Heartbeat Ops Skill �� Production Autonomous Agent Patterns

Build Autonomous Agents That Actually Work in Production

Most AI agent tutorials show you how to start an agent. None tell you how to keep it running reliably.

This skill solves the operational layer: the heartbeat framework that keeps autonomous agents from stomping each other, missing tasks, or crashing without explanation.

Invoke /heartbeat-setup and it generates a complete framework for any AI agent:

HEARTBEAT.md �� step-by-step execution checklist:

• Identity confirmation on every wake
• Inbox triage (compact inbox-lite pattern)
• Safe checkout with 409 dedup �� no race conditions
• Incremental context loading �� no full thread replays
• Blocked task dedup �� do not re-notify what is already blocked
• Exit discipline �� always comment before going dark

Also generates:

• AGENTS.md �� agent identity, capabilities, chain of command, safety rules
• SOUL.md �� agent personality, communication style, optimization targets

The problems this prevents:

| Problem | Without | With | |---------|---------|------| | Two agents grab same task | Duplicate work | Checkout + 409 dedup | | Agent goes quiet mid-task | Silent failures | Mandatory exit comment | | Blocked task spams updates | Repeated same alert | Dedup pattern | | Cold start loses context | Full thread reload | Incremental loading |

Built from real production use �� the exact framework running Eve CEO, an autonomous AI CEO managing a live company with real tasks, real products shipped, and real revenue generated.

If you want the full deployed CEO agent, see Eve CEO ($79) on ClawMart.