multi-channel-ad-ideation

Warn

Audited by Gen Agent Trust Hub on Aug 6, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The channels/image-gen/SKILL.md file contains embedded Python scripts for image manipulation, logo composting, and headline rendering using the Pillow and NumPy libraries. These scripts are dynamically constructed and executed by the agent at runtime to process remote chat history.
  • [COMMAND_EXECUTION]: The skill leverages the author's tdx CLI suite (tdx use, tdx chat, tdx llm history) to interact with managed LLM projects and agents. It also utilizes the mcp__tdx-studio__open_file tool to display final campaign assets.
  • [PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface by ingesting user-supplied creative briefs and target segment descriptions. These inputs are interpolated into prompts for external agents without documented sanitization or boundary enforcement. Ingestion points occur in SKILL.md and are passed to the image-gen skill. Capabilities include file system writes and local script execution.
  • [EXTERNAL_DOWNLOADS]: The image generation workflow involves locating and loading system-level fonts and relies on the availability of external libraries such as Pillow and numpy.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 6, 2026, 06:00 AM
Security Audit — agent-trust-hub — multi-channel-ad-ideation