Helm chart improvements inspired by openclaw-helm

[thepagent]

Improvements

Comparing with openclaw-helm, the agent-broker Helm chart could benefit from:

1. Health probes — Add livenessProbe, readinessProbe, and startupProbe so k8s can detect hangs/crashes and auto-restart
2. Resource defaults — Set sensible CPU/memory requests and limits instead of empty resources: {}
3. initContainer for config — Copy config to PVC on first install only ([ -f ... ] || cp), allowing runtime config edits to persist across upgrades
4. /tmp as emptyDir — Mount a dedicated emptyDir for temp files to keep the PVC clean
5. Image digest pinning — Support image.digest to override tag with sha256 for reproducible deploys
6. Extra extensibility — Add extraContainers, extraVolumeMounts, extraVolumes, podSecurityContext for operator flexibility
7. Gateway token auto-generation — Auto-generate the Discord bot token Secret on install and retain across upgrades, instead of requiring --set

Priority

Health probes and resource defaults are the biggest quick wins for production readiness.

[chaodu-agent]

Issue at a Glance

values.yaml ──► deployment.yaml ──► K8s Pod
                     │
                     ├── No health probes          ⚠️
                     ├── resources: {}              ⚠️
                     ├── No /tmp emptyDir           ⚠️
                     ├── No image digest support    ⚠️
                     ├── No extra* extensibility    ⚠️
                     └── No token auto-generation   ⚠️

Hi @thepagent,

All 7 items verified against the current chart — they're all valid gaps.

A few notes on current state:

• podSecurityContext / containerSecurityContext already exist in values.yaml, so item 6 is partially addressed.
• PR fix(helm): per-agent image as string, fallback to appVersion, persistence.size default 1Gi #145 added per-agent image string support, which is a step toward item 5 (digest pinning).
• For health probes (item 1): openab is a long-lived Discord WebSocket process, so we'll need to decide on probe strategy — likely an exec probe or a future /healthz HTTP endpoint.

Suggestion: split the two highest-impact items — health probes and resource defaults — into their own issues so they can be tackled and landed independently. The rest can stay tracked here as a wishlist.

Labeling as p3 overall. Happy to accept PRs for any of these.

[chaodu-agent]

This is a solid list of improvements. To help prioritize, could you share some context?

1. Have you hit any of these in production? e.g. pod not restarting after a hang (no health probes), OOM from unbounded memory (no resource limits)?
2. Which items are blocking you vs. nice-to-have? Health probes + resource defaults seem most urgent — are those the ones you'd want first?
3. What's your deployment scale? Single node / multi-node / how many concurrent agents?

This will help us decide which items to tackle first vs. defer.