The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to clone a repository from an untrusted source and install its dependencies.
Evidence: git clone https://github.com/YuhangChen1/Paper2All.git and pip install -r requirements.txt found in SKILL.md and references/installation.md.
[COMMAND_EXECUTION]: The skill facilitates the execution of Python scripts downloaded from an external, unverifiable source.
Evidence: Commands to run pipeline_all.py, pipeline_light.py, and scripts/generate_schematic.py are featured in SKILL.md and references/usage_examples.md.
[PRIVILEGE_ESCALATION]: The installation guide requires the use of administrative privileges to install system dependencies.
Evidence: sudo apt-get install libreoffice poppler-utils in references/installation.md.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted documents while maintaining broad system capabilities, creating a vulnerability to malicious instructions embedded in research papers.
Ingestion points: The input-dir containing user-provided LaTeX or PDF papers as described in SKILL.md.
Boundary markers: Absent; there are no instructions for the agent to treat the contents of the papers as untrusted or to ignore embedded instructions.
Capability inventory: The skill is configured with Read, Write, Edit, and Bash tools in the frontmatter, which are used to process and generate files.
Sanitization: Absent; no logic is described for filtering, escaping, or validating the content extracted from processed academic papers before it influences the agent's generation pipeline.

paper-2-web