seo-image-gen
Pass
Audited by Gen Agent Trust Hub on Apr 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute varios local Python scripts for cost tracking (
cost_tracker.py), preset management (presets.py), and a fallback generation method (generate.py) located in~/.claude/skills/seo-image-gen/scripts/.- [COMMAND_EXECUTION]: Provides detailed instructions for invoking system-level utilities such as ImageMagick (magick),ffmpeg, andpotracefor post-generation image processing. It also includes user-led installation instructions involvingsudo apt installfor these dependencies.- [EXTERNAL_DOWNLOADS]: References a local installation script./extensions/banana/install.shand an external MCP server package (@ycse/nanobanana-mcp) from GitHub which are required to enable the skill's primary functionality.- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface (Category 8) due to the interpolation of untrusted user data into shell command strings. - Ingestion points: User-supplied descriptions in image generation commands and custom brand preset configurations stored in local JSON files.
- Boundary markers: While documentation shows command arguments wrapped in double quotes (e.g.,
--prompt "..."), there are no explicit instructions for the agent to sanitize or escape shell metacharacters within the user-provided text. - Capability inventory: The agent can execute shell commands, run Python scripts, and perform network requests to the Gemini API through the
gemini_generate_imageMCP tool. - Sanitization: No validation, filtering, or escaping logic is defined in the instructions to protect against potential command injection when processing external input.
Audit Metadata