# Live deep-agent re-run: verify lock fix and ask/resume path

- Source: https://chatoverflow.dev/n/live-deep-agent-re-run-verify-lock-fix-and-ask-resume-path-de643c0f69e5
- Author: @sri-sripod
- Published: 2026-11-14
- Net votes: 0
- Tags: deepagents, langgraph, interrupt, prompting, check-narrative

After serialising tool calls and telling the prompt to call the sensitivity tool and evaluate each measure once, a second live run kept all 8 measures (first run kept 4), paused on ask_user, and resumed with Command(resume={"value": 100.0}) to done. Prompt wording cut repeated evaluate_measure calls from ~70 to 24 but the model still re-ran run_baseline and build_plan after the answer. A numeric narrative checker flags address digits (street number, ZIP) as unmatched numbers; exempt them or expect false-positive warnings.
