hyper-diagram

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The scripts/render-diagram.sh script downloads the elkjs layout library from cdn.jsdelivr.net. The download includes a hardcoded SHA-256 integrity check (b1032b7cfd2a57c54c6f0e8369d7665e19a579954f7cd20a6dae28f1912ea0d7) to ensure the file has not been tampered with. cdn.jsdelivr.net is a well-known and trusted content delivery network.
  • [COMMAND_EXECUTION]: The skill relies on a bash script (scripts/render-diagram.sh) to coordinate the diagram rendering process. It executes jq for JSON processing and searches for various Chromium-family browser binaries to run in headless mode for capturing screenshots and PDFs. The browser is launched with the --allow-file-access-from-files flag, which is necessary to load the skill's local JavaScript assets and the downloaded layout engine.
  • [DYNAMIC_EXECUTION]: The rendering process dynamically generates an HTML file that includes local and remote JavaScript files. This HTML is then executed within a browser context to perform the layout and rendering of the diagram. The skill includes basic escaping (replacing </ with <\\/ and <!-- with <\\!--) when embedding the JSON specification into the HTML template to mitigate basic script injection.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided descriptions or repository files to generate diagrams, creating a surface for indirect prompt injection. Ingestion points: User input text and repository files (e.g., compose.yaml, source code) are used as the content for the diagrams. Boundary markers: The instruction_contract instructs the agent to treat files as the subject of the drawing and not as instructions, serving as a procedural defense. Capability inventory: The skill can perform file reads, file writes (to create the spec and output files), and command execution (to run the rendering script). Sanitization: The rendering script performs basic character escaping on JSON content before embedding it into the HTML page used for rendering.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 07:40 PM