The Agent Skills Directory

[DYNAMIC_EXECUTION]: The scripts scripts/build_figure.py, scripts/overlay_labels.py, and examples/poster_substrate.py utilize the importlib.util module to dynamically load and execute Python code from computed relative paths. Specifically, the _load_module and _load_validate_fonts functions are used to load components like compose.py, export.py, and validate_fonts.py from a sibling skill directory (scientific-figure).
[COMMAND_EXECUTION]: The examples/poster_substrate.py script uses subprocess.run to execute other Python scripts within the skill's directory structure (such as generate_figure.py and overlay_labels.py) using the system's Python executable.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting data from external files and user arguments which are interpolated into model prompts.
Ingestion points: scripts/build_figure.py and scripts/generate_figure.py read configuration and subject data from --spec, --theme, and positional command-line arguments.
Boundary markers: There are no explicit delimiters or instructions provided to the model to ignore potentially malicious content embedded within the user-supplied strings during prompt construction.
Capability inventory: The skill environment includes capabilities for file system writes, shell command execution via subprocesses, and dynamic Python module execution.
Sanitization: The skill performs basic validation for specific fields (like hex colors and image sizes) but lacks general sanitization for the natural language content used in prompt construction.

ai-full-figure