Configuration

Hordr is zero-config: a .beans.yml with just a beans: block is enough. Defaults cover implementer, tester, reviewer, and merger roles, all on the opencode harness. Add a hordr: section only to override.

Minimal

.beans.yml

beans:
  # …beans settings…

That’s it. primary_branch defaults to develop.

Full shape

hordr:
  primary_branch: develop
  agents:
    implementer:
      harness: opencode
      persona: |
        You implement ONE task bean assigned to you.
        …
    tester:
      harness: opencode
    reviewer:
      harness: claude        # any harness hordr can launch

Fields

Field	Default	Notes
primary_branch	develop	The branch hordr finish merges into.
agents	(map)	Role name → {harness, persona?}. Merged over defaults.
company	(unset)	Path to an Agent Companies package (AGENTS.md, skills).

Roles

A role is any key under agents. Each maps to:

• harness — the agent binary to launch (opencode, claude, …). What hordr can resolve and shell out to.
• persona — the prompt identity. Optional when a company context is active (then the persona comes from the company’s AGENTS.md).

User-configured agents take precedence over the built-in defaults; defaults fill the gaps. So you can override one role and inherit the rest.

Built-in defaults

The defaults are minimal, fleet-shaped instructions: do one task, verify, commit code + bean together, signal done, stop. They also tell the agent not to re-read the bean (it’s embedded in the prompt) and not to re-run beans prime / hordr prime (it’s already primed via AGENTS.md).

See src/config/defaults.ts for the exact text — copy and adapt the persona for a role to customize its behavior.

Company context (optional)

For larger setups, point company at an Agent Companies package root holding COMPANY.md, agent AGENTS.md files, and skills. With a company active, role personas resolve from the company manifest instead of inline persona: strings.

hordr:
  company:
    path: ./AGENTS