architecture-diagram-gpt-imagegen
Pass
Audited by Gen Agent Trust Hub on May 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes an instruction to run a "local helper" (Codex CLI) to render images when the user provides an affirmative response. This constitutes a command execution surface where the agent is tasked with invoking external software based on user prompts.
- [PROMPT_INJECTION]: The skill demonstrates an indirect prompt injection surface by ingesting untrusted data and interpolating it into a generated prompt file.
- Ingestion points: The skill accepts Mermaid diagrams, voice descriptions, and project data from the user (SKILL.md).
- Boundary markers: The skill wraps user-provided content in Mermaid code blocks within the output file, providing minimal separation between data and instructions.
- Capability inventory: The agent has the capability to create directories, write files to the filesystem (
~/diagram-prompts/), and execute local commands (SKILL.md). - Sanitization: There are no explicit instructions for the agent to escape, sanitize, or validate the user's input before it is written to the filesystem or processed by the local helper tool.
- [COMMAND_EXECUTION]: The skill generates filenames using a "slug" derived from user-provided topics. If the agent fails to strictly enforce the "kebab-case" instruction, this could potentially be exploited for path traversal or to overwrite sensitive files outside of the intended directory.
Audit Metadata