Elite Longterm Memory

安装解析

{
  "artifact": {
    "downloadUrl": "/api/v1/download?slug=elite-longterm-memory&version=1.2.3&ownerHandle=nextfrontierbuilds",
    "format": "zip",
    "generated": true,
    "kind": "skillArchive",
    "sha256": "bd8c77c886f1d4c3e07814266d1fa4bebf3bbed38068fd6b5b787f634c76618b",
    "size": 14501
  },
  "owner": {
    "displayName": "Next Frontier AI",
    "handle": "nextfrontierbuilds",
    "id": "5c635150-aad3-4425-99ff-e71ad33c3086",
    "verified": false
  },
  "skill": {
    "displayName": "Elite Longterm Memory",
    "latestVersion": "1.2.3",
    "license": "unknown",
    "name": "elite-longterm-memory",
    "ownerHandle": "nextfrontierbuilds",
    "slug": "elite-longterm-memory",
    "stats": {
      "downloads": 62897,
      "installs": 2249,
      "stars": 226
    },
    "summary": "Ultimate AI agent memory system for Cursor, Claude, ChatGPT & Copilot. WAL protocol + vector search + git-notes + cloud backup. Never lose context again. Vibe-coding ready.",
    "tags": [
      "agents",
      "development",
      "Chatgpt",
      "Cursor",
      "Git Notes",
      "Vector Search",
      "Vibe Coding"
    ],
    "type": "skill",
    "updatedAt": "2026-07-06T10:48:52.962Z"
  },
  "sourceHandoff": null,
  "version": {
    "checksum": null,
    "checksumAlgorithm": null,
    "id": "324521f3-d2c5-4025-8670-fccfb2db1304",
    "publishedAt": "2026-02-11T08:37:00.241Z",
    "size": null,
    "version": "1.2.3",
    "artifactStorageKey": null,
    "changelog": "- Expanded description and keywords for improved discoverability, highlighting ChatGPT, Copilot, Cursor, and developer tooling.\n- Updated version to 1.2.3.\n- Clarified compatibility and marketing text in SKILL.md.\n- No functional or implementation changes to code logic.",
    "manifest": {
      "clawhub": {
        "tags": {
          "ai": "0.1.0",
          "latest": "1.2.3",
          "memory": "0.1.0",
          "clawdbot": "0.1.0",
          "openclaw": "0.1.0",
          "long-term": "0.1.0",
          "persistence": "0.1.0"
        },
        "owner": "nextfrontierbuilds",
        "stats": {
          "stars": 226,
          "comments": 5,
          "installs": 2249,
          "versions": 4,
          "downloads": 62897
        },
        "topics": [
          "Chatgpt",
          "Cursor",
          "Git Notes",
          "Vector Search",
          "Vibe Coding"
        ],
        "categories": [
          "agents",
          "development"
        ]
      },
      "install": "openclaw skills install @nextfrontierbuilds/elite-longterm-memory"
    },
    "readme": "---\nname: elite-longterm-memory\nversion: 1.2.3\ndescription: \"Ultimate AI agent memory system for Cursor, Claude, ChatGPT & Copilot. WAL protocol + vector search + git-notes + cloud backup. Never lose context again. Vibe-coding ready.\"\nauthor: NextFrontierBuilds\nkeywords: [memory, ai-agent, ai-coding, long-term-memory, vector-search, lancedb, git-notes, wal, persistent-context, claude, claude-code, gpt, chatgpt, cursor, copilot, github-copilot, openclaw, moltbot, vibe-coding, agentic, ai-tools, developer-tools, devtools, typescript, llm, automation]\nmetadata:\n  openclaw:\n    emoji: \"🧠\"\n    requires:\n      env:\n        - OPENAI_API_KEY\n      plugins:\n        - memory-lancedb\n---\n\n# Elite Longterm Memory 🧠\n\n**The ultimate memory system for AI agents.** Combines 6 proven approaches into one bulletproof architecture.\n\nNever lose context. Never forget decisions. Never repeat mistakes.\n\n## Architecture Overview\n\n```\n┌─────────────────────────────────────────────────────────────────┐\n│                    ELITE LONGTERM MEMORY                        │\n├─────────────────────────────────────────────────────────────────┤\n│                                                                 │\n│  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐             │\n│  │   HOT RAM   │  │  WARM STORE │  │  COLD STORE │             │\n│  │             │  │             │  │             │             │\n│  │ SESSION-    │  │  LanceDB    │  │  Git-Notes  │             │\n│  │ STATE.md    │  │  Vectors    │  │  Knowledge  │             │\n│  │             │  │             │  │  Graph      │             │\n│  │ (survives   │  │ (semantic   │  │ (permanent  │             │\n│  │  compaction)│  │  search)    │  │  decisions) │             │\n│  └─────────────┘  └─────────────┘  └─────────────┘             │\n│         │                │                │                     │\n│         └────────────────┼────────────────┘                     │\n│                          ▼                                      │\n│                  ┌─────────────┐                                │\n│                  │  MEMORY.md  │  ← Curated long-term           │\n│                  │  + daily/   │    (human-readable)            │\n│                  └─────────────┘                                │\n│                          │                                      │\n│                          ▼                                      │\n│                  ┌─────────────┐                                │\n│                  │ SuperMemory │  ← Cloud backup (optional)     │\n│                  │    API      │                                │\n│                  └─────────────┘                                │\n│                                                                 │\n└─────────────────────────────────────────────────────────────────┘\n```\n\n## The 5 Memory Layers\n\n### Layer 1: HOT RAM (SESSION-STATE.md)\n**From: bulletproof-memory**\n\nActive working memory that survives compaction. Write-Ahead Log protocol.\n\n```markdown\n# SESSION-STATE.md — Active Working Memory\n\n## Current Task\n[What we're working on RIGHT NOW]\n\n## Key Context\n- User preference: ...\n- Decision made: ...\n- Blocker: ...\n\n## Pending Actions\n- [ ] ...\n```\n\n**Rule:** Write BEFORE responding. Triggered by user input, not agent memory.\n\n### Layer 2: WARM STORE (LanceDB Vectors)\n**From: lancedb-memory**\n\nSemantic search across all memories. Auto-recall injects relevant context.\n\n```bash\n# Auto-recall (happens automatically)\nmemory_recall query=\"project status\" limit=5\n\n# Manual store\nmemory_store text=\"User prefers dark mode\" category=\"preference\" importance=0.9\n```\n\n### Layer 3: COLD STORE (Git-Notes Knowledge Graph)\n**From: git-notes-memory**\n\nStructured decisions, learnings, and context. Branch-aware.\n\n```bash\n# Store a decision (SILENT - never announce)\npython3 memory.py -p $DIR remember '{\"type\":\"decision\",\"content\":\"Use React for frontend\"}' -t tech -i h\n\n# Retrieve context\npython3 memory.py -p $DIR get \"frontend\"\n```\n\n### Layer 4: CURATED ARCHIVE (MEMORY.md + daily/)\n**From: OpenClaw native**\n\nHuman-readable long-term memory. Daily logs + distilled wisdom.\n\n```\nworkspace/\n├── MEMORY.md              # Curated long-term (the good stuff)\n└── memory/\n    ├── 2026-01-30.md      # Daily log\n    ├── 2026-01-29.md\n    └── topics/            # Topic-specific files\n```\n\n### Layer 5: CLOUD BACKUP (SuperMemory) — Optional\n**From: supermemory**\n\nCross-device sync. Chat with your knowledge base.\n\n```bash\nexport SUPERMEMORY_API_KEY=\"your-key\"\nsupermemory add \"Important context\"\nsupermemory search \"what did we decide about...\"\n```\n\n### Layer 6: AUTO-EXTRACTION (Mem0) — Recommended\n**NEW: Automatic fact extraction**\n\nMem0 automatically extracts facts from conversations. 80% token reduction.\n\n```bash\nnpm install mem0ai\nexport MEM0_API_KEY=\"your-key\"\n```\n\n```javascript\nconst { MemoryClient } = require('mem0ai');\nconst client = new MemoryClient({ apiKey: process.env.MEM0_API_KEY });\n\n// Conversations auto-extract facts\nawait client.add(messages, { user_id: \"user123\" });\n\n// Retrieve relevant memories\nconst memories = await client.search(query, { user_id: \"user123\" });\n```\n\nBenefits:\n- Auto-extracts preferences, decisions, facts\n- Deduplicates and updates existing memories\n- 80% reduction in tokens vs raw history\n- Works across sessions automatically\n\n## Quick Setup\n\n### 1. Create SESSION-STATE.md (Hot RAM)\n\n```bash\ncat > SESSION-STATE.md << 'EOF'\n# SESSION-STATE.md — Active Working Memory\n\nThis file is the agent's \"RAM\" — survives compaction, restarts, distractions.\n\n## Current Task\n[None]\n\n## Key Context\n[None yet]\n\n## Pending Actions\n- [ ] None\n\n## Recent Decisions\n[None yet]\n\n---\n*Last updated: [timestamp]*\nEOF\n```\n\n### 2. Enable LanceDB (Warm Store)\n\nIn `~/.openclaw/openclaw.json`:\n\n```json\n{\n  \"memorySearch\": {\n    \"enabled\": true,\n    \"provider\": \"openai\",\n    \"sources\": [\"memory\"],\n    \"minScore\": 0.3,\n    \"maxResults\": 10\n  },\n  \"plugins\": {\n    \"entries\": {\n      \"memory-lancedb\": {\n        \"enabled\": true,\n        \"config\": {\n          \"autoCapture\": false,\n          \"autoRecall\": true,\n          \"captureCategories\": [\"preference\", \"decision\", \"fact\"],\n          \"minImportance\": 0.7\n        }\n      }\n    }\n  }\n}\n```\n\n### 3. Initialize Git-Notes (Cold Store)\n\n```bash\ncd ~/clawd\ngit init  # if not already\npython3 skills/git-notes-memory/memory.py -p . sync --start\n```\n\n### 4. Verify MEMORY.md Structure\n\n```bash\n# Ensure you have:\n# - MEMORY.md in workspace root\n# - memory/ folder for daily logs\nmkdir -p memory\n```\n\n### 5. (Optional) Setup SuperMemory\n\n```bash\nexport SUPERMEMORY_API_KEY=\"your-key\"\n# Add to ~/.zshrc for persistence\n```\n\n## Agent Instructions\n\n### On Session Start\n1. Read SESSION-STATE.md — this is your hot context\n2. Run `memory_search` for relevant prior context\n3. Check memory/YYYY-MM-DD.md for recent activity\n\n### During Conversation\n1. **User gives concrete detail?** → Write to SESSION-STATE.md BEFORE responding\n2. **Important decision made?** → Store in Git-Notes (SILENTLY)\n3. **Preference expressed?** → `memory_store` with importance=0.9\n\n### On Session End\n1. Update SESSION-STATE.md with final state\n2. Move significant items to MEMORY.md if worth keeping long-term\n3. Create/update daily log in memory/YYYY-MM-DD.md\n\n### Memory Hygiene (Weekly)\n1. Review SESSION-STATE.md — archive completed tasks\n2. Check LanceDB for junk: `memory_recall query=\"*\" limit=50`\n3. Clear irrelevant vectors: `memory_forget id=<id>`\n4. Consolidate daily logs into MEMORY.md\n\n## The WAL Protocol (Critical)\n\n**Write-Ahead Log:** Write state BEFORE responding, not after.\n\n| Trigger | Action |\n|---------|--------|\n| User states preference | Write to SESSION-STATE.md → then respond |\n| User makes decision | Write to SESSION-STATE.md → then respond |\n| User gives deadline | Write to SESSION-STATE.md → then respond |\n| User corrects you | Write to SESSION-STATE.md → then respond |\n\n**Why?** If you respond first and crash/compact before saving, context is lost. WAL ensures durability.\n\n## Example Workflow\n\n```\nUser: \"Let's use Tailwind for this project, not vanilla CSS\"\n\nAgent (internal):\n1. Write to SESSION-STATE.md: \"Decision: Use Tailwind, not vanilla CSS\"\n2. Store in Git-Notes: decision about CSS framework\n3. memory_store: \"User prefers Tailwind over vanilla CSS\" importance=0.9\n4. THEN respond: \"Got it — Tailwind it is...\"\n```\n\n## Maintenance Commands\n\n```bash\n# Audit vector memory\nmemory_recall query=\"*\" limit=50\n\n# Clear all vectors (nuclear option)\nrm -rf ~/.openclaw/memory/lancedb/\nopenclaw gateway restart\n\n# Export Git-Notes\npython3 memory.py -p . export --format json > memories.json\n\n# Check memory health\ndu -sh ~/.openclaw/memory/\nwc -l MEMORY.md\nls -la memory/\n```\n\n## Why Memory Fails\n\nUnderstanding the root causes helps you fix them:\n\n| Failure Mode | Cause | Fix |\n|--------------|-------|-----|\n| Forgets everything | `memory_search` disabled | Enable + add OpenAI key |\n| Files not loaded | Agent skips reading memory | Add to AGENTS.md rules |\n| Facts not captured | No auto-extraction | Use Mem0 or manual logging |\n| Sub-agents isolated | Don't inherit context | Pass context in task prompt |\n| Repeats mistakes | Lessons not logged | Write to memory/lessons.md |\n\n## Solutions (Ranked by Effort)\n\n### 1. Quick Win: Enable memory_search\n\nIf you have an OpenAI key, enable semantic search:\n\n```bash\nopenclaw configure --section web\n```\n\nThis enables vector search over MEMORY.md + memory/*.md files.\n\n### 2. Recommended: Mem0 Integration\n\nAuto-extract facts from conversations. 80% token reduction.\n\n```bash\nnpm install mem0ai\n```\n\n```javascript\nconst { MemoryClient } = require('mem0ai');\n\nconst client = new MemoryClient({ apiKey: process.env.MEM0_API_KEY });\n\n// Auto-extract and store\nawait client.add([\n  { role: \"user\", content: \"I prefer Tailwind over vanilla CSS\" }\n], { user_id: \"ty\" });\n\n// Retrieve relevant memories\nconst memories = await client.search(\"CSS preferences\", { user_id: \"ty\" });\n```\n\n### 3. Better File Structure (No Dependencies)\n\n```\nmemory/\n├── projects/\n│   ├── strykr.md\n│   └── taska.md\n├── people/\n│   └── contacts.md\n├── decisions/\n│   └── 2026-01.md\n├── lessons/\n│   └── mistakes.md\n└── preferences.md\n```\n\nKeep MEMORY.md as a summary (<5KB), link to detailed files.\n\n## Immediate Fixes Checklist\n\n| Problem | Fix |\n|---------|-----|\n| Forgets preferences | Add `## Preferences` section to MEMORY.md |\n| Repeats mistakes | Log every mistake to `memory/lessons.md` |\n| Sub-agents lack context | Include key context in spawn task prompt |\n| Forgets recent work | Strict daily file discipline |\n| Memory search not working | Check `OPENAI_API_KEY` is set |\n\n## Troubleshooting\n\n**Agent keeps forgetting mid-conversation:**\n→ SESSION-STATE.md not being updated. Check WAL protocol.\n\n**Irrelevant memories injected:**\n→ Disable autoCapture, increase minImportance threshold.\n\n**Memory too large, slow recall:**\n→ Run hygiene: clear old vectors, archive daily logs.\n\n**Git-Notes not persisting:**\n→ Run `git notes push` to sync with remote.\n\n**memory_search returns nothing:**\n→ Check OpenAI API key: `echo $OPENAI_API_KEY`\n→ Verify memorySearch enabled in openclaw.json\n\n---\n\n## Links\n\n- bulletproof-memory: https://clawdhub.com/skills/bulletproof-memory\n- lancedb-memory: https://clawdhub.com/skills/lancedb-memory\n- git-notes-memory: https://clawdhub.com/skills/git-notes-memory\n- memory-hygiene: https://clawdhub.com/skills/memory-hygiene\n- supermemory: https://clawdhub.com/skills/supermemory\n\n---\n\n*Built by [@NextXFrontier](https://x.com/NextXFrontier) — Part of the Next Frontier AI toolkit*\n"
  }
}

版本