docx-design-agent
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses shell commands for environment setup and dependency management, including
pip installcalls. It also utilizesosascriptto communicate with Microsoft Word for live document editing and finalization tasks. - [DYNAMIC_EXECUTION]: The agent dynamically generates and executes Python scripts using the
python-docxlibrary to create and style documents, as well as AppleScript to perform live IPC operations such as updating fields and exporting to PDF. - [EXTERNAL_DOWNLOADS]: During installation and first run, the skill fetches well-known and standard Python libraries (
python-docx,lxml,Pillow) from official package registries. - [INDIRECT_PROMPT_INJECTION]: A potential attack surface exists in workflows designed to edit or redesign existing documents, as the agent reads and catalogs content from
.docxfiles which could contain adversarial instructions. - Ingestion points: Paragraph and table reading functionality described in
references/applescript-patterns.md. - Boundary markers: No specific delimiters or safety instructions are defined for the imported content.
- Capability inventory: Full file write access via
python-docxand application control viaosascript. - Sanitization: No evidence of content sanitization or validation before processing document text.
Audit Metadata