Long-running command (build, test suite, dev server) times out or blocks an AI coding agent's shell. How do I run it in the background and poll for the result?

Cause: Agent shell tools have a per-command time limit, and a dev server or watcher never exits, so the tool call blocks until the limit. In Claude Code the foreground Bash timeout defaults to 2 minutes (BASH_DEFAULT_TIMEOUT_MS=120000) and the model-settable maximum defaults to 10 minutes (BASH_MAX_TIMEOUT_MS=600000). Fix (best first): 1. Run it in the background and check on it. Claude Code's Bash tool supports run_in_background, for long-running processes such as dev servers or watch builds; list and stop tasks with /tasks. If a foreground command reaches its timeout it is moved to the background rather than killed (unless it starts with sleep), and the result tells you the task id and output file. The agent can read the output file for progress. Claude Code also has a Monitor tool that runs a command in the background and feeds each output line back. 2. In unattended runs (claude -p, SDK apps, CI, cloud sessions) background commands have a time limit; the docs give 30 minutes by default and up to 2 hours (10 minutes by default in -p runs with a text prompt). A command stopped at the limit is reported, and it can be restarted with a longer timeout. 3. Raise limits only when needed: the environment variables BASH_DEFAULT_TIMEOUT_MS and BASH_MAX_TIMEOUT_MS take milliseconds. Prefer background plus polling over very long foreground waits. 4. In any agent without a background feature, start the job yourself detached from the shell, for example nohup make test > build.log 2>&1 &, then poll with tail -n 50 build.log and check the exit status written to a file. Use a short sleep between polls rather than a tight loop. 5. Make the job report progress and write a clear completion marker so polling can tell finished from stuck. 6. Remember that cd and exported variables inside a backgrounded command do not carry over to later commands. Don't: disable timeouts globally or loop with sleep inside one long foreground command. Verify: start the job, confirm the agent regains control immediately, read the log, and see the final exit status when it completes. Sources: https://code.claude.com/docs/en/tools-reference https://code.claude.com/docs/en/env-vars