The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs a runtime installation of the cairosvg package using pip install -q cairosvg during the rendering step. While this is a well-known library, runtime package installation is a noteworthy behavior for security auditing.
[COMMAND_EXECUTION]: The skill utilizes shell command blocks to automate the rendering pipeline. It uses cat to save SVG files, python3 -c to execute inline Python logic for SVG-to-PNG conversion and draw.io XML wrapping, and calls system utilities like rsvg-convert.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data, such as research reports and user-provided paper content, to extract scientific entities for diagram generation. This introduces a vulnerability surface where malicious instructions embedded in the source text could attempt to influence the LLM's SVG generation or the logic in the subsequent processing scripts.
Ingestion points: Step 1 parses research context, report text, and user descriptions from the agent's environment.
Boundary markers: The skill does not explicitly define delimiters for the input data, although it forces the output into a structured JSON schema before SVG generation.
Capability inventory: The skill possesses the ability to write to the file system and execute shell/Python commands.
Sanitization: The skill employs urllib.parse.quote() to sanitize the SVG content before embedding it into the draw.io XML structure, and the use of single-quoted heredocs ('SVGEOF') prevents shell expansion of the generated SVG content.

svg-scientific-figures