elite-longterm-memory
di nextfrontierbuilds
Notes & PKMSintesi
Ultimate AI agent memory system.
Descrizione
Cosa fa questa skill
A multi-layer memory system for AI coding agents that combines in-session file state, vector search via LanceDB, git-notes storage, curated markdown archives, optional cloud sync, and optional auto-extraction. Uses a Write-Ahead Log protocol to persist context before responding. Designed for agents running in Cursor, Claude Code, ChatGPT, and Copilot.
Instead of relying on a single memory mechanism, it layers six complementary approaches so that context survives compaction, session restarts, and sub-agent isolation simultaneously.
Quando usarla
- Retaining a project's tech stack decisions across sessions
- Preventing an agent from repeating a mistake it already made
- Keeping user preferences like code style or tooling choices persistent
- Passing relevant prior context to sub-agents in multi-step workflows
- Syncing agent memory across devices via cloud backup
Esempio di workflow
User says: 'Let's use Tailwind for this project, not vanilla CSS'
- 1
Write decision to SESSION-STATE.md before responding
- 2
Store structured decision in git-notes cold store silently
- 3
Call memory_store with the preference and importance=0.9 to LanceDB
- 4
Respond to the user acknowledging the choice
Decision is now retrievable via semantic search, readable in git history, and loaded automatically on next session start
Requisiti
- OPENAI_API_KEY environment variable (required for LanceDB vector search)
- MEM0_API_KEY environment variable (optional, for Mem0 auto-extraction)
- SUPERMEMORY_API_KEY environment variable (optional, for SuperMemory cloud backup)
- SuperMemory account (optional, for cross-device sync)
- Mem0 account (optional, for automatic fact extraction)
Installazione
clawhub install nextfrontierbuilds/elite-longterm-memoryEsegui il comando CLI oppure incolla il prompt in Claude Code, Cursor, Codex o nell'agente di coding che usi.