AgentSkillsCN

store

将重要事实、决策或偏好存储为持久记忆

SKILL.md
--- frontmatter
name: store
description: Store important facts, decisions, or preferences as persistent memories
user_invocable: true
arguments:
  - name: content
    description: Optional content to store. If omitted, extract key facts from the conversation.
    required: false

/openmem:store

Store memories from the current conversation into persistent memory.

Instructions

  1. •

    Determine what to store:

    • •If the user provided $ARGUMENTS content, store that directly.
    • •If no content was provided, analyze the current conversation and extract the most important:
      • •Decisions made (type: decision)
      • •Preferences expressed (type: preference)
      • •Key facts established (type: fact)
      • •Constraints identified (type: constraint)
      • •Plans discussed (type: plan)
      • •Incidents or issues encountered (type: incident)
  2. •

    For each item to store, call the memory_store MCP tool with:

    • •text: The full content of the memory
    • •type: One of fact, decision, preference, incident, plan, constraint
    • •entities: A list of key entities (project names, people, technologies, etc.)
    • •confidence: How certain we are (0-1, default 1.0)
    • •gist: A short one-line summary for quick retrieval
  3. •

    If you identify relationships between new memories or between new and existing memories, use memory_link to connect them.

  4. •

    Report back to the user what was stored, showing:

    • •The memory text (abbreviated if long)
    • •The type and entities
    • •The memory ID (so they can reference it later)
  5. •

    If you think the new memory supersedes or contradicts an existing one, mention this and offer to use memory_supersede or memory_contradict to manage the relationship.