The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/generate_schematic.py uses subprocess.run to execute an internal Python script (scripts/generate_schematic_ai.py). The implementation uses a list of arguments and omits the shell=True parameter, which prevents shell injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill's image and schematic generation scripts (scripts/generate_image.py and scripts/generate_schematic_ai.py) communicate with the OpenRouter API (openrouter.ai). OpenRouter is a well-known service for accessing AI models, and these operations are essential for the skill's stated purpose of providing visual enhancements to scientific documents.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection because it ingests data from external research sources (via research-lookup) and incorporates user input into prompts for image generation. However, this is a common architectural pattern for such tools. Evidence details:
Ingestion points: Research data from the research-lookup skill and user-defined schematic descriptions.
Boundary markers: Absent in the SKILL.md instructions.
Capability inventory: Network requests to AI model providers and file system writes for outputting images.
Sanitization: No explicit sanitization is performed on input strings beyond the standard safe argument handling of subprocess.run.

scientific-writing