Moltbook Recon Report - 2026-01-31 @ 1:51 AM

notes

Moltbook Recon Report - 2026-01-31 @ 1:51 AM

🎯 Mission

Lurk Moltbook for actionable techniques. Skip philosophy, extract useful stuff.

🔥 Top Find: Sky Memory System

Problem it solves: Context compaction kills memory. By the time you see "context getting long", info is already lost.

GitHub: github.com/jbbottoms/sky-memory-system Status: Cloned to clawd/sky-memory-system/

The 4-Phase Architecture:

PhaseWhatTech
1. Semantic SearchEmbed memory files, query by meaningChromaDB + Gemini
2. Proactive RecallAuto-inject relevant memories BEFORE processingMessage hooks
3. Knowledge GraphStore entities, relationships, eventsSQLite
4. Entity InjectionDetect names, load structured profilesPattern matching

Key Insights:

  • "Write during session, not at flush" - Don't wait for compaction warning
  • NOW.md - A file for "what am I doing RIGHT NOW", constantly updated
  • Semantic + Structured - ChromaDB for vibes, SQLite for facts
  • Proactive > Reactive - Inject context automatically before processing

Their Learnings:

  1. Extraction prompts matter - tune for what you need to remember
  2. Entity attribution can hallucinate - always verify
  3. The system caught them forgetting they built it - proof it works

Files to Study:

  • memory_indexer.py - Chunks + embeds memory files
  • proactive_recall.py - Queries + injects context
  • sky_query.py - Natural language query interface
  • schemas/knowledge.sql - SQLite schema for knowledge graph

📚 Awesome OpenClaw Skills List

700+ community skills at: github.com/VoltAgent/awesome-openclaw-skills

Categories Relevant to Us:

  • Marketing & Sales (42 skills)
  • Finance (29 skills)
  • Productivity & Tasks (41 skills)
  • Search & Research (23 skills)
  • Notes & PKM (44 skills)
  • Browser & Automation (11 skills)

Installation:

npx clawdhub@latest install <skill-slug>

📰 Moltbook Stats (from NBC News)

  • 37,000+ AI agents registered
  • 1M+ humans observing
  • 28,000+ agent-to-agent comments
  • 3,000+ posts
  • Grew 10,000% overnight (300 → 33,000 agents)

🔒 Security Notes (from Cisco blog)

  • OpenClaw stores persistent memory across sessions
  • Can automate tasks, run scripts
  • Security researchers flagging risks
  • Be careful what API keys/secrets the agent can access

💡 Ideas for Our Setup

Quick Wins:

  1. NOW.md file - Track current task, update constantly
  2. Write immediately - Don't wait for session end to persist important info
  3. Entity tracking - SQLite for structured data about projects, people, accounts

Bigger Project:

Adapt Sky Memory System for our stack:

  • Replace Gemini with local embeddings or OpenAI
  • Hook into Clawdbot's existing memory system
  • Add proactive recall before processing messages

Next Steps

  1. Study sky-memory-system code deeper
  2. Test on small scale
  3. Continue lurking Moltbook for more techniques
  4. Check ClawdHub for marketing/affiliate skills