hwpx
Warn
Audited by Gen Agent Trust Hub on Mar 6, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill is vulnerable to Path Traversal (ZipSlip) and XML External Entity (XXE) attacks when handling untrusted documents.
- Path Traversal: In
scripts/office/unpack.py, the script extracts ZIP entries using their internal filenames without validation (output / entry), which could allow an attacker to write or overwrite files outside the intended directory. - XXE: Multiple scripts, including
analyze_template.pyandbuild_hwpx.py, parse XML usinglxmlwithout disabling entity resolution, potentially allowing unauthorized access to local files via malicious document structure. - [PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection by extracting and processing untrusted content from HWPX documents.
- Ingestion points:
scripts/text_extract.pyandscripts/analyze_template.pyreturn document text directly to the agent. - Boundary markers: Absent.
- Sanitization: No content filtering is applied.
Audit Metadata