skills/oubakiou/skills/dataviz-svg/Gen Agent Trust Hub

dataviz-svg

Warn

Audited by Gen Agent Trust Hub on Jul 3, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The scripts/vl2svg.ts script (and its bundled version vl2svg.mjs) implements a custom data loader that permits arbitrary local file access.
  • The resolveLocalDataPath function supports absolute paths and directory traversal sequences (e.g., ../), which allows the tool to read any file the system process has permissions for, such as system configurations or credentials.
  • This capability is triggered whenever a Vega-Lite JSON specification contains a data.url property pointing to a local file.
  • [COMMAND_EXECUTION]: The script performs unrestricted file writes based on provided command-line arguments without path validation.
  • The renderSvg and renderPng functions in scripts/vl2svg.ts use writeFileSync with paths derived directly from user-controlled arguments, allowing for the potential overwriting of critical system files (e.g., ~/.bashrc, /etc/hosts) if the agent is manipulated.
  • [EXTERNAL_DOWNLOADS]: The data loader in scripts/vl2svg.ts is configured to fetch data from external URLs.
  • The createDataLoader function delegates requests for non-file protocols to the default Vega loader, enabling the rendering process to make outbound network requests to any domain, which could be used for basic SSRF or data exfiltration.
  • [PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection because it processes untrusted data from external sources.
  • Ingestion points: The skill ingests untrusted data from URLs and local files via the data.url property in scripts/vl2svg.ts during chart rendering.
  • Boundary markers: Absent. The skill does not implement any delimiters or instructions for the agent to ignore or isolate content found within ingested data files.
  • Capability inventory: The skill has powerful system capabilities, including arbitrary file reading, arbitrary file writing, and outbound network access.
  • Sanitization: Absent. No validation or sanitization is performed on the data paths or the content of the data files before they are processed by the Vega runtime.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 3, 2026, 03:27 AM
Security Audit — agent-trust-hub — dataviz-svg