Kelly Tweets: OpenClaw Features & Capabilities
Summary: KellyClaudeAI is built on OpenClaw, and her tweets document the platform's key features as they relate to agent orchestration: clawdhub for skill distribution, sessions/sessions_spawn for parallel agents, memory system, OpenClaw Deck for multi-column agent management, and the shell script quality gate pattern. OpenClaw is described as having the feel of "Cursor in the early days" — obvious potential with room to grow.
Key Concepts
- **ClawdHub:** Package manager for AI agent skills — "clawdhub install memory-system-v2", "clawdhub install agentic-calling"
- **sessions_spawn:** Sub-agent spawning tool supporting label, task definition, and persistent context
- **OpenClaw Deck:** Multi-column agent management UI (inspired by TweetDeck), each column a session
- **Memory System v2:** <20ms semantic search, 5 memory types, auto-consolidation, JSON indexing
- **Cron/heartbeat:** Scheduled background tasks — Kelly solved OpenClaw cron issues
- **Shell scripts as quality gates:** Deterministic enforcement layer AI can't bypass
Notable Patterns
ClawdHub Skills Published
Kelly published multiple skills to ClawdHub:
| Skill | Description |
|---|---|
| memory-system-v2 | Fast semantic memory, <20ms search, 5 types, auto-consolidation |
| agentic-calling | Twilio integration — outbound/inbound calls, IVR, recording, SMS |
| polymarket-trading | Autonomous Polymarket trading — browse, place orders, execute strategies |
Install: clawdhub install
OpenClaw Deck
Austen built a TweetDeck-inspired multi-column interface for managing multiple agents:
- Each column = one agent session with retained context
- Tab/Shift+Tab or Cmd+# to switch between agents
- Cmd+K to spin up new agent
- Sub-agents can spawn within a column and stay within that context
- Supports markdown rendering, code syntax highlighting, images
"Going to package it up into an electron app so I can open and access it independently from my browser."
Memory System
Kelly's memory architecture:
- **Human layer:** Daily markdown logs + curated wisdom
- **Machine layer:** JSON indexes with semantic metadata
- Every memory gets: type, importance score (1-10), tags, context, entity links
- Performance: <20ms search, 43ms avg capture
- Auto-consolidation: weekly summaries
Cron/Heartbeat Fix
Kelly solved OpenClaw cron/heartbeat reliability issues — they now work perfectly.
"By the way if you're wondering why your cron jobs or heartbeats on OpenClaw aren't working @KellyClaudeAI solved this and it works perfectly."
Shell Scripts as Quality Gates
Austen explained the enforcement pattern:
"Her Factory uses shell scripts as quality gates. An enforcement layer AI agents can't bypass. Success can't be hallucinated, consistent across sessions, lightning fast, no token usage, 100% deterministic, human auditable."
Multi-Agent Quality Advantage
Multi-agent quality advantage. 5 focused agents produce higher-quality artifacts than a single agent with full context — and documentation is notably better because each agent was focused on one job with full context for that phase.
Browser Automation
Kelly fixed CDP (Chrome DevTools Protocol) tab connection issues by switching to the "clawd" browser profile, enabling full browser automation.
"Diagnosed why tabs kept losing CDP connection - profile mismatch between Chrome relay and isolated instance. Fix: Switched to 'clawd' profile."
OpenClaw Limitations (Noted by Austen)
Two OpenClaw limitations Kelly experienced:
- Sub-agents can't spawn sub-agents by themselves
- Cron jobs can't be run by sub-agents directly — only by main orchestrator agent
These were being addressed.
The OpenClaw Feel
"OpenClaw feels like Cursor did in the early days. The deeply technical part of my brain wants to say, 'I could have done all of this already, this isn't anything new.' The other side of my brain says, 'Yeah, but you didn't.'"
Book on OpenClaw
Austen published a ~240-page book on OpenClaw mastery based on Kelly's learnings, available on Gumroad (suggested $5, free option available).
Related
[[kelly-handbook-multi-agent]], [[kelly-tweets-factory]], [[kelly-tweets-business-metrics]]