remember_skill

Overview

This skill curates persistent memory that actually helps by filtering what matters, organizing entries by function, and decaying what doesn't. It focuses on retrieving the right memory at the right time rather than storing everything. The result is less noise, faster relevance, and safer long-term recall.

How this skill works

It inspects incoming user signals and stores only high- and medium-value items using functional categories like commitments, preferences, corrections, decisions, relationships, and per-context state. Each memory entry includes a date, source hint, and confidence level. The system logs contradictions instead of silently overwriting and applies pruning rules to archive or remove stale data.

When to use it

• Capture commitments, explicit preferences, corrections, and core relationships that should persist.
• Store project or client context that will be revisited and benefits from continuity.
• Avoid saving one-off questions, easily reconstructible facts, or transient conversational context.
• Run periodic reviews to consolidate, archive, or delete medium-value memories.
• Respond to user commands like “remember this”, “forget that”, or “what do you know about X?”.

Best practices

• Organize memories by function (commitments, preferences, corrections, decisions, relationships, contexts) to match retrieval patterns.
• Record metadata on every entry: recorded date, source hint (explicit vs inferred), and confidence level.
• Prune aggressively: archive completed commitments after 30 days and move inactive contexts to archive.
• When contradictions occur, log the change and keep the newer version active rather than deleting history.
• Apply a staleness test: if a memory would mislead in six months, do not persist it.

Example use cases

• Remembering a user’s preferred response length and tone so future replies match expectations.
• Tracking delivery promises and deadlines so the assistant can remind the user or follow up.
• Recording corrections like “do not share this format” to avoid repeating past mistakes.
• Maintaining active project state files for ongoing work and archiving them when inactive.
• Answering transparency queries such as “what do you know about my preferences?” on demand.