html-export
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes HTML files, which represent an untrusted data surface for the AI agent.
- Ingestion points: Positional arguments provided as
<input...>are processed byscripts/lib/to-png.cjs,scripts/lib/to-pdf.cjs, andscripts/lib/to-video.cjs. - Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are used when the agent is instructed to process or verify the exported content.
- Capability inventory: The skill executes subprocesses (
ffmpeginscripts/lib/ffmpeg.cjs), performs file system writes (fs.writeFileSyncinscripts/lib/result.cjs), and controls a browser environment via Playwright. - Sanitization: Input filenames are sanitized in
scripts/lib/paths.cjs, but the content of the HTML is rendered without sanitization. - [DATA_EXFILTRATION]: The skill has a risk of data exposure through local file inclusion. As explicitly noted in the documentation, Chromium runs without an OS sandbox in this implementation, which allows a malicious HTML file to use elements like iframes to embed local files (e.g., SSH keys or credentials) into the generated PNG, PDF, or video files.
- The skill provides an
--offlineflag which mitigates network-based exfiltration, but it does not prevent the exposure of local data into the exported files themselves.
Audit Metadata