The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external data formats including SVG source, PPTX XML, and Mermaid diagrams. This functionality introduces a surface for indirect prompt injection where malicious instructions could be embedded in the metadata or text labels of the source files to influence the agent's behavior during the restyling process.
Ingestion points: As defined in SKILL.md, the skill reads raw source code and structure from SVG, PPTX, and Mermaid inputs provided by the user.
Boundary markers: The instructions explicitly require the agent to extract the structural elements (nodes, connections, groups) and present them as a list for user verification before any modifications are applied.
Capability inventory: The skill utilizes xml.etree.ElementTree for asset parsing and executes local Python scripts for design validation.
Sanitization: The skill enforces strict numerical and accessibility gates via derive.py and check_fit.py, which validate the final output against WCAG contrast ratios and layout constraints.
[COMMAND_EXECUTION]: The skill workflow involves running local Python scripts (derive.py, check_fit.py) to verify theme colors and layout fit. These scripts use standard libraries and operate on local data to ensure the design remains within safe parameters. The execution of these scripts is a core, legitimate part of the skill's validation logic.

visual-craft