The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill defines a workflow that ingests potentially untrusted data from an external search tool.
Ingestion points: Data enters the workflow via the paper-search tool and is saved to /workspace/projects/papers.json.
Boundary markers: The templates do not include explicit markers or delimiters to separate ingested data from instructions in subsequent processing steps.
Capability inventory: The workflow utilizes Python scripts, shell commands, and LaTeX compilation, providing a wide capability surface.
Sanitization: Data cleaning is limited to string truncation in the Python processing script, which does not prevent malicious content from influencing downstream logic.
[Dynamic Execution] (LOW): The skill uses shell heredocs to generate and execute Python scripts at runtime. While these scripts follow static templates, they are assembled and executed dynamically.
[Command Execution] (SAFE): The use of standard utilities such as pdflatex, python3, and cat is appropriate for the skill's stated academic research purpose.

academic-workflow