kubectl
Pass
Audited by Gen Agent Trust Hub on Nov 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill allows the agent to ingest data from Kubernetes clusters which may contain untrusted or adversarial content intended to influence agent behavior.
- Ingestion points: The agent reads output from
kubectl logs,kubectl get, andkubectl describe(SKILL.md). - Boundary markers: Absent. There are no specific instructions or delimiters provided to prevent the agent from following instructions embedded in log streams or resource configurations.
- Capability inventory: The skill exposes high-impact tools including
kubectl apply(filesystem/cluster writes),kubectl delete(resource destruction), andkubectl exec(arbitrary command execution). - Sanitization: Absent. Data retrieved from the cluster is processed without filtering or validation for malicious prompts.
- [COMMAND_EXECUTION]: The skill documents and enables the use of
kubectl exec, which allows the agent to spawn interactive shells or run arbitrary commands inside remote containers. - [DATA_EXFILTRATION]: The skill provides commands such as
kubectl get secretsandkubectl describe secretwhich allow the agent to access sensitive credentials and configuration data stored within the Kubernetes cluster. - [EXTERNAL_DOWNLOADS]: The skill uses the
install_commandto fetch the officialkubectlbinary via the Homebrew package manager (brew install kubectl). This targets a well-known service and repository.
Audit Metadata