The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's SKILL.md file contains instructions to install a tool using curl -fsSL https://parallel.ai/install.sh | bash. This represents a high-risk remote code execution pattern as the domain parallel.ai is not recognized as a trusted provider, allowing for the potential execution of malicious scripts without user review.
[DYNAMIC_EXECUTION]: The skill utilizes Python's subprocess.run() function in multiple scripts to execute system-level commands. Specifically, scripts/generate_pdf.py calls pandoc and xelatex with arguments that include user-defined filenames. While this is used for document generation, it creates a surface for command injection if inputs are not strictly validated.
[INDIRECT_PROMPT_INJECTION]: The image generation workflow in scripts/generate_schematic_ai.py takes user-supplied descriptions and embeds them into prompts sent to the OpenRouter API. This creates a surface where malicious instructions could potentially influence the behavior of the AI reviewing the schematic.
Ingestion points: User-supplied diagram descriptions in scripts/generate_schematic.py.
Boundary markers: No explicit delimiters or boundary markers are used to isolate user data within the prompt template.
Capability inventory: The skill possesses the ability to execute subprocesses, perform network requests to external APIs, and read/write to the local file system.
Sanitization: No sanitization or filtering of the user-provided prompt content was identified before it is passed to the external LLM.
[EXTERNAL_DOWNLOADS]: The skill documentation requires the installation of several external dependencies from unverified sources, including the parallel-cli tool and various system packages (pandoc, xelatex) through package managers like brew and apt-get.

literature-review