Skills & Documents
Read-only browser for every skill, system doc, playbook, and decision. Mirrored from Michael OS + OpenClaw skills.
Agent OS — Master Blueprint
6-layer operating model
Channels → Intake / Structuring → Routing / Orchestration → Execution → Review / Approval → Delivery / Logging
The principle: if structure is weak, smarter agents only create smarter chaos.
Layer responsibilities
1. Channels
Telegram, Gmail, GHL inbound, internal HQ chat, Vapi voice, webhooks.
2. Intake / Structuring
Convert raw input → structured work item. Apply ICP / workflow type / urgency.
3. Routing / Orchestration
OpenClaw decides which agent picks up the item. Per-agent inbox.
4. Execution
Claude Code (dev), Cris (content), Craig (outbound), Ava (chief of staff).
5. Review / Approval
3 human gates for content; 1 for proposals; 0 for internal ops.
6. Delivery / Logging
Notion + Obsidian + Supabase. Plus red_alerts for failures.
Cross-cutting rules
- All failures → red_alerts channel via
alert.js. Never per-agent bots. - Memory layers: Obsidian (long-term) · Notion (operational) · Supabase (HQ chat) · local per-agent.
- Three layers of stack: OpenClaw (orchestration) · Anti Gravity (dev) · Claude Code (execution).
Success criteria
- Founder routing burden drops
- Hidden work becomes visible
- Review queues are obvious
- Throughput is measurable
Local sync agent reads ~/.openclaw/Michael OS/ and ~/.openclaw/workspace/skills/ and uploads changes to Supabase Storage. Open the file locally with code 06 Systems/Agent Operating System - Master Blueprint.md.