subdomain-takeover
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute various network diagnostic, security, and cloud management commands.
- Evidence: Mentions of
dig,curl,aws s3 mb,heroku create,subjack,nuclei, andsubzyin sections 2, 3, and 4. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources and maintains the capability to execute commands based on that data, creating an attack surface where external content could influence agent behavior.
- Ingestion points: DNS records and HTTP response bodies fetched via
digandcurlas described in detection and confirmation procedures (SKILL.md). - Boundary markers: Absent. There are no instructions for the agent to use delimiters or ignore instructions embedded in the external data it retrieves.
- Capability inventory: High-privilege command execution capability including cloud resource creation (
aws,heroku) and DNS resolution tools. - Sanitization: Absent. No mention of sanitizing or validating external data before it is processed or used in subsequent commands.
Audit Metadata