gpt-image-sticker-pack

Pass

Audited by Gen Agent Trust Hub on Nov 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill relies on scripts/export.sh to process images into platform-specific formats. This script executes various command-line utilities including ffmpeg, ffprobe, magick, and gifsicle. The script is properly implemented with variable quoting to mitigate command injection risks.
  • [DATA_EXFILTRATION]: Instructions within SKILL.md direct the agent to upload generated image files to s3.varg.ai. This domain is a verified vendor resource belonging to vargHQ and is a necessary part of the sticker generation workflow.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied images as inputs for the gpt_image_2_6 model. While this constitutes an attack surface for indirect prompt injection, it is the fundamental purpose of the skill, and no specific exploitative patterns or unsafe interpolation of untrusted text were observed.
  • Ingestion points: User-provided image files (e.g., ./me.jpg) are passed to the image tool via the files parameter.
  • Boundary markers: None identified; image models generally do not use text-based delimiters for visual input.
  • Capability inventory: The skill has access to the local filesystem and shell command execution via the Bash tool to run the export scripts.
  • Sanitization: No explicit sanitization of image data is performed by the skill's logic, relying instead on the model provider's internal safety filters.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 14, 2026, 09:49 PM