Working with orq-lite flows

orq-lite runs coding-agent workflows durably. Behavior is data: versioned v2 flows compose activities and subflows, shipped with schemas, policies and prompts inside verified packs. You change a flow, validate it, and run it — there is no imperative runtime to patch.

Two things you need before any flow runs: a team.json that maps roles to agents, and a flow to run. guide.md in the repo root is the full setup narrative (gates, conventions, objective writing, GitHub watch mode). This skill is the working reference for the flow lifecycle.

1. team.json

Validate() enforces exactly this, and rejects the config before a run is created:

• at least one agent
• every role: non-empty agents, all resolving to declared agents; non-empty prompt and result_path; timeout_seconds > 0
• each agent declares either provider or cmd, never both; cmd contains {{PROMPT}}, and extra_args requires a provider

Minimum viable config for the shipped pack: