Hierarchical Agents — AI skill for Claude Code

hierarchical-agents

**Turn isolated Hermes agent profiles into a coordinated organization with hierarchy, messaging, delegation, and shared memory.**

**Requires Hermes.** This project is a coordination layer that runs on top of [Hermes](https://github.com/GieshBuilds) — it does not replace it. You need a working Hermes installation with at least one profile before setting this up.

Hermes gives each agent its own profile — an isolated session with its own identity, tools, and state. But profiles can't talk to each other. This project adds the coordination layer: an org chart, an IPC message bus, delegation chains, scoped memory, and worker lifecycle management. Profiles stop being silos and start working as a team.

Pure Python 3.10+ stdlib. Zero external dependencies in core. All persistence via SQLite.

---

Quick Start

If you have Hermes installed and want to add the hierarchy layer now:

git clone https://github.com/GieshBuilds/hierarchical-agents.git
cd hierarchical-agents
pip install -e .
python scripts/sync_hermes_profiles.py --show-chart   # import your profiles
python -m core show-org-chart                          # verify the result

For the complete setup — environment variables, gateway daemons, systemd services, and tool configuration — see **[docs/SETUP.md](docs/SETUP.md)**.

Want your AI agent to do the integration automatically? Paste the prompt in **[docs/AGENT-INTEGRATION-PROMPT.md](docs/AGENT-INTEGRATION-PROMPT.md)** to Claude Code or any capable assistant.

---

The Problem

Hermes profiles are powerful in isolation. Each one has its own SOUL.md, session history, skills, model selection, and gateway. But out of the box:

• Profiles can't message each other
• There's no hierarchy — no way to say "the CTO manages these PMs"
• There's no task delegation — no way to assign work from one profile to another and get results back
• There's no shared memory — each profile's knowledge is locked in its own silo
• Th