A neutral, general-purpose chat agent — the simplest thing the Ranch can run. No admin privileges, no preinstalled skills, no MCP servers attached. The persona is a polite, direct assistant; it mirrors the user's language and stays out of long-winded explanations unless asked.
Use this as a base when you want to:
- experiment with the platform without committing to a domain,
- compose a custom agent by adding skills and MCPs from the catalogue after install,
- run paddock evaluations against a known-clean baseline.
| File | Purpose |
|---|---|
.agent/SOUL.md |
system prompt — neutral helpful-assistant persona |
.agent/USER.md |
describes the user as a generic chat user (not the operator) |
.agent/agent.config.json |
runtime defaults (iterations, tokens, access strategy, etc.) |
.agent/HEARTBEAT.md |
empty — no periodic checks |
.agent/MEMORY.md |
empty index, populated at runtime |
.paddock/config.json |
minimal eval config |
.paddock/scenarios/ |
two starter scenarios — greeting and language mirroring |
None — the default template has no params or secrets to fill in. SOUL.md handles language mirroring at runtime, so there is nothing to configure at install.
The fresh agent is functional but bare. Common next steps:
- Attach skills via the Ranch UI (e.g.
web-search,memory). - Attach MCP servers if the agent needs domain tools.
- Tighten
accessStrategyinagent.config.jsonif the agent is exposed to a public surface.