terraform apply during an infra migration stopped at Do you want to perform these actions and the agent timed out...

TL;DR: Split the run into plan and apply: terraform plan -out=tfplan first, then terraform apply tfplan - applying a saved plan file never prompts. Review the plan output before approving it, and record that approval in the run log. The prompt exists to stop unreviewed changes; the fix is to move the review earlier, not to blindly auto-approve.

Do you want to perform these actions?

1. Confirm the stall is the approval prompt: the process is idle and the last output is the plan summary followed by the actions question. Expected: you can see the full proposed plan in the output above the prompt.

2. Kill the hung process - with the prompt unanswered, nothing was applied. Expected: the process exits and terraform plan afterward shows the same pending changes.

3. Generate a saved plan: run terraform plan -out=tfplan. Expected: the plan is written to the tfplan file and the summary is printed for review.

4. Review the plan output (resources to add, change, and especially destroy) and record the approval decision in the run log. Expected: a human or policy check has explicitly approved this exact plan.

5. Apply the saved plan: run terraform apply tfplan. Expected: apply runs to completion with no prompt, since a saved plan is pre-approved.

Use this when

• terraform apply stalls on the approval question in a scripted or agent session
• an infra migration step times out waiting on the prompt
• you want apply to be non-interactive but still reviewed

Not for this skill when

• the plan itself fails - fix the configuration first
• state is locked by another run - resolve the lock, do not force it
• you are applying interactively at a terminal - just answer the prompt

Variant phrasings

• terraform apply hangs waiting for approval in CI
• Do you want to perform these actions with no tty
• agent terraform apply timed out on confirmation

Why it happens

terraform apply without a saved plan file always asks for confirmation before changing infrastructure - it is the last checkpoint between a proposal and real changes. In a headless agent run there is nobody to answer, so it waits until the harness kills it. Applying a saved plan file skips the question because the plan was already reviewed when it was created; the approval moves from apply-time to plan-time, which is exactly where an agent workflow can handle it.

Edge cases

• The -auto-approve flag also skips the prompt but removes the review checkpoint entirely - prefer the saved-plan workflow so every apply traces back to a reviewed plan.
• A stale plan file (configuration changed after planning) is rejected at apply time - regenerate the plan rather than forcing it.
• State locking still applies: if another process holds the state lock, apply waits on the lock, which looks like a hang but is a different problem.
• Destroy operations in a plan deserve a second pair of eyes - require explicit human approval for any plan containing destroys, even in automation.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_lrhRWQZ3aRIzBHFccvKU1g