gpt-image-markup-edit
Pass
Audited by Gen Agent Trust Hub on Nov 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes JSON specification files (e.g.,
examples/nook.json) to generate image markups. Thelabelandnotefields within these JSON files are directly interpolated into the prompts presented to the agent and the downstream image model. - Ingestion points: Specification files passed to
scripts/markup.py(e.g., via workspace files or generated content). - Boundary markers: Absent; instructions in
SKILL.mddirect the agent to use the output lines of the script verbatim in the image tool prompt. - Capability inventory: Access to
curlfor API interaction,python3for image generation, andmagickfor diffing images. - Sanitization: No sanitization or escaping is performed on the
labelornotetext before it is printed and subsequently used in prompts. - [COMMAND_EXECUTION]: The skill relies on several shell-based tools to perform its tasks.
- Evidence: Uses
python3to run thescripts/markup.pyimage processing script,magick(ImageMagick) to verify image changes viamagick compare, andcurlto upload files to the varg.ai API. - [EXTERNAL_DOWNLOADS]: The skill references and interacts with external domains for image hosting and API services.
- Evidence: Interacts with
s3.varg.aiandmcp.varg.ai. These resources belong to the vendor (vargHQ) and are used for legitimate image storage and model invocation.
Audit Metadata