The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill clones a repository from a personal GitHub account (https://github.com/YuhangChen1/Paper2All.git) and executes its internal scripts (pipeline_all.py, pipeline_light.py) to process user data. This creates a significant supply chain risk as the code is sourced from an account outside the author's official organization.
[COMMAND_EXECUTION]: Installation instructions require the use of administrative privileges via sudo apt-get install to install system dependencies such as LibreOffice and Poppler utilities, which presents a privilege escalation risk.
[EXTERNAL_DOWNLOADS]: The installation process involves downloading and installing unverified Python dependencies from an external requirements.txt file within the cloned repository.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It ingests untrusted academic papers in LaTeX or PDF format and processes them using LLMs to generate structured content and drive execution pipelines. This ingestion point, combined with Bash execution tools, creates an attack surface if a paper contains malicious instructions designed to hijack the agent's behavior during transformation.
Ingestion points: User-supplied paper directories containing LaTeX files (main.tex) and PDF documents.
Boundary markers: None identified; the system processes extracted text directly through LLM prompts.
Capability inventory: Full Bash, Read, Write, and Edit permissions are used to manage the pipeline and file transformations.
Sanitization: No explicit sanitization or validation of the ingested LaTeX/PDF content is described prior to LLM processing.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to run local Python scripts with complex arguments, including references to non-standard model versions (e.g., gpt-4.1) which may trigger unexpected LLM behavior.

alterlab-paper-2-web