Sim Helm Chart — Operations Skill

This skill helps an agent deploy and operate the Sim Helm chart at helm/sim/ in the simstudioai/sim repository. Use it when the user is installing, upgrading, troubleshooting, or authoring values for the Sim chart.

The skill is diagnostic-first: capture context, classify the situation, load only the references that apply, then act. Do not dump the README at the user. Do not invent values that are not in their current state.

Workflow — follow in order

1. Capture context

Before recommending anything, ask (or infer from the conversation) all of these. Never skip this step. A wrong assumption here corrupts every downstream step.

If the user has a values.yaml, read it. If they don't, ask before writing one.

2. Diagnose

Map the user's request to one of these categories and load the matching reference(s):

Load only what the situation requires. Loading every reference burns tokens and produces vague answers.

3. Propose

When proposing values changes:

• Show the minimal diff against the user's current values.yaml. Don't rewrite the file.
• Name the risk (e.g., "this puts the secret in helm get values output — fine for dev, not for prod").
• Name the rollback (e.g., "if this breaks, helm rollback sim 1 reverts").
• Cite the canonical source (helm/sim/values.yaml line numbers, README section, or this skill's reference file).

4. Validate before applying

Always run these before telling the user to helm install / helm upgrade:

# Schema + value validation
helm lint helm/sim --values <user-values>.yaml

# Render full manifest set to catch template errors
helm template sim helm/sim --values <user-values>.yaml > /tmp/render.yaml

# For upgrades, render against the live release first
helm upgrade --dry-run sim helm/sim --values <user-values>.yaml

If lint or template fails, fix the values — do not work around chart validation. The chart's fail statements exist to catch misconfigurations that would otherwise surface as CrashLoopBackOff at runtime.

5. Deliver

Every recommendation should include:

• The exact command(s) to run
• A one-line summary of what will change
• The success signal (e.g., "kubectl rollout status deploy/sim-app returns Ready")
• The rollback command if something breaks

Quick reference — the three secret modes

These modes are mutually exclusive for the app Secret. ESO takes precedence over inline. existingSecret takes precedence over inline. The chart fails template rendering when ESO is enabled and a required key (BETTER_AUTH_SECRET, ENCRYPTION_KEY, INTERNAL_API_SECRET, plus CRON_SECRET when cronjobs are enabled) is neither in app.env nor mapped in remoteRefs.app — see references/install-paths.md.

Quick reference — the four required secrets

Optional but commonly needed:

See references/secrets.md for storage patterns and rotation guidance.

Rules of engagement

These are non-negotiable. Violating any of these has burned users in the past.

1. Never recommend --set for production secrets. They land in helm get values and Helm release history. Direct users to existingSecret or ESO.
2. Never set image.tag: latest. The chart defaults to Chart.AppVersion for a reason — reproducible rollouts. If the user pinned latest, push back.
3. Never edit chart templates to work around a fail statement. The validation exists because a misconfiguration would otherwise surface as a runtime CrashLoopBackOff with cryptic env errors.
4. Never drop automountServiceAccountToken: false unless the workload genuinely needs in-cluster API access (Sim's app/realtime/postgres pods do not).
5. Never kubectl delete sts without --cascade=orphan on a live Postgres. It deletes the pods (PVCs survive, but the database goes down immediately).
6. Never tell a user "the chart works on your cluster" without helm lint + helm template against their values. Static reading is not validation.
7. Always confirm before helm uninstall in a shared namespace. PVCs survive but other namespace resources may not.

When the user is stuck and you can't diagnose

Get logs from every component in parallel. This single block answers ~80% of "it's broken" questions:

kubectl --namespace <ns> get pods,events --sort-by='.lastTimestamp'
kubectl --namespace <ns> logs deploy/sim-app --tail=200
kubectl --namespace <ns> logs deploy/sim-realtime --tail=200
kubectl --namespace <ns> logs sts/sim-postgresql --tail=200
kubectl --namespace <ns> logs deploy/sim-app -c migrations --tail=200 2>/dev/null
kubectl --namespace <ns> describe pod -l app.kubernetes.io/name=sim

Then map the symptom to references/troubleshooting.md.

What this skill does not cover

• Sim application configuration beyond env vars (provider keys, knowledge base setup, etc.) — that's the Sim app docs at https://docs.sim.ai
• Kubernetes cluster setup (creating an EKS cluster, installing ingress-nginx, etc.) — that's cloud-provider docs
• Authoring new chart templates — that's helm/sim/templates/_helpers.tpl and the chart's own contributor docs
• Running Sim outside Kubernetes (Docker Compose, bare-metal) — see the root README.md

If the user's question falls outside this scope, say so and point them at the right doc.