team down

Stops the team: asks every running seat to exit, closes its workspace, stops the watch, and stops the herdr session. A seat is only asked when it is free — idle, with an empty input box — so a working, blocked or half-typed seat is left running and named. --dry-run prints the plan and runs nothing.

A seat the file renamed after up launched it is stopped all the same, under the name it was launched with and the CLI the state records for it: what runs, not what the file now calls it.

Synopsis

team down [--dry-run] [--wait] [--abandon] [--session <name>] [--file <path>]

What it reads and writes

Reads the team file, this machine's approval store, the session's state (.agents/team.state.json, for the watch's pid and the seats' recorded panes and CLIs), and herdr: whether the session is running, its agents, each pane's screen and status, and each pane's foreground processes. Writes .agents/team.state.json (the seats it stopped are dropped), .agents/team.log, and, through herdr: the exit in each pane, the workspaces it closes, the watch's process, the session it stops and the stopped session it clears — its own, just stopped. A team file that no longer validates is replaced by the last copy that did, with a notice printed first: down must keep working when the file breaks.

Who may run it

The owner, the coordinator's seat, and the operator's seat. A seat's own call is refused by the --abandon flag, which only the owner may use. A seat that may stop the team never stops the coordinator's or the operator's seat — only the owner does.

Flags

What it prints

claude-keeper: stopped
watch: stopped
session beacon: stopped and cleared

A seat it stops prints <seat>: stopped; the watch prints watch: stopped; the session prints session <session>: stopped and cleared — herdr keeps a stopped session listed, so down clears the one it has itself just stopped, in the same run, and the next up starts from the beginning. up never deletes a session: one stopped any other way keeps its refusal, with the command to run. A clear that does not happen — herdr still reports the session running, or the delete fails — prints session <session>: stopped; it did not clear, run \herdr session delete `instead, anddownstill exits 0: the stop itself succeeded. A seat it does not stop prints one skip` line and is named in the last line instead of the session being stopped:

claude-beacon: is working (`--wait` waits for it); left running
session beacon: not stopped, 1 agent left in it

It prints session <session> is not running: nothing to stop and exits 0 when there is nothing to stop at all.

Refusals

Exit codes

• 0 — the seats it could stop were stopped, the watch and the session with them; or there was nothing to stop; or --dry-run printed its plan. A seat left running because it was busy is not a failure.
• 1 — refused, or a step failed: a seat's exit was not typed, its workspace did not close, it timed out leaving its pane, or the watch or the session did not stop.
• 2 — the invocation, the team file or herdr can't be read.

Examples

format: 1
project: beacon
coordinator: claude-keeper
operator: claude-keeper

workspace:
  mode: shared

seats:
  - role: coordinator
    name: claude-keeper
    label: coordinator
    cli: claude-code
    vendor: anthropic
    model: Claude Opus
    version: "5.5"
    launch: claude --model claude-opus-5-5

  - role: implementer
    name: claude-beacon
    label: implementer
    cli: claude-code
    vendor: anthropic
    model: Claude Opus
    version: "5.5"
    launch: claude --model claude-opus-5-5

Two seats idle, a watch running: this is the whole plan.

$ team down --dry-run ; echo "exit $?"
+ herdr --session beacon pane run w1:p1 /exit
  wait until claude-keeper's pane is back at its shell (30 s at most); on a time-out it is left as it is
+ herdr --session beacon workspace close w1
+ herdr --session beacon pane run w2:p1 /exit
  wait until claude-beacon's pane is back at its shell (30 s at most); on a time-out it is left as it is
+ herdr --session beacon workspace close w2
+ kill 4242
    (the watch)
+ herdr session stop beacon
    (stopped, then cleared: the session this run stopped, so a later `up` starts from the beginning)
dry run: nothing was run
exit 0

An implementer's seat is not the coordinator's: it cannot stop the team.

$ team down ; echo "exit $?"
team down: only the owner, the coordinator or the operator stops the team; this call is claude-beacon
exit 1

The coordinator's own call stops the team but never its own seat, so the session stays up:

$ team down --dry-run ; echo "exit $?"
  skip claude-keeper: left running; only the owner stops the coordinator's or the operator's seat
+ herdr --session beacon pane run w2:p1 /exit
  wait until claude-beacon's pane is back at its shell (30 s at most); on a time-out it is left as it is
+ herdr --session beacon workspace close w2
+ kill 4242
    (the watch)
  skip session beacon: not stopped, 1 agent left in it
dry run: nothing was run
exit 0

The file renamed claude-beacon after up launched it. The seat still runs under the name it was launched with, and the state holds the CLI it was launched with, so the plan is the same as before the rename — only the file's copy of the name changed:

format: 1
project: beacon
coordinator: claude-keeper
operator: claude-keeper

workspace:
  mode: shared

seats:
  - role: coordinator
    name: claude-keeper
    label: coordinator
    cli: claude-code
    vendor: anthropic
    model: Claude Opus
    version: "5.5"
    launch: claude --model claude-opus-5-5

  - role: implementer
    name: claude-relay
    label: implementer
    cli: claude-code
    vendor: anthropic
    model: Claude Opus
    version: "5.5"
    launch: claude --model claude-opus-5-5

$ team down --dry-run ; echo "exit $?"
+ herdr --session beacon pane run w1:p1 /exit
  wait until claude-keeper's pane is back at its shell (30 s at most); on a time-out it is left as it is
+ herdr --session beacon workspace close w1
+ herdr --session beacon pane run w2:p1 /exit
  wait until claude-beacon's pane is back at its shell (30 s at most); on a time-out it is left as it is
+ herdr --session beacon workspace close w2
+ kill 4242
    (the watch)
+ herdr session stop beacon
    (stopped, then cleared: the session this run stopped, so a later `up` starts from the beginning)
dry run: nothing was run
exit 0

For the owner, running that plan stops everything, and clears the session it stopped — the next team up starts a fresh one, with no step in between:

$ team down ; echo "exit $?"
claude-keeper: stopped
claude-beacon: stopped
watch: stopped
session beacon: stopped and cleared
exit 0

Once the session is down there is nothing to stop, and saying so is not an error:

$ team down ; echo "exit $?"
session beacon is not running: nothing to stop
exit 0

Even with the team file broken, down still reads the last copy that validated and stops the team:

format: 1
project: beacon
seats: [

$ team down ; echo "exit $?"
team.yaml is invalid (line 3: "[" is not closed on its line); using the copy of 2026-10-04T09:00:00.000Z
session beacon is not running: nothing to stop
exit 0