memory-management

Create a small always-visible index for high-frequency facts and store detailed domain knowledge in focused files loaded on demand. Record each entry with its scope, source evidence, date, and review condition, and follow the environment's documented discovery rules such as AGENTS.md or a memory directory.

Store stable terminology, confirmed architecture and conventions, explicit team preferences, dated decisions with rationale, recurring commands, and active constraints. Never store secrets, credentials, unnecessary personal data, speculation, or temporary task state.

Update the authoritative entry instead of appending contradictions, and preserve decision history only when rationale still matters. Remove or archive entries that are obsolete, duplicated, or unverifiable, and verify drift-prone facts before consequential work.

No. Memory should only be added with explicit user authorization, and inferred personal data must not be saved. The Skill requires reporting what was added, changed, moved, or removed after any maintenance.

File-based memory can drift, duplicate facts across layers, and persist unauthorized data if not audited. When the host environment controls memory through a dedicated API or policy, that mechanism should be followed instead of writing arbitrary files.