The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to "redesign" or "refine" existing Draw.io diagrams, which involves reading and processing XML content. If these files contain malicious instructions within cell value attributes, the agent could be influenced by them.
Ingestion points: The agent reads existing .drawio files (XML format) as instructed in the workflow (e.g., assets/learning-roadmap.drawio).
Boundary markers: No specific boundary markers or instructions are provided to the agent to treat XML cell content as untrusted data or to ignore instructions embedded within the diagram text.
Capability inventory: The agent has the capability to execute a bundled Python script (scripts/validate_drawio.py), invoke the drawio CLI tool for exports, and write files to the workspace.
Sanitization: The provided validation script (scripts/validate_drawio.py) checks for XML structural integrity and duplicate IDs but does not perform any sanitization or filtering of the text content within the diagrams for prompt injection patterns.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute a locally provided Python script and an external CLI tool.
Evidence: The workflow in SKILL.md requires running python3 scripts/validate_drawio.py <file.drawio>. Additionally, references/drawio-xml.md suggests using the drawio CLI for exporting diagrams to SVG or PNG formats.

drawio-roadmap-ppt