Deprecated 2026-05-10. The healthcheck CronJob was suspended on pai-m1 because its state-change posts were noise more than signal. The daily diagnostic loop is now pai-self-improver, which mines OpenObserve for recurring patterns and proposes memory changes for human approval.

The healthcheck agent file is still on disk for ad-hoc invocation (claude --agent healthcheck) when you want a quick pod / ArgoCD / Vault read. The old shell-only healthcheck.sh cron lives at infra/ai-agents/cronjobs/scripts/healthcheck.sh if you ever want to flip it back on temporarily.

Overview

The health check agent (claude --agent healthcheck) runs a systematic check of every workload in the pai-m1 K8s cluster. It uses kubectl for pod/app status, the OpenObserve MCP for log error scanning, and the Linear MCP for bug filing with deduplication.

Source: .claude/agents/healthcheck.md

Usage

claude --agent healthcheck

No arguments needed. The agent runs all checks and outputs a summary table.

What It Checks

Check	Method	Pass Condition
Pod health	kubectl get pods -A	All pods Running/Completed
ArgoCD sync	kubectl get applications -n argocd	All apps Synced + Healthy
CronJob health	kubectl get jobs -n ai-agents	Latest run per CronJob succeeded
Log errors	o2_error_summary (1h)	No error-level logs
Vault status	vault status	sealed=false, initialized=true
Log ingestion	o2_list_streams	k8s_logs stream has recent docs

Linear Integration

When issues are found, the agent:

1. Searches open Linear issues labeled health-check
2. If an existing issue matches, adds a comment instead of duplicating
3. If a similar resolved issue exists, creates new issue referencing it
4. New issues use title format: [health-check] namespace/workload: description
5. Max 5 issues per run — related errors are batched

Expected Workloads

The agent knows what should be running and flags missing workloads:

• argocd: server, repo-server, redis, applicationset-controller
• ai-agents: pai-responder, journalist cronjobs, pai-morning cronjob
• openobserve: openobserve-standalone, vector
• vault: vault, vault-agent-injector
• cloudflared: cloudflared
• kube-system: coredns, traefik, local-path-provisioner

Future: Scheduled Runs

Currently manual. Plan is to run as a K8s CronJob (like the journalist) on a schedule, posting results to Discord and filing bugs automatically.