email-creative-builder
Pass
Audited by Gen Agent Trust Hub on Aug 20, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted data from external sources, which presents a surface for indirect prompt injection targeting the AI agent.
- Ingestion points: The skill scrapes content from external destination URLs and processes user-provided CSV exports or pasted emails as described in SKILL.md.
- Boundary markers: The instructions explicitly warn the agent to treat these inputs as untrusted and ignore any instructions embedded within them, although no technical delimiters are enforced on the data itself.
- Capability inventory: The skill is configured to execute a local script (registry-events.py) and perform file-write operations to the local memory/ directory.
- Sanitization: High-level stylistic cleaning is mentioned (humanizer-slop.md), but there is no technical sanitization or validation of the untrusted input content for security purposes.
- [COMMAND_EXECUTION]: The skill is designed to invoke a local script named registry-events.py to propose updates to a claims registry. This script execution is a documented part of the skill's logic for maintaining marketing claim integrity.
Audit Metadata