email-protocol-attack

Fail

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: CRITICALINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSOBFUSCATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze data from external mail servers (e.g., banner grabbing, user enumeration, and protocol capabilities) using tools such as nc, swaks, and smtp-user-enum. This data is externally controlled and could contain prompt injection strings intended to manipulate the agent's reasoning.
  • Ingestion points: Shell command outputs from nc, smtp-user-enum, swaks, and openssl in SKILL.md and accompanying guides.
  • Boundary markers: Absent; the skill does not define specific delimiters to separate untrusted tool output from agent instructions.
  • Capability inventory: The skill has access to shell execution (Bash), file system operations (Write, Edit), and network capabilities (WebFetch, WebSearch).
  • Sanitization: No sanitization or validation logic is present for the output gathered from remote network services.
  • [EXTERNAL_DOWNLOADS]: The skill includes links to book.hacktricks.xyz for technical documentation on email pentesting. While these are widely used resources in the security community, automated scanners have flagged these paths as potentially associated with malicious patterns.
  • [OBFUSCATION]: The skill contains pedagogical examples of homoglyph domains (e.g., аpple.com using the Cyrillic 'а') within SKILL.md to demonstrate how attackers bypass email authentication filters. While used for education, these patterns represent the use of confusable characters to imitate legitimate targets.
Recommendations
  • Contains 4 malicious URL(s) - DO NOT USE
Audit Metadata
Risk Level
CRITICAL
Analyzed
Sep 19, 2026, 04:36 PM