seo-image-gen
Fail
Audited by Gen Agent Trust Hub on Jul 27, 2026
Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The
references/prompt-engineering.mdfile contains a dedicated section titled "Safety Filter Rephrase Strategies" which provides explicit instructions on how to evade AI safety filters (e.g., bypassing blocks for violence, public figures, and NSFW content) through abstraction and context shifting. - [COMMAND_EXECUTION]: The skill frequently instructs the agent to execute shell commands. This includes high-privilege commands like
sudo apt install imagemagickinreferences/post-processing.md, and various image manipulation pipelines usingmagickandffmpeg. These commands often incorporate file paths derived from tool outputs without explicit sanitization. - [REMOTE_CODE_EXECUTION]: The skill requires the execution of an installation script located at
./extensions/banana/install.sh, which is not part of the provided skill files. Additionally,references/cost-tracking.mdandreferences/presets.mdreference external Python scripts (cost_tracker.py,presets.py) that are expected to be present in the user's environment but are not provided, posing a risk of executing unverified code. - [EXTERNAL_DOWNLOADS]: The documentation directs users to install external software and MCP servers from unverified GitHub repositories (
github.com/YCSE/nanobanana-mcpandgithub.com/AgriciDaniel/banana-claude).
Recommendations
- AI detected serious security threats
Audit Metadata