and workflows.

Ideal for graduate students or self-directed learners working through derivation-heavy texts, especially the Geometry of Physics by Frankel. Also a strong fit for any DeepSeek Harness user who wants the AI to shift from passive recitation to active, step-by-step co-derivation. No extra dependencies required; the preset is ready to use immediately after installation.