The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from a Kubernetes environment, such as pod names, namespaces, and cluster events. This represents a potential surface for indirect prompt injection if a malicious actor controls metadata within the monitored cluster. However, this is an inherent risk of infrastructure monitoring tasks.
Ingestion points: External data is ingested through tools like list-pods, get-infra-issues, and get-recent-changes which retrieve strings from the Kubernetes API (SKILL.md).
Boundary markers: None identified in the skill instructions.
Capability inventory: The skill utilizes read-only tools to fetch cluster topology and issues.
Sanitization: The instructions do not specify sanitization for the retrieved strings, relying on the agent's default processing.
[NO_CODE]: The skill consists exclusively of Markdown documentation and configuration metadata. No executable scripts, such as Python, JavaScript, or Shell scripts, are provided within the analyzed file.
[COMMAND_EXECUTION]: The documentation highlights a required misspelling (_precent) for sorting node resource usage. This is documented as a requirement for interacting with the specific API and does not introduce a security vulnerability or command injection risk.

kubesense-infra