ppt-component-atlas
Warn
Audited by Gen Agent Trust Hub on Jul 16, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The script
scripts/export-component-html.mjscontains logic to download a JavaScript file from a remote repository (https://raw.githubusercontent.com/WiseWong6/wise-labs/main/html-ppt-components/catalog-data.js) and execute it using the Node.jsvm.runInNewContextmethod when the--verify-sourceflag is invoked. - [COMMAND_EXECUTION]: The skill's operational flow involves the agent executing a local Node.js script (
scripts/export-component-html.mjs) via the system shell, where user-supplied queries are passed directly as command-line arguments. - [EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch catalog updates from the author's public GitHub repository, which is a well-known service but remains an external dependency for code execution.
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and processes component descriptions and labels from an external catalog file. If the catalog data (either local or remote) were to be modified maliciously, it could influence the agent's behavior during the selection and generation process.
- Ingestion points: The
public/catalog-data.jsfile and the remote GitHub source URL. - Boundary markers: No explicit delimiters or instructions are used to prevent the agent from obeying natural language instructions embedded within the catalog data.
- Capability inventory: The skill possesses the ability to execute shell commands, perform network fetches, and write files to the local filesystem.
- Sanitization: While the script performs basic text normalization for matching purposes, it does not include robust sanitization or escaping for the data that is subsequently presented to the agent.
Audit Metadata