Executive OS

Go from blank agent to working Chief of Staff in 10 minutes. This isn't a prompt template — it's a complete operating system distilled from weeks of real production use managing two email accounts, dual calendars, project tracking, and daily operations for a startup founder.

🧠 Self-Improving Agent The secret weapon: a structured self-review system where your agent logs mistakes and corrections permanently. It reads past failures on every boot and actively avoids repeating them. Most AI agents make the same errors forever — this one compounds.

📦 What's Included (11 files)

• SOUL.md — Personality, safety rules, hard boundaries, communication style
• AGENTS.md — Operating procedures, memory protocol, heartbeat system, group chat rules
• IDENTITY.md — Agent identity and role definition
• USER.md — Owner profile template (preferences, accounts, pain points)
• MEMORY.md — Two-tier memory architecture (curated long-term + raw daily logs)
• TOOLS.md — Local tool notes, account details, API configurations
• HEARTBEAT.md — Autonomous monitoring checklist (email, calendar, tasks)
• TODO.md — Task queue with priority levels, synced to external project management
• Self-review system — Mistake logging with tagged categories for pattern recognition
• Setup guide — Step-by-step installation with configuration walkthrough

⚡ Proven in Production:

• Dual email triage (Gmail + Outlook) with smart archiving and VIP protection
• Daily briefings with newsletter intelligence, calendar sync, and weather
• Automated meeting prep — researches attendees before external meetings
• Post-compaction recovery — survives context resets without losing state
• Heartbeat-driven monitoring that catches issues before you notice them
• Task management synced to Notion with priority-based surfacing

What makes Executive OS different from a generic EA prompt:

• Battle-tested patterns — not theory, but operational knowledge from hundreds of real interactions and edge cases
• Anti-pattern awareness — knows what NOT to do: never send external emails without approval, never archive documents, never guess which email account
• Self-correction loop — structured mistake logging means the agent improves permanently, not just per-session
• Safety-first design — hard rules for external communications, VIP sender protection, quiet hours respect
• Proactive, not reactive — heartbeat system checks email, calendar, and tasks autonomously throughout the day

Designed for OpenClaw. Works with Claude Opus, Sonnet, or any supported model.