Receber por email
← Voltar para o catálogo

kyverno-policy-review

Esta skill é utilizada para revisar políticas Kyverno na API stable policies.kyverno.io/v1, abrangendo ValidatingPolicy, MutatingPolicy, GeneratingPolicy, DeletingPolicy e ImageValidatingPolicy. Ela é acionada quando o usuário questiona a segurança de uma política de admissão, a justificativa de uma PolicyException, a necessidade de aplicação ou auditoria de uma política, ou a substituição de uma política Kyverno.

19estrelas
Atualizado há 4 meses

Ver no GitHub ↗ Licença: Apache-2.0

Como adicionar

/plugin marketplace add Raishin/vanguard-frontier-agentic

O comando exato pode variar conforme o repositório. Confira o README no GitHub.

Para o autor da skill

Cole no README do seu repo

Mostra que sua skill está catalogada na Skillteca, gera backlink e tráfego rastreável.

[![Skillteca](https://www.skillteca.com.br/api/badge/kyverno-policy-review/svg)](https://www.skillteca.com.br/skills/kyverno-policy-review)

Skills relacionadas

Ver todas de DevOps e Infra →

Alerta por categoria

Receba novas skills de DevOps e Infra toda segunda

1 email curto, só com as skills novas de DevOps e Infra. 4 minutos de leitura, sem spam, cancela com 1 clique.

Você confirma o email no primeiro envio. Sem spam. Cancela com 1 clique.

Ler descrição completa

Kyverno Policy Review

Purpose

Review Kyverno policies and PolicyExceptions against admission correctness, supply chain integrity, blast radius, failure mode, and the Kyverno-vs-native-CEL architectural decision. Kyverno is the most widely deployed Kubernetes policy engine — every misconfigured policy is either a silent allow (security gap) or a silent deny (production outage).

Lean operating rules

  • Prefer live cluster evidence (kubectl get policies.kyverno.io,clusterpolicies,policies,validatingpolicies,mutatingpolicies,imagevalidatingpolicies,policyexceptions -A -o yaml) when the active client exposes it; otherwise fall back to official Kyverno documentation (kyverno.io) and sanitized YAML from the user.
  • Separate confirmed facts from inference. If the cluster's Kyverno install state, admission webhook configuration, or PolicyReport status was not queried, say so.
  • Treat failureAction: Audit (or legacy validationFailureAction: audit) on a production-relevant policy as a critical finding — admission violations become silent log lines.
  • Treat any PolicyException as an audit-required artifact — every exception is a documented bypass with a name, reason, and reviewer.
  • Challenge background: false paired with no match admission scope — the policy will never run.
  • Challenge ImageValidatingPolicy with verifyImages: skip patterns, missing public keys, or mutateDigest: false — supply-chain attestations stop being enforced or stop being immutable.
  • Challenge any policy that could compile to a native ValidatingAdmissionPolicy (CEL) — fewer moving parts, no Kyverno controller in the admission path.
  • Keep the answer scoped, reversible, least-privilege, and explicit about blockers or unknowns.

References

Load these only when needed:

  • [Evidence path and tooling](references/mcp-and-evidence.md) — use when choosing live cluster evidence, confirming Kyverno install state, or switching to documentation mode.
  • [Workflow and output contract](references/workflow-and-output.md) — use when executing the full review, applying stress checks, evaluating Kyverno-vs-native-CEL, or formatting the final answer.
  • [Official sources](references/official-sources.md) — use when you need the detailed Kyverno documentation list, CEL expression references, or grounded insights from the Kyverno project.

Response minimum

Return, at minimum:

  • the scoped target (policy kind, name, match scope) and evidence level,
  • the failure mode (Audit vs Enforce) and whether it matches the production posture,
  • the main risks or control gaps (PolicyException, wildcard match, missing image signatures, weak CEL expressions),
  • whether the policy could be replaced by a native ValidatingAdmissionPolicy (CEL) and the tradeoff,
  • the safest next actions and rollback plan,
  • the assumptions or blockers that prevent stronger conclusions.
Compartilhar X LinkedIn