skills/mukul975/anthropic-cybersecurity-skills/performing-cloud-penetration-testing/Gen Agent Trust Hub
performing-cloud-penetration-testing
Pass
Audited by Gen Agent Trust Hub on Mar 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [SAFE]: The core functionality of the skill and the provided Python script (
scripts/agent.py) involves standard security enumeration of cloud environments using official APIs (boto3) to identify common misconfigurations. - [COMMAND_EXECUTION]: The skill instructs the agent to execute various cloud CLI commands (AWS, Azure, gcloud) for reconnaissance and testing. While these are powerful commands, they are appropriate for the skill's stated purpose of penetration testing.
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from the cloud environment (such as resource names or configuration values) which are then processed by the agent. Specifically, the
scripts/agent.pyscript outputs these values to the console and a JSON file without sanitization or protective boundary markers. - Ingestion points: AWS API responses for S3 bucket names, security group details, and Lambda environment variables in
scripts/agent.py. - Boundary markers: Absent in the script output.
- Capability inventory: Filesystem write and AWS service enumeration in
scripts/agent.py. - Sanitization: None observed for the data retrieved from cloud APIs.
- [EXTERNAL_DOWNLOADS]: The skill references well-known and reputable third-party security tools such as ScoutSuite, Prowler, and Pacu. These are standard tools in the cybersecurity industry and their use is consistent with the skill's objectives.
Audit Metadata