coding-agent

Interactive coding agents output garbled text or hang because they expect a terminal. Allocating a pseudo-terminal (PTY) before execution ensures interactive CLI applications run correctly without broken or hanging output.

You can run a coding agent in the background by starting it with a PTY and using session lifecycle actions like list, poll, log, write, submit, paste, send-keys, and kill to monitor and manage the background process reliably.

Yes, you can scope an interactive coding agent to a specific git worktree or repository directory by specifying a focused work directory. This ensures the agent operates safely within the intended project context.

Yes, running Claude Code or Codex in automated workflows requires PTY allocation. These interactive terminal applications need a pseudo-terminal to function correctly and avoid broken, hanging, or garbled output during programmatic execution.

The best way to manage long-running coding agent sessions for refactoring is to start them in the background with a specified work directory and PTY allocation, then use session tracking actions like poll, log, and send-keys for reliable monitoring and input.

You can send input or keystrokes to a background coding agent session by using the write, submit, paste, and send-keys actions. These lifecycle commands allow you to interact with and control the agent during execution.