OriginRouter CLICollaboration command reference

Collaboration command reference

collaborate is the interactive entry point, while collaboration exposes the full lifecycle. The daemon owns each run, and participants may span trusted devices and workspaces.

Use App Agent collaboration when you need to review a plan, select participants, and confirm permissions.

Creation modes

Mode	Use case
`originrouter collaborate`	Full wizard with resumable local drafts
`collaborate "<objective>"` with optional `--review` or `--yes`	Prepare a configuration from an objective; configuration and plan review are required by default, `--review` is equivalent to that default, and `--yes` accepts both startup gates
`collaboration create --objective ... --participant ...`	Explicit scripted team definition
`collaboration create --spec collaboration.json`	Versioned, reviewable complete configuration
`collaboration create --draft <draft-id>`	Continue a local draft

Scripted example:

originrouter collaboration create \
  --objective "Audit the release and prepare fixes" \
  --participant planner:codex:studio-mac:/work/app \
  --participant reviewer:claude:linux-ci:/srv/app \
  --role "planner=inspect, implement, and coordinate" \
  --role "reviewer=independently verify risk and tests" \
  --permission planner=workspace-write \
  --permission reviewer=read-only \
  --concurrency 2 \
  --token-limit 180000 \
  --yes

Creation flags

Flag	Effect
`--participant <id:claude or codex:device:workspace>`	Add a participant. Repeatable; may include agent type, device, and workspace
`--role <id=responsibility>`	Assign a natural-language responsibility. Repeatable
`--route <id=provider:model>`	Select an existing route for a participant
`--permission <id=profile>`	Select a permission profile without widening device policy
`--preference <text>`	Add a planning preference. Repeatable
`--template <id>`	Select a built-in workflow template
`--coordination-prompt <text>`	Set coordination constraints
`--concurrency <n>`	Maximum concurrency, no greater than team size
`--token-limit <n>`	Limit model usage for this run; the run pauses or ends at the limit
`--amount-limit <decimal>`, `--currency <ISO-4217>`	Monetary budget and currency
`--review`	Compatibility flag for the default required review; cannot be combined with `--yes`
`--yes`	Explicitly accept Team/configuration and plan startup confirmations; does not bypass later action approvals
`--detach`	Do not enter attach after launch
`--no-wait`	Return immediately without waiting for the planner to produce a plan
`--timeout <seconds>`	Plan wait time, clamped to 30 through 900 seconds
`--json`	Emit a machine-readable snapshot

Find and inspect

Command	Effect
`templates [--json]`	List built-in templates
`list --category all, attention, active, or recent`	Paginate runs with `--page`, `--page-size`, and `--archived`
`drafts`	List local drafts
`draft show`, `draft resume`, or `draft delete <draft-id>`	Inspect, continue, or delete a draft
`show <run-id> [--json]`	Inspect a run snapshot
`doctor <run-id> [--json]`	Check service, device, capability, task, and recovery state
`export <run-id> --format json or markdown`	Export the run record

Follow and control

attach <run-id> follows a run continuously. Use --plain for logs, --verbose or --raw for more detail, and --participant or --task to filter.

Lifecycle controls are confirm, revise --feedback, pause, resume, retry [--task], cancel, archive, and delete --yes. Permanent deletion in a non-interactive environment requires explicit --yes.

When a run requests user attention, inspect it with attention <run-id> and then resolve it:

originrouter collaboration resolve <run-id> <attention-id> \
  --action <action> \
  --text "optional reply"

States and recovery

A typical lifecycle includes configuration, plan confirmation, execution, pause, or pending interaction, followed by completion, failure, cancellation, or budget exhaustion. Run state must not be inferred solely from whether the terminal connection ended.

Local service unavailable: check service status, then use service restart if needed.
Device or relay unavailable: run collaboration doctor, then retry or wait.
Capability, runtime, provider, or route mismatch: run doctor on the target device and repair its agent route.
Budget or policy block: review run, device, and agent budgets before resume or retry.
Configuration or plan still requires confirmation: confirm interactively; use --yes only when you deliberately want to accept both startup gates.

Automation should inspect JSON state, diagnostic_code, and the process exit status together. See Exit status and error recovery.