The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute scientific processing commands. Specifically, it provides instructions for generating schematics using a local Python script and converting PDF files into images. The script scripts/generate_schematic.py uses the subprocess module to manage these tasks.
[INDIRECT_PROMPT_INJECTION]: The skill processes external manuscripts, research papers, and grant proposals. These materials could contain hidden instructions designed to manipulate the AI's review conclusions or exfiltrate the OPENROUTER_API_KEY. Ingestion points: Manuscript text reading and presentation slide analysis (SKILL.md). Boundary markers: No explicit instructions to ignore embedded prompts in documents. Capability inventory: Network access (OpenRouter API), file system write access, and shell execution (Bash tool). Sanitization: No validation or sanitization of document content is performed before processing by LLMs.
[DYNAMIC_EXECUTION]: The script scripts/generate_schematic.py uses subprocess.run to dynamically invoke scripts/generate_schematic_ai.py at runtime. While these are components of the same skill, the pattern of dynamic script invocation represents an expanded attack surface.

peer-review