joelclaw System Health Check
Run scripts/health.sh for a full system health report with 1-10 score.
~/Code/joelhooks/joelclaw/skills/joelclaw-system-check/scripts/health.shWhat It Checks (16 components)
| Check | What | Green (10) | Yellow (5-7) | Red (1-3) |
|---|---|---|---|---|
| k8s cluster | pods in joelclaw namespace | 4/4 Running, 0 restarts | partial pods | no pods |
| pds | AT Proto PDS on:9627 | version + collections | pod running, host publish degraded | pod not running |
| worker | system-bus on:3111 | 16+ functions | responding, low count | down |
| inngest server | :8288 reachable | responding | — | down |
| redis/gateway | Redis + gateway session queues | connected, low pending queue | connected, backlog rising | unavailable |
| typesense/otel | Typesense health + OTEL query path | healthy + queryable | healthy, query degraded | unavailable |
| tests | bun test in system-bus | 0 fail | — | failures |
| tsc | tsc --noEmit | clean | — | type errors |
| repo sync | monorepo HEAD vs origin/main | in sync | ahead/behind | repo unavailable |
| memory pipeline | joelclaw inngest memory-health | healthy checks | degraded checks | failing checks |
| pi-tools | extension deps installed | all 3 deps | — | missing |
| git config | user.name + email set | set | — | missing |
| active loops | joelclaw loop list | queryable | query degraded | unavailable |
| gogcli | Google Workspace auth | account authed, token valid | token stored, no password | not configured |
| disk | free space + loop tmp | <80% used | — | >80% |
| stale tests | __tests__/ + acceptance tests | clean | — | present |
When to Run
- Session start — orient on system state before doing work
- After loops complete — verify nothing broke
- After infra changes — k8s, worker, Redis config
- When something feels off — quick triage
Fixing Common Issues
Repo drift: cd ~/Code/joelhooks/joelclaw && git fetch origin && git status -sb
pi-tools broken: cd ~/.pi/agent/git/github.com/joelhooks/pi-tools && bun add @sinclair/typebox @mariozechner/pi-coding-agent @mariozechner/pi-tui @mariozechner/pi-ai
PDS unreachable: curl -fsS http://localhost:9627/xrpc/_health then kubectl get deploy,svc,pods,pvc -n joelclaw | rg 'bluesky-pds|NAME' (or if pod down: kubectl rollout restart deployment/bluesky-pds -n joelclaw)
Worker down: joelclaw inngest restart-worker --register
Stale tests: rm -rf ~/Code/joelhooks/joelclaw/packages/system-bus/__tests__/ && find ~/Code/joelhooks/joelclaw/packages/system-bus/src -name "*.acceptance.test.ts" -delete
Loop tmp bloat: rm -rf /tmp/agent-loop/loop-*/ (only when no loops are running)
Inngest Hung-Run Quick Triage
When a run appears stuck after first step:
joelclaw run <run-id>If trace shows Finalization failure with "Unable to reach SDK URL":
- Verify registration/health:
joelclaw inngest status - Verify function is present where expected:
joelclaw functions | rg -i "manifest-archive|<function-name>" - Check for stale app registrations in Inngest UI/API and remove stale SDK URLs.
- Assume possible handler blocking (not just network): review recent step code for filesystem/Redis/subprocess blocking before step response.