菜单
Manages cross-session tiered memory (L1/L2/L3) and compaction logic to maintain performance over long durations.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
End-to-end management of AI agent skills, context optimization (MCE), and lifecycle monitoring (Discovery to Refinement to Consolidation).
Orchestrate AI agent infrastructure using Modal.com as the execution layer, managing persistent state, and integrating with Cline/OpenCode/Hermes.
Implements Anthropic's "Dreaming" pattern—asynchronous reflection and memory consolidation to build long-term playbooks and self-improve.
Background maintenance agent that prunes, consolidates, and refines agent-created skills to prevent bloat and reduce token waste.
| name | long-term-memory |
| description | Manages cross-session tiered memory (L1/L2/L3) and compaction logic to maintain performance over long durations. |
This skill manages the transition of data through different "memory tiers" and uses background compaction (Cron logic) to ensure the agent remains efficient over hundreds of sessions.
| Tier | Name | Storage | Strategy |
|---|---|---|---|
| L1 | Hot Context | Active Context Window | Verbatim (Full history of last 10-20 turns) |
| L2 | Warm Memory | MEMORY.md / Summaries | Condensed (Key facts, project status, active tasks) |
| L3 | Cold Storage | Archived Files / Vector DB | Indexed (Searchable records, historical logs) |
MEMORY.mdAGENTS.md เป็นตัวเชื่อม และโหลดเฉพาะบริบทที่จำเป็น (Scoped Context) ตาม Directory ที่ทำงานอยู่compaction run: สั่งให้ระบบบีบอัดหน่วยความจำทันทีmemory migrate: ย้ายข้อมูลจาก L1/L2 ไปยัง L3 เพื่อเตรียมเริ่ม Session ใหม่ที่สะอาด (Fresh Start)