Run a set of independent tasks as a parallel agent team

Usage

team_run(
  tasks,
  model = NULL,
  n_workers = NULL,
  permission_mode = "dont_ask",
  cwd = getwd(),
  parent_rules = NULL,
  parent_policy = NULL,
  security_context = NULL
)

Arguments

tasks: Character vector of task prompts (one sub-agent per task).
model: Character. Model spec each agent uses. Defaults to the CODEAGENT_MODEL env var or "claude-sonnet-4-6".
n_workers: Integer or NULL. Number of parallel daemons. Defaults to min(length(tasks), parallelly::availableCores()) so it never exceeds the container's cgroup CPU quota (each daemon is a heavy R process).
permission_mode: Character. Permission mode for each agent (default "dont_ask" since parallel agents cannot prompt interactively – NOT "bypass", so user-defined deny rules are still honoured).
cwd: Character. Working directory for each agent.
parent_rules: List. Permission rules inherited from the parent agent.
parent_policy: List. Tool capability policies inherited from parent.
security_context: Internal immutable parent security snapshot. When supplied it takes precedence over the legacy permission arguments.

Value

A list (same length/order as tasks), each element either the agent's text result or an [Error] ... string.