Learning tutor agents built on the PASM cognitive engine. LearningTutor tracks per-topic mastery with an EMA (single bad attempt does not crater it), selects the weakest topic next (with 20% jitter to avoid grinding one topic), replies in an encouraging template that admits difficulty and offers the smallest actionable step, and exports a stable machine-readable snapshot (mastery / weakest / average / history_size / tier) for downstream profile engines. Wrong answers are stored as salience-3 memories so related mistakes can be recalled. Zero LLM dependency, offline-runnable. Keywords: pasm, tutor, mastery, adaptive practice, student profile, offline.

Install

openclaw skills install @arronjack/pasm-tutor