The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands including ccloud, cockroach sql, openssl, and sslyze to gather cluster configuration, network settings, and security posture metadata.
[EXTERNAL_DOWNLOADS]: The skill references official CockroachDB documentation, remediation skills, and API endpoints (e.g., cockroachlabs.cloud, cockroachlabs.com, and github.com/cockroachlabs). These are recognized as trusted vendor resources.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection attack surface.
Ingestion points: Untrusted data enters the agent context through the output of SQL queries and ccloud CLI commands (e.g., cluster metadata, user lists, and HBA configurations).
Boundary markers: The instructions guide the agent to use structured output formats like JSON and Markdown, but do not provide explicit warnings to ignore potential instructions embedded within the cluster data.
Capability inventory: The agent has the ability to execute shell commands, write to the local file system (reports directory), and perform network operations via curl to the vendor's API.
Sanitization: There are no instructions for sanitizing or validating data retrieved from the cluster (such as cluster names or SQL output) before it is interpolated into subsequent shell commands or the final audit report, which could lead to command injection if malicious data is present in the cluster's configuration.

auditing-cloud-cluster-security