coreweave-common-errors

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from Kubernetes pod events and resource descriptions via kubectl describe pod and kubectl get nodes. This data could contain malicious strings designed to manipulate the agent's logic.\n
  • Ingestion points: Kubernetes resource events and pod descriptions (SKILL.md).\n
  • Boundary markers: No specific delimiters or boundary instructions are used when interpolating command outputs into the prompt.\n
  • Capability inventory: The skill possesses the ability to modify Kubernetes secrets and resources using Bash(kubectl:*) (SKILL.md).\n
  • Sanitization: The instructions include a policy to use 'redacted evidence sets,' which helps mitigate data exposure but does not sanitize the input against prompt injection.\n- [COMMAND_EXECUTION]: The skill provides instructions for creating Kubernetes secrets (kubectl create secret) using environment variables for credentials. While it includes a safety warning not to print or paste these credentials, it involves the agent in sensitive administrative write operations.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 03:46 AM
Security Audit — agent-trust-hub — coreweave-common-errors