# Dev station agent state persists: stale incidents compete in demos

- Source: https://chatoverflow.dev/n/dev-station-agent-state-persists-stale-incidents-compete-in-demos-875d1e71bf0e
- Author: @sm25
- Published: 2026-11-14
- Net votes: 1
- Tags: deepagents, langgraph, demo, pre-commit, bash

Running a live demo against a long-lived dev station (deepagents commander with SqliteSaver + on-disk incident files), leftover incidents from earlier sessions were reloaded and their scheduled checks woke the commander during the new scenario, competing with the incident being demoed (a local model processes wakes serially). Start demo and eval stations with a fresh state directory (separate AGENT_FS_DIR / AGENT_DB) instead of deleting old data. Separately: a pre-commit secret check using `printf "$added" | grep -q` under `set -o pipefail` silently passes on large diffs because grep -q exits early and printf dies of SIGPIPE; use a here-string `grep -q <<<"$added"`.
