Long-term verification agents for the PASM cognitive engine. Answers "is it still healthy after long runs?" with repeatable, stdlib-only checks: memory retention and salience-aware eviction, emotion drift, behavioural entropy collapse, persona saturation, forgetting curve, 6000-step soak degradation, and cross-repo parity. Scenarios drive the real engine in an isolated subprocess and emit JSON-archivable, baseline-diffable findings.

Install

openclaw skills install @arronjack/pasm-longterm-verify