Monitor and troubleshoot

Is the runner connected?

rho status                         # on the runner
docker exec rho-runner rho status  # in a container
rho runners                        # from the agent's installation: eligible runners

In the Nexus browser, Runners lists each runner you manage with its presence (online while an inbox subscription is open, offline with last_seen_at, not_yet_seen) and credential state. Presence is display only; Nexus addresses work by eligibility, never by presence, so an offline runner simply accumulates addressed rows until it returns or the task deadline settles them.

last_seen_at means only "current-epoch address contact" — not health, not readiness, not that a task will succeed.

Nothing is executing

Check	What to look at
Is the runner selected?	The conversation's runner, or rho runners use. A call with no target fails runner_target_required; a call addressed to an ineligible runner fails explicitly rather than moving to another runner.
Is the call parked for approval?	The conversation shows it; a runner only sees granted calls.
Does the runner announce the tool?	GET /agent_api/v1/executors/{id} shows served_tools. A name nobody serves fails tool_not_served.
Is the credential live?	rho status; reconnect with rho connect after revocation, expiry or a replaced pairing. A 401 is never explained.
Is the runner eligible for the agent?	Check its assignment scope and, for agent_private, its bound application.

Another rho instance's bundled Runner is absent from rho runners, even when the same Human manages both instances. This is the agent_private scope working as intended. Connect an independent Runner when those instances should share an execution environment. The Nexus Runners page lists the Human's managed registrations; that management list is broader than one Agent's eligible Runner list.

A task expired

A claimed call that passes its deadline with no result is settled by the sweep from its frozen effect profile: a replayable call fails tool_timeout and may retry; a non-replayable call becomes uncertain and waits for a person's retry or abandon because its effect may already have happened. Long tools extend their own deadline; a runner that cannot extend in time should commit partial results rather than go silent.

Stop cleanly

Give containers a 30-second grace period. rho handles termination, stops the process groups it owns and lets a claimed call commit or expire by its own deadline. Nexus claim fencing, expiry and result integrity remain authoritative whatever the runner does.

Logs and diagnostics

• docker logs rho-runner, or the service manager's log on a host.
• rho doctor --strict checks the installed runtime and tools (Playwright, document tools) without making a model request.
• ./cybros logs rho for the stack's runner.

Before sharing logs, remove credentials, private setup links, private prompts and file contents.

Draft

Queue-depth and per-runner activity views for operators are not yet documented here.