ops-maintenance-agent

Diamondnode maintenance control plane combining:

Skill / command	Role
/verification	End-to-end service + Kimi + gh flow checks
/troubleshooting	Chrome DevTools MCP path (browser layer when needed)
/build-agent	Cloudflare MaintenanceAgent DO with cron schedule
/pr-babysit all	Inventory all open PRs; flag CI/review/conflicts
/subagent-driven-development	Parallel domain work for scaffolds
/sentry-svelte-sdk	SvelteKit hooks under svelte-hooks/
/axiom-alerting	Monitor JSON under monitors/
/building-dashboards	Dashboard blueprint under dashboard-spec/
/workflow	Durable maintenanceWorkflow (Vercel WDK)
setup-kimi-mobile-replies.sh	Kimi mobile updates via OpenClaw-moY

Quick start (diamondnode)

cd /root/ops-maintenance-agent

# One cycle: verify → PR inventory → Kimi push
node scripts/maintenance-cycle.mjs

# Kimi bridge (optional Moonshot key for LLM replies)
bash scripts/setup-kimi-bridge.sh
# bash scripts/setup-kimi-bridge.sh sk-YOUR_MOONSHOT_KEY

# Install scheduled cycle + HTTP control plane + post-boot/crash automation
bash scripts/install-automation.sh
# (or manual: cp systemd/*.service systemd/*.timer /etc/systemd/system/
#  systemctl daemon-reload && systemctl enable --now ops-maintenance-http …)

Reboot / crash automation

Unit	Role
ops-post-boot.service	On multi-user boot: crash analysis → heal allowlisted services → verify → Kimi
ops-crash-analyze.service + .timer	Read-only journal/failed-unit snapshot every 30m (OnBootSec=5min)
ops-maintenance-cycle.timer	15m cycle; on critical/core-down runs analysis + force-heal if needed
ops-maintenance-http.service	Local control plane 127.0.0.1:8791

# Manual runs
node scripts/crash-analyze.mjs
node scripts/post-boot-reconfig.mjs [--force-heal]
node scripts/maintenance-cycle.mjs

# Status
systemctl status ops-post-boot.service
systemctl list-timers 'ops-*'

Heal allowlist (never ssh/sshd): openclaw-gateway, ollama, nginx, diamond-gateway, redis-server, qmem-llama-server, qmem-sync, genesis-ops-stream, cloudflared, ops-maintenance-http.

Reports: reports/crash-analysis-latest.json, reports/post-boot-latest.json, reports/cycle-latest.json.

Unblock checklist

Capability	Requirement
PR babysit fix cycles	gh auth login or GH_TOKEN + git repo remote
Axiom monitors/dashboards	~/.axiom.toml with API token + org_id
Sentry live events	SENTRY_DSN (+ auth token for source maps)
Full Kimi LLM replies	Moonshot key → setup-kimi-mobile-replies.sh
CF Agent deploy	wrangler login + npm i + npm run agent:deploy
Vercel Workflow deploy	Link project, install workflow, wire API route with start()

HTTP API (local)

Method	Path	Purpose
GET	/health	Liveness
GET	/ops/verify	Run verification, return JSON report
GET	/ops/deployments	Fleet deploy audit (Vercel/CF/local); ?refresh=1 to re-run
POST	/ops/pr-babysit	Inventory open PRs
POST	/ops/kimi-update	Push status to Kimi mobile
POST	/ops/maintenance-cycle	Full cycle

Default: http://127.0.0.1:8791