The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external inputs such as research papers, user descriptions, and screenshots to generate editable diagrams. This inherent functionality creates an attack surface for indirect prompt injection.
Ingestion points: The skill ingests content from prompts, papers, and codebases to determine labels and entities for the diagrams (as detailed in SKILL.md and references/drawio-workflow.md).
Boundary markers: The instructions emphasize fidelity to the source and provide a diagram brief structure, although they do not implement strict technical delimiters for all input types.
Capability inventory: The skill possesses the capability to write files, execute local Python scripts for validation, and use browser automation for screenshots.
Sanitization: The skill relies on structured XML generation using Python's xml.etree.ElementTree and includes helpers for escaping rich text labels to mitigate injection into the diagram metadata.
[EXTERNAL_DOWNLOADS]: The scripts/check_skill_update.py script connects to GitHub's raw content service to fetch version information from the developer's repository. This is a standard and transparent update check targeting a well-known service.
[COMMAND_EXECUTION]: The skill requires the execution of its own bundled Python scripts for geometric validation (validate_visual_quality.py), diagram sanity checking (validate_drawio.py), and local preview serving (serve_drawio_preview.py). These executions are part of the skill's core functionality and operate within the project context.

drawio-diagram-builder