00-email-marketing-guardrails
Pass
Audited by Gen Agent Trust Hub on Jul 23, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No malicious patterns, such as credential exfiltration, obfuscation, or unauthorized network access, were detected in the skill instructions or associated scripts.
- [COMMAND_EXECUTION]: The
eval/run_harness.pyscript usessubprocess.run()to manage Docker containers. This is used exclusively for creating isolated, network-disabled testing environments to evaluate the skill's adherence to policy. The script includes checks to ensure the runner executable remains within the repository structure. - [PROMPT_INJECTION]: The skill processes potentially untrusted email content and recipient lists, which introduces an indirect prompt injection surface. However, the skill acts as a defensive policy layer to block non-compliant data.
- Ingestion points: Data enters the agent context via email content drafts and address source records referenced in
SKILL.md. - Boundary markers: The skill does not explicitly define delimiters to separate untrusted content from instructions.
- Capability inventory: The skill is limited to making decision outputs (e.g.,
BLOCK,HOLD,SEND) and does not possess capabilities for file writing or network interaction. - Sanitization: No explicit sanitization or filtering is defined for the input data.
Audit Metadata