baoyu-infographic
Pass
Audited by Gen Agent Trust Hub on Mar 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes Bash shell commands (
test -f) inSKILL.mdto check for configuration files in the project-local.baoyu-skills/directory and the user's$HOMEdirectory. These commands are restricted to checking file existence for the skill's own preferences and do not process untrusted input or perform sensitive system changes.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8) because it incorporates external, untrusted content into the instructions for an image generation agent.\n - Ingestion points: External data is ingested from user-provided file paths or direct clipboard pastes during the Step 1.2 analysis phase in
SKILL.md, and preferences are loaded fromEXTEND.md.\n - Boundary markers: The prompt template in
references/base-prompt.mdlacks delimiters or protective instructions to prevent embedded commands in the source content from being interpreted by the model.\n - Capability inventory: The skill has file system write access for saving analysis and prompt files, and it can invoke external image generation skills.\n
- Sanitization: No validation or escaping of the source material is performed before it is included in the final generation prompt.
Audit Metadata