dataviz-svg
Fail
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: HIGHDATA_EXFILTRATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill's path resolution logic in
scripts/vl2svg.tsis vulnerable to directory traversal. TheresolveLocalDataPathfunction usespath.resolve(base, uri)without validating that the resulting path remains within the intended directory. This allows the agent to read arbitrary files (e.g.,../../etc/passwd) by specifying them in thedata.urlproperty of a Vega-Lite JSON specification. - [COMMAND_EXECUTION]: The skill permits writing files to arbitrary locations. The
renderSvgandrenderPngfunctions inscripts/vl2svg.tsaccept user-provided output paths and callmkdirSync(outDir, { recursive: true })andwriteFileSync(path.resolve(outputPath), ...)without any path validation or restriction, enabling the overwriting of sensitive system files or configuration. - [DATA_EXFILTRATION]: The skill provides a clear path for data exfiltration by combining arbitrary local file access with the ability to make network requests to non-whitelisted domains via the Vega data loader.
- [EXTERNAL_DOWNLOADS]: The skill fetches data from external URLs during the rendering process using the built-in Vega loader. It also references and uses assets from well-known sources such as the
resvg-jsandgoogle-fontsGitHub repositories. - [DYNAMIC_EXECUTION]: The skill performs dynamic execution of binary code by initializing a WebAssembly module (
resvg.wasm) at runtime using theinitWasmfunction inscripts/vl2svg.ts. - [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection by processing untrusted external data that is subsequently rendered into images for the agent to interpret.
- Ingestion points: Data is ingested from local files or remote URLs defined in the
data.urlfield of the Vega-Lite specification processed byvl2svg.ts. - Boundary markers: No explicit delimiters or boundary markers are used to separate untrusted data from the rendering instructions.
- Capability inventory: The skill has access to
readFileSync,writeFileSync,mkdirSync, and network fetching capabilities. - Sanitization: There is no evidence of sanitization, filtering, or validation performed on the ingested data or the resolved file paths.
Recommendations
- AI detected serious security threats
Audit Metadata