pptx-official
Pass
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes external command-line utilities for document processing and validation. \n
- Evidence: Found in
ooxml/scripts/pack.py,ooxml/scripts/validation/redlining.py, andscripts/thumbnail.py, which invoke tools likesoffice,pdftoppm, andgit. \n - Context: The execution pattern is safe as it uses argument lists to prevent shell injection and limits file access to the local workspace and temporary directories. \n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its core function of analyzing content from external files. \n
- Ingestion points: Untrusted text and XML are extracted from user-supplied presentations via
markitdownandooxml/scripts/unpack.py. \n - Boundary markers: There are no instructions to use delimiters or ignore potential commands within the extracted presentation content. \n
- Capability inventory: The skill has broad permissions, including file system access and shell command execution, which could be targeted by adversarial document content. \n
- Sanitization: XML parsing is hardened against structure-based attacks, but the semantic content is not sanitized before being processed by the AI agent.
Audit Metadata