heartbeat-monitor
External uptime and liveness check for the bot, memory pipelines, and local API. Use when the user wants to monitor whether the agent is alive, set up a health check, get notified when the bot goes do
SKILL.mdoptional-skills/devops/heartbeat-monitor/SKILL.md
GitHubHeartbeat Monitor
Check that the running stack is healthy and report a single green/red verdict.
What to probe
The framework runs a local API on port 4322 (see
.claude/sections/03_* / orchestration). Probe the pieces that matter:
| Component | Probe |
|---|---|
| Local API | GET http://127.0.0.1:4322/health returns 200 |
| Chat process | process for run_chat.sh is alive |
| Memory heartbeat | last heartbeat timestamp is < 2× the interval old |
| Disk | vault partition has free space |
Run
bash optional-skills/devops/heartbeat-monitor/scripts/healthcheck.sh
# exit 0 = all green, exit 1 = something is down
The script prints a per-check table and sets a non-zero exit code on any failure, so it composes with cron / CI / an external uptime service.
Escalation
- For local use, surface the red verdict in chat.
- For unattended monitoring, point an external uptime service (Healthchecks.io,
UptimeRobot, a cron on another host) at this script or at the
/healthroute — a monitor running inside the same process it watches can't report its own death. - This skill only reads state; it does not restart anything. Pair with
log-triageto explain why a check went red.