converting-email-templates

Pass

Audited by Gen Agent Trust Hub on May 5, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute a local Python script (scripts/converter.py) to perform deterministic template conversions. The script is bundled with the skill and uses standard libraries for string manipulation.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it processes untrusted HTML templates provided by users.
  • Ingestion points: Raw HTML template files or inline HTML pasted into the chat (SKILL.md Workflow Steps 1 and 3).
  • Boundary markers: No explicit boundary markers or 'ignore' instructions are used when passing template content to the LLM for conversion.
  • Capability inventory: The agent can read and write files on the local filesystem via the conversion script and shell access.
  • Sanitization: While the script uses regex for safety-neutral patterns, the 'Deep mode' allows the LLM to process arbitrary HTML without specific sanitization against embedded instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
May 5, 2026, 05:35 PM
Security Audit — agent-trust-hub — converting-email-templates