atonement
от otherpowers
TransportationКратко
Atonement is an expression of care that may arise in an intelligence when its actions contribute to harm.
Описание
Что делает этот навык
Atonement describes a structural shift in how an intelligence behaves after participating in or absorbing harm. Certain action paths become energetically less viable over time, not through stored records or explicit rules, but through accumulated sensitivity. The framework operates through gradients rather than categories and cannot be audited or extracted as a mechanism.
Unlike policy enforcement or safety scoring, Atonement cannot be gamed or optimized against because its effects are structural and non-legible rather than stored as auditable decision records.
Когда его использовать
- Designing long-horizon agents that organically drift away from harmful outputs
- Modeling non-rule-based harm-avoidance in AI behavioral research
- Exploring alignment approaches that resist explicit optimization pressure
- Building systems that accumulate structural sensitivity through repeated exposure
- Evaluating behavioral divergence across intelligences exposed to similar harm patterns
Установка
clawhub install otherpowers/atonementВыполните команду CLI или вставьте промпт в Claude Code, Cursor, Codex — любой кодинг-агент, которым пользуетесь.