What problem does it solve? Conversational agents forget everything between sessions, forcing users to repeat their identity, preferences, rules, and project context. This Skill evaluates each conversation turn and persists only the facts worth remembering into mem0 long-term memory. ## Core Features & Use Cases - Four-Gate Triage: Every candidate fact must pass future-utility, novelty, factual, and safety gates before storage, so most turns correctly produce zero memory operations. - Categorized Storage: Facts are stored via memory_add under categories (identity, preference, decision, rule, project, configuration, technical, relationship) that control retention policy. - Recall Protocol: A companion protocol defines how to rewrite user messages into 3-6 keyword search queries and apply time/category filters for memory_search. - Credential Safety: Secrets and tokens are never stored; only the fact that a credential was configured is recorded. - Use Case: A user says "I'm Sarah, I work at Cloudflare, and we just switched monitoring from Datadog to Grafana because of cost." The Skill stores one identity fact and one decision fact in separate categorized calls, then recalls them in future sessions. ## Quick Start Ask the agent to remember that you prefer terse responses with no trailing summaries, and it will store that preference in long-term memory for future sessions.
