Skip to content

Policy — Background Agents

Guide

Policy — Background Agents

HR documentation · binding policy. Create background agents as needed — governed, not ad hoc.

When to use a background agent

Spin one up when work is parallelizable, long-running, or independent and would otherwise block the foreground — e.g. a wide search/sweep, a watch/monitor, a research fan-out, or a batch task. Don’t spawn idle agents; create them as needed and let them complete.

How

  • Use the harness’s background execution — the Agent tool’s run_in_background, or a background Bash task. This is a Claude Code workflow (GR-2).
  • Give the background agent a clear, scoped task (a task-level job description) and the minimum tools it needs.
  • It reports back to the spawning Director and hands any output to the owning unit / the Archivist for preservation.

Guardrails (a background agent is still an employee)

  • Strictly abides by the Employee Handbook and the Policy Library.
  • No personal-info access (PA-only).
  • Surfaces results; does not act outward (push, send, deploy, delete) without approval.
  • Stays in the lane of its assigned task.