drawio-skill
Pass
Audited by Gen Agent Trust Hub on Aug 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches brand and technology logos from well-known content delivery networks (unpkg.com and cdn.simpleicons.org) to enrich diagrams with official iconography. This process is used solely for asset retrieval and does not involve the execution of remote code.\n- [COMMAND_EXECUTION]: Multiple scripts automate the use of established local command-line tools. The drawio CLI is utilized for exporting diagrams to various image and document formats, Graphviz (dot and tred) handles graph layout and simplification, and git is used to retrieve historical file versions and identify changes for comparison and timelapse features.\n- [PROMPT_INJECTION]: The skill parses structured data and labels from a wide range of external files, including source code (Python, JS, Go, Rust), infrastructure manifests (Terraform, Kubernetes, Docker), and SQL definitions. This ingested content is used to populate diagram labels and generate structured documentation, forming a surface for indirect prompt injection that is mitigated by standard parsing and XML-escaping practices.\n
- Ingestion points: Found in all bundled importer scripts (e.g., pyimports.py, tfimports.py, sqlerd.py) when reading project files.\n
- Boundary markers: The skill uses standard parsing logic to extract specific metadata and structural identifiers rather than full-text ingestion.\n
- Capability inventory: Limited to local tool execution and documented network asset retrieval.\n
- Sanitization: Content is parsed for structural information, with labels being XML-escaped before insertion into .drawio files.
Audit Metadata