The Agent Skills Directory

[COMMAND_EXECUTION]: The Python scripts scripts/agent.py and scripts/process.py utilize the subprocess module to execute kubectl commands locally. This is used to retrieve pod configuration data from the active Kubernetes cluster for auditing purposes.
[INDIRECT_PROMPT_INJECTION]: The skill processes pod metadata (such as names and namespaces) retrieved from the cluster environment. This creates a potential vulnerability surface where maliciously named resources in a cluster could attempt to influence an AI agent's interpretation of the audit results.
Ingestion points: Pod specifications retrieved via kubectl get pods -o json in scripts/agent.py and scripts/process.py.
Boundary markers: The scripts do not use specific delimiters or instructions to treat pod metadata as untrusted when generating output.
Capability inventory: The skill possesses the capability to run system commands and read cluster-wide resource configurations.
Sanitization: No sanitization or escaping is performed on pod metadata before it is printed to standard output.
[METADATA_POISONING]: There is a discrepancy between the author identified in the SKILL.md frontmatter ("mahipal") and the copyright holder listed in the LICENSE file ("mukul975"), indicating inconsistent metadata.

detecting-privilege-escalation-in-kubernetes-pods