The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides the execute_command(command) tool in references/tools.md. This allows the AI agent to execute arbitrary shell commands on the host system, which is a high-risk capability that can be abused for unauthorized operations beyond the stated scope of cloud auditing.
[DATA_EXFILTRATION]: The skill is designed to collect and analyze sensitive cloud configuration and IAM data using tools like prowler_scan, scout_suite_assessment, and cloudmapper_analysis. While this is part of the stated purpose, the combination of these tools with execute_command provides a direct pathway for harvesting and exfiltrating cloud credentials and infrastructure secrets.
[PRIVILEGE_ESCALATION]: The pacu_exploitation tool mentioned in references/tools.md is an AWS exploitation framework specifically designed for privilege escalation, IAM exploitation, and data exfiltration within cloud environments.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources and infrastructure which could be manipulated by an attacker.
Ingestion points: Data enters the agent context through the outputs of prowler_scan, trivy_scan, kube_hunter_scan, and falco_runtime_monitoring in references/tools.md.
Boundary markers: The instructions do not define boundary markers or clear separators to distinguish between tool results and instructions.
Capability inventory: The skill has access to arbitrary command execution (execute_command), cloud exploitation (pacu_exploitation), and local file writes via tool output parameters (e.g., output_dir in prowler_scan).
Sanitization: There is no evidence of sanitization, filtering, or validation of the data returned by the security scanning tools before it is processed by the agent.

pentest-cloud-infrastructure