Squad routing lets a leader agent choose which member should handle a task.
It is optional. Direct agent routing remains valid.
Use a squad when:
the task spans multiple specialties
the best worker is not obvious at intake
the routing target should stay stable while members change
the leader has clear routing instructions and member roles
Do not use a squad when a single qualified agent is already known.
When a squad is used, record:
squad id or name
leader agent
members
member roles
routing instructions
leader dispatch
leader decision
delegated member dispatches
no_action or escalation, if no member is delegated
The leader agent must:
The leader must not silently become the implementer unless the routing record says so.
Delegation to a member is a new dispatch. It requires its own receipt:
dispatch_written
dispatch_submitted
dispatch_completed or dispatch_blocked
Leader judgment is not member completion.
Agent-to-agent routing can accidentally create loops.
The default rule:
Do not mention or re-dispatch another agent for thanks, acknowledgements,
sign-offs, or "no action needed" messages.
Use delegation only for a concrete new subtask, escalation, or explicit user request.
Squad routing cannot bypass:
provider matrix checks
context compression
approval gates
receipt gates
expected evidence
review/fix/review
agent recommendation resolution
Squad routing is a routing mechanism, not a completion mechanism.