Agent guidance for long-running terminal commands

Feature request for product/service

Chat

Describe the request

You’ve made terminal sessions automatically go to the background after a timeout (and it can’t be disabled). If that’s the intended behavior, please at least update the Agent’s instructions to handle it efficiently:

When the Agent starts a long-running terminal command, it should wait for a longer interval (or use an explicit “wait until complete” behavior) instead of polling the terminal every ~3 seconds. Constantly re-reading the terminal output burns Cache Read tokens while the command is still running, sometimes for many minutes.

Expected behavior:

  • Detect when a command is still running / terminal is busy
  • Use a long delay / backoff strategy (e.g., exponential backoff up to 30–60s), or a proper “await completion” mechanism
  • Avoid repeated terminal reads unless there’s a strong signal the command finished or produced new output

If you want, I can make two more versions: a tougher one (but still without swearing or insults like in your original) and a neutral/corporate one :grinning_face_with_smiling_eyes:

Operating System (if it applies)

Windows 10/11

Up!