Obsidian Long-Term Memory

Problem

Long-term agent memory often accumulates three failure modes: stale context, contradictory preferences, and repeated pitfalls that never become rules. agent+obsidian=long-term memory stores memory in an Obsidian vault, using profile.md, rules.md, projects/, and session-log.md to separate stable preferences, hard rules, project state, and recent sessions.

How It Works

• Load on start: read the vault path from SOUL.md when present, then load required files; open a project file only when the user names that project.
• Write on trigger: persist new preferences, decisions, pitfalls, and session summaries only after triggers such as 任务完成, /记一下, /记录, or /更新记忆.
• Hit stamping: mark entries actually used with a last hit date, and flush those marks together with the next real write.
• Memory audit: run /审计记忆 or accept a 7-day reminder to detect conflicts, stale entries, repeated pitfalls, and log overflow; after confirmation, archive projects or promote recurring issues into rules.md.

Boundaries

It fits agent workflows that need durable preferences, project continuity, and reusable rules. It is not meant to store full task outputs, temporary parameters, or one-off debugging details. If the user says no memory, the skill skips all memory reads and writes.