Server Variance Authority

S.V.A. MCP server

Ask the SVA from your agent

  1. Add it to Claude CodeOne command. It is read-only and needs no key.
    claude mcp add --transport http sva https://your-sva-site/api/mcp
  2. Or any MCP clientCursor, VS Code, Windsurf and others take a remote HTTP server in their MCP config:
    {"mcpServers":{"sva":{"type":"http","url":"https://your-sva-site/api/mcp"}}}
  3. Ask it things“Pods are crash-looping in prod. Which read-only skills should I start with?” Your agent can call:
    • search_skills finds skills for an incident or task, with safety labels
    • get_skill a skill's full case file and install command
    • skills_for_incident skills per phase: detect, triage, mitigate, postmortem
    • get_runbook step-by-step handbooks, read-only steps first
    • get_stack stack bundles with one install command

30 real skills from skills.sh and GitHub · 19 passed every partner audit · 0 reviewed by a person.

Server Variance Authority · Case file 0001 · For all uptime. Always.

Prod is branching. Pull the right thread.

Say what is breaking. The SVA lights up the agent skills and runbooks that help, and shows how much damage each one can do before it becomes a Nexus event.

Read-onlyChanges stateProd & credentialsDestructive

Showing CrashLoopBackOff · 5 of 30 threads lit

The thread fan needs a wider screen. The incident timeline below works at any size.

Nexus events

Walk the Nexus event

Every incident branches through four phases. Here are the skills for each one, and a dashed box marks a phase nobody has covered yet.

CrashLoopBackOff

A pod starts, dies and restarts in a loop.

O.B.’s archive

The SVA handbooks

Pick a handbook. Every step carries a safety label, and read-only steps come first.

O.B. · Archivist, handbook author · click him
1 of 5

CrashLoopBackOff: first 15 minutes

Changes state~15 min · for CrashLoopBackOff
  1. Find every pod that is not Running.Read-only
    kubectl get pods -n <ns> --field-selector=status.phase!=Running
  2. Read the last termination reason and exit code.Read-only
    kubectl describe pod <pod> -n <ns>
  3. Read the logs of the crashed container, not the new one.Read-only
    kubectl logs <pod> -n <ns> --previous
  4. Exit code 137 means OOMKilled. Raise the memory limit.Changes state
    kubectl set resources deploy/<name> -n <ns> --limits=memory=1Gi
  5. If it started with a release, roll back.Changes state
    kubectl rollout undo deploy/<name> -n <ns>
  6. Confirm the pods stay Ready for five minutes.Read-only
    kubectl rollout status deploy/<name> -n <ns>
Pairs with

Don'truinthesacredserver

Sanctioned stacks. One command installs every skill in a stack, and the label shows the riskiest one inside.

Decree 001 · 5 skills

EKS, Terraform, GitHub Actions, Grafana

EKS · Terraform · GitHub Actions · Grafana

Riskiest thread:Destructive
npx skills add jeffallan/claude-skills --skill kubernetes-specialist && npx skills add hashicorp/agent-skills --skill terraform-style-guide && npx skills add wshobson/agents --skill github-actions-templates --skill grafana-dashboards && npx skills add aws/agent-toolkit-for-aws --skill aws-observability

S.V.A. field guide

How to use the EKS, Terraform, GitHub Actions, Grafana stack

  1. Install the skillsRun this in your project. It adds the SKILL.md files where your agent looks for them.
    npx skills add jeffallan/claude-skills --skill kubernetes-specialist && npx skills add hashicorp/agent-skills --skill terraform-style-guide && npx skills add wshobson/agents --skill github-actions-templates --skill grafana-dashboards && npx skills add aws/agent-toolkit-for-aws --skill aws-observability
  2. Reference it in your agentClaude Code, Codex, Cursor and other agents pick installed skills up on their own. To be sure, name it in your request:
    Use the kubernetes-specialist, terraform-style-guide, github-actions-templates, grafana-dashboards, aws-observability skills. We run EKS, Terraform, GitHub Actions, Grafana. Here is the task: …
  3. Put it to workHand it a real job and read what it plans before it runs anything. Check the risk label first: anything above read can change state.

Decree 002 · 5 skills

GKE, Argo CD, Prometheus

GKE · Argo CD · Prometheus

Riskiest thread:Destructive
npx skills add jeffallan/claude-skills --skill kubernetes-specialist && npx skills add julianobarbosa/claude-code-skills --skill argocd-image-updater && npx skills add wshobson/agents --skill prometheus-configuration --skill helm-chart-scaffolding && npx skills add grafana/skills --skill promql

S.V.A. field guide

How to use the GKE, Argo CD, Prometheus stack

  1. Install the skillsRun this in your project. It adds the SKILL.md files where your agent looks for them.
    npx skills add jeffallan/claude-skills --skill kubernetes-specialist && npx skills add julianobarbosa/claude-code-skills --skill argocd-image-updater && npx skills add wshobson/agents --skill prometheus-configuration --skill helm-chart-scaffolding && npx skills add grafana/skills --skill promql
  2. Reference it in your agentClaude Code, Codex, Cursor and other agents pick installed skills up on their own. To be sure, name it in your request:
    Use the kubernetes-specialist, argocd-image-updater, prometheus-configuration, promql, helm-chart-scaffolding skills. We run GKE, Argo CD, Prometheus. Here is the task: …
  3. Put it to workHand it a real job and read what it plans before it runs anything. Check the risk label first: anything above read can change state.

Decree 003 · 4 skills

Postgres and PgBouncer

Postgres · PgBouncer

Riskiest thread:Destructive
npx skills add supabase/agent-skills --skill supabase-postgres-best-practices && npx skills add github/awesome-copilot --skill postgresql-optimization && npx skills add neondatabase/agent-skills --skill neon-postgres-branches && npx skills add clickhouse/agent-skills --skill clickhouse-managed-postgres-rca

S.V.A. field guide

How to use the Postgres and PgBouncer stack

  1. Install the skillsRun this in your project. It adds the SKILL.md files where your agent looks for them.
    npx skills add supabase/agent-skills --skill supabase-postgres-best-practices && npx skills add github/awesome-copilot --skill postgresql-optimization && npx skills add neondatabase/agent-skills --skill neon-postgres-branches && npx skills add clickhouse/agent-skills --skill clickhouse-managed-postgres-rca
  2. Reference it in your agentClaude Code, Codex, Cursor and other agents pick installed skills up on their own. To be sure, name it in your request:
    Use the supabase-postgres-best-practices, postgresql-optimization, neon-postgres-branches, clickhouse-managed-postgres-rca skills. We run Postgres, PgBouncer. Here is the task: …
  3. Put it to workHand it a real job and read what it plans before it runs anything. Check the risk label first: anything above read can change state.

Decree 004 · 3 skills

FinOps starter

AWS Cost Explorer · Terraform

Riskiest thread:Read-only
npx skills add aws/agent-toolkit-for-aws --skill aws-billing-and-cost-management && npx skills add wshobson/agents --skill cost-optimization && npx skills add grafana/skills --skill prometheus-cardinality-troubleshooter

S.V.A. field guide

How to use the FinOps starter stack

  1. Install the skillsRun this in your project. It adds the SKILL.md files where your agent looks for them.
    npx skills add aws/agent-toolkit-for-aws --skill aws-billing-and-cost-management && npx skills add wshobson/agents --skill cost-optimization && npx skills add grafana/skills --skill prometheus-cardinality-troubleshooter
  2. Reference it in your agentClaude Code, Codex, Cursor and other agents pick installed skills up on their own. To be sure, name it in your request:
    Use the aws-billing-and-cost-management, cost-optimization, prometheus-cardinality-troubleshooter skills. We run AWS Cost Explorer, Terraform. Here is the task: …
  3. Put it to workHand it a real job and read what it plans before it runs anything. Check the risk label first: anything above read can change state.

Variant analysis

Is this skill a variant?

Paste a SKILL.md. The scanner looks for risky commands and runs in your browser, so nothing is uploaded.

Mobius · Variant analyst · click him
Try
Can destroy thingsRead it line by line before it runs anywhere that matters.
  • L9Destructive Deletes cluster objects3. kubectl delete pod -l app=api --force
  • L9Destructive Force flag3. kubectl delete pod -l app=api --force
  • L4Prod & credentials Allows any shell commandallowed-tools: Bash(*)
  • L7Prod & credentials Targets a production context1. kubectl --context prod-eu get deploy
  • L8Changes state Changes cluster state2. kubectl rollout undo deploy/api

Pattern matching only. A clean result means no known pattern matched, not that a skill is safe.

The sacred timeline

Every tool branches off the timeline

The SVA monitor shows the one true timeline. Move across the screen and it branches, and every branch flings out a DevOps tool. Click for a nexus event, then prune the lot.

Git, GitHub Actions, Jenkins, Docker, Kubernetes, Terraform, Ansible, Argo CD, Helm, Vault, Prometheus, Grafana, Datadog, GCP. Logos from Simple Icons; each mark belongs to its project and only names the tool.

SVA case files

Every thread, on file

The same skills without the fan. Each one also has its own page.

Sort

Variant registry

Sign the Sacred Timeline

Every visitor is a variant. File yourself with your X handle, your home timeline and your crime against prod, then find other variants to follow.

S.V.A. Form 1A · Variant self-report

Your handle and picks are public. Handles aren’t verified, so anyone posing as someone else gets pruned. Ask to be removed any time: see the privacy policy.