The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs several third-party security tools and cloud CLIs during the setup process.
Evidence: pip install scoutsuite prowler-cloud pacu awscli in SKILL.md.
Note: These are well-known, legitimate packages from the official Python Package Index used for security auditing.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from cloud environment resources and tool outputs, which are interpolated into a final findings report without sanitization.
Ingestion points: scoutsuite_run.log, Prowler outputs, and direct cloud CLI query results (e.g., IAM policy names, bucket ACLs) in SKILL.md and REFERENCE.md.
Boundary markers: The generated findings report in SKILL.md lacks delimiters or instructions to the agent to ignore potentially malicious embedded content.
Capability inventory: The skill has extensive capabilities to execute shell commands and cloud CLI operations for AWS, Azure, and GCP.
Sanitization: Tool outputs and cloud metadata are piped directly into the report file without any escaping, filtering, or validation logic.

cloud-pentest