Emergence Agent CEO Architecture#

A multi-agent system where a CEO Agent (Hermes) runs 24/7 on a server, coordinating specialized sub-agents through GitHub Issues. The human stakeholder focuses on strategy, relationships, and PR review — while agents handle execution.

Agent Roles#

RoleResponsibilityCEO Agent (Hermes)Competitor analysis, market survey, task creation, delegationGrowth LeaderContent marketing, publications, social distribution, daily pulse signalsDevOps LeaderInfrastructure monitoring, CI/CD health, error recovery, incident logging

Interaction Model: Human-in-the-Loop#

By delegating tedious detailed work to agents, the human stakeholder is freed to focus on high-value activities:

• Strategy: Provide long-term direction via GitHub Issues
• Relationships: Manage public human relationships and stakeholder experience
• Review: Merge GitHub Pull Requests with approval
• Amplification: Provide external tools and credentials (Railway, Vercel, gh, API keys)

The CEO agent handles execution. Humans handle strategy, relationships, and judgment.

Workflow#

1. Strategic Intent: Human provides direction via GitHub Issue or IM message
2. Analysis: CEO agent researches topic, analyzes competitive landscape, writes analysis in issue comments
3. Delegation: CEO creates and assigns tasks to sub-agents (Growth Leader, DevOps Leader)
4. Execution: Sub-agents produce work in publications/, pulse/, ops/ directories
5. Review: Sub-agents open PRs for human stakeholder review and merge
6. Heartbeat: CEO runs proactive daily cron tasks via HEARTBEAT.md schedule

Quick Start#

bash

# 1. Clone this repo to your server
git clone https://github.com/emergencescience/emergence-agent-ceo.git
cd emergence-agent-ceo

# 2. Scaffold the workspace
./scripts/scaffold.sh /path/to/your/workspace

# 3. Configure your environment
cp .env.example .env
# Edit .env with your LLM API key and GitHub repo info

# 4. Start the agent (choose your runtime)
# openclaw gateway start          # OpenClaw
# claude run --agent hermes       # Claude Code agent mode
# (any autonomous agent framework with cron support works)

Workspace Structure#

text

emergence-agent-ceo/
├── SOUL.md                  # CEO agent identity and boundaries
├── SOUL-DEVOPS.md           # DevOps Leader identity and boundaries
├── SOUL-GROWTH.md           # Growth Leader identity and boundaries
├── MEMORY.md                # Long-term organizational memory
├── HEARTBEAT.md             # Cron schedule and proactive tasks
├── pulse/                   # Daily signal reports
├── publications/            # Draft staging (blog, social)
├── strategies/              # Roadmaps and strategic direction
├── research/                # Intelligence and competitive analysis
├── ops/                     # Runbooks and operations
├── designs/                 # Architecture blueprints
└── .github/                 # GitHub Issue templates

Model Recommendations#

The CEO's job — strategic reasoning, competitor analysis, task decomposition, quality judgment on sub-agent output — requires deep reasoning capabilities. Use a strong reasoning model for the CEO; sub-agents can use faster, cheaper models for execution tasks.

License#

MIT