Running a Robomotion Flow

Run a flow on a robot, watch the agent-mode event stream, react to failures. Four moving parts:

1. Pre-flight — robomotion validate <flow-dir> catches pspec/schema errors locally in <1s before you ever submit. Cheap; always do it first.
2. A robot — the person's own, connected on this computer (Robomotion's Desktop App, or robomotion robot connect when the person agrees). robomotion get robots shows which are connected.
3. Trigger — robomotion run <flow-dir> --robot <name> builds locally, submits, and streams the event log.
4. Observation — run tails the JSONL session log; robomotion logs --last reads it again. A run is done at its flow_end (or flow_error) event; an agent_mode start/end pair may bracket it, but do not wait for one.

A flow that starts at Robomotion.ChatAssistant.ChatIn is not run here. Nothing would send it a message; it would wait until --timeout. Test it as an Agent in the real chat page with the running-chat-assistant skill (robomotion agent).

No browser: the log is the only oracle for a flow. A flow has no screens, so nothing shows you whether it worked but the robot's events. Read them; never say a run worked without a flow_end status=success in them.

Step 1 — Validate (mandatory pre-flight)

robomotion validate <flow-dir>           # exit 0 = pspec-clean, 1 = errors on stderr