The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script assets/comparison_atlas_builder.py dynamically loads Python modules based on a configuration file. It adds a user-specified directory to the Python path using sys.path.insert and executes code via importlib.import_module. This design allows the execution of arbitrary Python code within the project root directory.- [COMMAND_EXECUTION]: The scripts/validate_research_figures.py script invokes external commands using subprocess.run(). Specifically, it calls the pdftoppm utility and allows the user to provide a custom path for the executable, which could be exploited to run arbitrary binaries.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local project environment, including text extraction from PDFs and reading source code or data files. This creates a vulnerability to indirect prompt injection if the ingested content contains malicious instructions intended to manipulate the agent's behavior.
Ingestion points: Automated scanning and reading of source files, data tables, and PDF documents in scripts/inspect_figure_project.py and scripts/validate_research_figures.py.
Boundary markers: There are no markers or instructions provided to the agent to distinguish between its own logic and data content from the analyzed files.
Capability inventory: The skill can execute shell commands and load external Python modules as described in the dynamic and command execution findings.
Sanitization: The skill lacks validation or sanitization of the content extracted from external project files before processing.

paperFig