k8s-triage

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill is primarily composed of kubectl shell commands for cluster diagnostics. This is the intended functionality of the tool and is constrained by instructions to verify permissions using kubectl auth can-i before proceeding.
  • [CREDENTIALS_UNSAFE]: The instructions contain specific safety guardrails forbidding the agent from reading Kubernetes Secrets and requiring it to redact passwords, tokens, or keys found in any command output before reporting to the user.
  • [INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests untrusted data from the cluster environment, such as pod names and event messages. However, this risk is mitigated by explicit instructions to 'Read first, ask before changing' and the requirement for user verification of all commands that modify the cluster state.
  • Ingestion points: Pod names, statuses, and event messages are read via kubectl get pods -A and kubectl events -A (SKILL.md).
  • Boundary markers: Absent for command output processing.
  • Capability inventory: State-changing capabilities include kubectl uncordon, kubectl rollout undo, and kubectl delete pod (SKILL.md).
  • Sanitization: Explicit instruction to ignore/redact secrets and tokens from output (SKILL.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 03:37 AM