AI Agents & Sessions
Run Long Tasks Overnight
By TermSquad · Last reviewed
You can leave a long task running in a persistent TermSquad session overnight. Prepare the task so it can make progress without repeated input, and check its output when you return.
Prepare the work
- Use a running computer with enough memory and disk space.
- Save or commit your current work so you can compare changes.
- Confirm the agent is connected and your provider has usable quota.
- Give the agent a clear scope, expected output, and validation command.
- Explain when it should stop and ask for help.
- Resolve initial permission requests.
- Confirm progress before disconnecting.
A useful instruction is:
Work only in this repository. Implement the agreed change, run the relevant tests, and leave a summary of changed files and test results. Stop if the task requires credentials, an external purchase, or a decision that is not in the brief.
Adapt the instruction to your project; do not promise the agent access it does not have.
Plan for interruption
Ask the agent to keep brief progress notes for a long, multi-step task. Use shared memory for durable decisions, and use version control or external copies for project data.
Leave normal approval settings unless you deliberately choose another mode. YOLO can remove some prompts, but it also changes the tool’s access controls and does not solve rate limits, crashes, or resource exhaustion.
Review in the morning
Reconnect to the original session. Check whether the task finished, stalled, or needs input. Inspect the diff and test results before using the output.
Read persistent-session limits, Guard, and agent permissions.
