The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes external security tools (Prowler and ScoutSuite) using the Python subprocess module to perform cloud environment scans.
Evidence: Implementation found in scripts/agent.py and scripts/process.py.
Note: The scripts correctly use list-based argument passing (e.g., subprocess.run(["prowler", ...])) rather than shell strings, which is a recommended practice to prevent shell command injection.
[EXTERNAL_DOWNLOADS]: The skill requires and recommends the installation of external security auditing packages from public registries.
Evidence: Mentions and executes installation of prowler and scoutsuite via pip in SKILL.md and references/api-reference.md.
Note: These are widely recognized open-source security tools (Prowler by Prowler Cloud and ScoutSuite by NCC Group) and their use is consistent with the skill's stated purpose of cloud security posture management.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection because it ingests findings from external cloud environments and processes them for the agent.
Ingestion points: scripts/process.py parses Prowler's JSON-OCSF output, which contains metadata like resource names, finding titles, and descriptions that can be controlled by a malicious actor in the cloud environment.
Boundary markers: Absent. The parsed findings are aggregated into a JSON report that is presented to the agent without specific delimiters or warnings to ignore embedded instructions.
Capability inventory: The skill has file-writing capabilities and can execute system commands via subprocess.
Sanitization: Absent. The script extracts strings directly from the scan output without filtering or escaping content for potential prompt injection patterns.

implementing-cloud-vulnerability-posture-management