Resume After Gateway Restart
PassAudited by ClawScan on May 10, 2026.
Overview
The skill is coherent with its restart/resume purpose, but users should understand that it schedules future one-shot agent jobs, reads session metadata, writes a resume note, and restarts the OpenClaw gateway service.
This skill appears safe for its stated purpose, but install it only if you want agents to be able to schedule a short-lived resume job and restart the OpenClaw gateway. Before using it, verify the destination session key, check `openclaw cron list` if needed, and keep the resume memory file limited to the task context needed after restart.
Findings (5)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
The gateway may restart shortly after the agent announces it, which can interrupt active agent work for a short time.
The skill intentionally proceeds with scheduling the restart after notifying the user, rather than requiring a second confirmation. This is disclosed and central to avoiding a failed handoff, but it is still a high-impact operational action.
Do not require a user 'go' message. Instead, schedule the restart to occur ~15s in the future
Use this skill only when you explicitly want the gateway restarted, and check scheduled cron jobs if you change your mind before the restart.
The skill may fail on systems that do not have the expected OpenClaw CLI, Python, or systemd user services available.
Registry metadata says no required binaries, but the script requires local commands including openclaw and python3. This is not hidden or suspicious by itself, but the dependency declaration is incomplete.
SESSION_JSON=$(openclaw sessions --active 1440 --json) MODEL_PROVIDER=$(python3 -c ...)
Before relying on it, verify that openclaw, python3, systemd-run, and systemctl are present and behave as expected on your system.
The skill can inspect session routing metadata so it can send the resume message back to the same Discord or Telegram destination.
The script reads local OpenClaw session metadata to infer the delivery route, model, and thinking level for the provided session key. This is purpose-aligned, but it uses local session state.
SESSION_JSON=$(openclaw sessions --active 1440 --json)
Use the intended session key and verify the inferred destination before restarting, especially in multi-channel or shared environments.
If the resume file contains stale, overly broad, or incorrect instructions, the resumed agent may continue the wrong task.
A future agent turn is instructed to read a persistent local memory file and act on its 'Next step'. This is the intended resume mechanism, but persistent task files should be kept scoped and accurate.
--message "Resume work after gateway restart. Read memory/post-restart-task.md. If it exists and Status is not handled, perform its Next step."
Keep memory/post-restart-task.md short and specific, avoid placing secrets in it, and review or mark it handled after the restart completes.
The agent will send an automatic 'back online' message and then resume the pending task after the restart.
The skill creates isolated future agent turns that persist across the gateway restart. The jobs are one-shot and deleted after running, which keeps the persistence bounded.
openclaw cron add ... --session isolated ... --delete-after-run
Run `openclaw cron list` before restarting if you want to confirm what will execute, and remove the jobs if automatic resumption is no longer desired.
