The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data for figure generation, which creates an attack surface for indirect prompt injection where malicious instructions could be embedded in data labels or titles.
Ingestion points: Data input parameters in kernel.py functions, such as ymat in bar_with_points, values in strip_with_median, and coordinate data in end_of_line_labels.
Boundary markers: No specific delimiters or safety instructions are defined to separate untrusted data from the plotting and rendering logic.
Capability inventory: The skill is capable of writing files to the disk (savefig) and instructions suggest the use of dynamic execution (exec) for loading local code.
Sanitization: No input validation or sanitization is performed on the data-driven labels or values before they are rendered into visual artifacts.
[DYNAMIC_EXECUTION]: The documentation in SKILL.md recommends loading the provided helper functions using the exec() function on the local kernel.py file. While the code itself is provided with the skill, exec() is a dynamic execution sink that bypasses standard module import safety mechanisms and could be exploited if the source file is manipulated.

figure-style