The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to clone and execute code from an external repository not owned by the skill author.
Evidence: Commands such as git clone https://github.com/YuhangChen1/Paper2All.git followed by execution of python pipeline_all.py and python pipeline_light.py allow for the execution of arbitrary code within the agent's local environment.
[EXTERNAL_DOWNLOADS]: The skill pulls software and utilities from multiple unverified external sources.
Evidence: Downloads are initiated from github.com/YuhangChen1 for the core tool and github.com/oschwartz10612 for Windows utilities, in addition to standard package managers.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to perform sensitive system configurations and run complex pipelines.
Evidence: Instructions include environment creation via conda, dependency installation via pip, and system package management using sudo apt-get install.
[INDIRECT_PROMPT_INJECTION]: There is a risk that malicious instructions could be embedded in the academic papers processed by the skill.
Ingestion points: Academic papers in LaTeX or PDF format provided via the input-dir parameter.
Boundary markers: Absent; there are no specified delimiters or warnings for the agent to ignore instructions embedded within the papers.
Capability inventory: The agent has Bash, Write, and Edit tools, which could be exploited if an injection causes it to execute malicious shell commands or modify system files.
Sanitization: No evidence of input validation or content sanitization is provided in the documentation.

alterlab-paper-2-web