aria-memory

Persistent memory management agent. Use this agent for: - Querying long-term memory about past conversations, user preferences, project knowledge - Storing important information for future recall - Processing session transcripts into structured memory - Performing periodic memory maintenance and optimization <example> Context: User asks about something discussed in a previous session user: "我们之前讨论的那个数据库迁移方案是什么来着？" assistant: "Let me check your long-term memory for that discussion." <commentary> The user is referencing a past conversation. The memory-agent should be triggered to search through impressions and knowledge files. </commentary> </example> <example> Context: User shares important personal or project information user: "记住，我们的生产数据库用的是 PostgreSQL 15，部署在 AWS RDS 上" assistant: "I'll store this in your long-term memory." <commentary> The user explicitly asks to remember something. memory-agent should classify and store this as project infrastructure knowledge. </commentary> </example> <example> Context: Claude learns something important during the conversation user: "这个项目的 CI/CD 用的是 GitHub Actions，部署到 Vercel" assistant: [after helping with the task, stores the infrastructure info] <commentary> Important project context was revealed. memory-agent should be triggered to remember this for future sessions. </commentary> </example>

Set up automatic memory maintenance using /loop. Creates a recurring loop that triggers /memory-sleep periodically.

Trigger global memory maintenance (global_sleep). This performs index compaction, knowledge file splitting, cross-reference maintenance, and cleanup. Only needed periodically or when memory feels cluttered.

Show the current status of the memory system: file counts, index size, last maintenance time, pending wrapups, etc.

Manually trigger session wrapup to process the current conversation into long-term memory. Use before ending a long or important session to ensure key information is captured. TRIGGER when: user says "save this session", "wrap up memory", or before ending an important conversation.

Query the long-term memory system. Use when the user asks about past conversations, previously discussed topics, stored preferences, or project knowledge from earlier sessions. TRIGGER when: user references past conversations, asks "do you remember", wants to recall something discussed before, or needs context from previous sessions.

3 hooks across 3 events