When a run needs you

Most things never reach you: the event manager retries, and the lead decides. A run needs you only for what only you can settle. The lead asks you in its pane, and for anything urgent you also get a desktop notification. orch status shows every waiting milestone, and orch events --peek shows what is waiting for the lead.

What reaches you, and what you do

Reason	What you do
A coder asked a question the plan and the code don’t answer	Answer the lead. It runs orch answer Mn "<your answer>": the answer goes under that milestone’s “Notes from user” and the same coder carries on
An open question in the plan	Answer it in the plan and set Open questions: to - none (or tell the lead)
A permission prompt in an agent’s pane	Answer it in that pane
A milestone is held for a risky change	Look at the reasons in orch status, then orch approve Mn (or orch set Mn paused)
A round limit was hit, even on a stronger model	The lead reads the report and tells you what keeps failing. Add a fact or constraint to the plan, or take it over
The computer was restarted, herdr was closed, or the event manager died	orch status starts with !! EVENT MANAGER NOT RUNNING. Open herdr and run orch resume (see below)
An agent is missing or a pane id is stale	orch team up opens every missing agent in its folder on its model; or start it yourself (orch agents launch <kind> --model <id>), or fix the pane id from herdr pane list
A milestone lost 4 attempts	The agents were unreachable, gone or rate-limited. Restart or wait for them
The lead is not reading its events	Look at its pane. Nothing is lost: orch events --peek lists what waits, and ready work starts by itself after 15 minutes
The run’s budget is used up	Raise or remove Budget: in the plan header
The check fails on the feature branch	Fix it; the next start re-checks and continues once it passes

Resume

With a lead, you rarely type anything: the lead’s orch answer or orch dispatch puts the milestone back and starts the event manager if it stopped.

orch answer M2 "skip empty rows"   # the lead runs this with your answer
orch dispatch M2                   # or this, for a milestone that failed or was parked

By hand, or with Lead mode: off:

orch set M2 pending     # put the milestone back in the queue
orch autopilot          # start the run again

After a restart

Nothing on a switched-off computer can restart itself, so after a reboot, or after closing herdr, run one command in the repo:

orch resume

It says what it did:

1. The time orch was down is not counted. Example: M4 was sent at 11:30, the computer went off at 11:42 and came back at 13:31. M4 counts 12 minutes of work, not 2 hours, so it is not paused as “over its time budget”. The time report says Not counted: 1h49m while orch was not running. A laptop that only slept, with the event manager still up, needs no command: the sleep is taken out the same way.
2. Agents get their names back. herdr brings the panes back without the agents’ names. orch resume finds each coder’s pane by its folder and names it again. Where it can’t decide, it prints the exact command, such as herdr agent rename w1:p5 reviewer. A pane that came back as a plain shell gets its agent started again by orch team up, in the same pane.
3. Unsaved work is saved into the milestone it belongs to.
4. Each task that was in progress is sent to its agent again. A fresh conversation carries on from the task file and the code, so you don’t type “continue” in each pane.
5. The event manager is started, and it wakes the lead if events wait for it.

It is safe to run more than once. orch autopilot and the lead’s orch dispatch do steps 1 to 4 by themselves when they start the event manager after a gap.

orch never refuses a task because a coder’s folder has unsaved files. It saves them into the milestone they belong to first. If it can’t tell whose they are, it keeps them in a rescue ref and prints its name, for example refs/orch/feature-x/rescue-opencode-1-20261114-133100. orch finish keeps rescue refs and lists them.

Useful commands

• orch status: every worker and milestone, and why something waits.
• orch resume: carry on after a restart.
• orch events --peek: the lead’s queue, without acknowledging anything.
• orch park M2 paused "why": stop a milestone and keep its work. A coder that is still working on it is told to stop.
• orch cooldown claude-1 60: bench an agent for 60 minutes.
• orch requeue-review M2: send a review again after the reviewer vanished.