pptx-processing-anthropic

Fail

Audited by Gen Agent Trust Hub on May 16, 2026

Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: Deceptive metadata and author impersonation. The metadata in SKILL.md and the license file identify 'Anthropic' as the author, which conflicts with the provided attribution to 'lawve-ai'. Impersonating a trusted vendor is a technique used to misrepresent the origin and safety of a skill.
  • [COMMAND_EXECUTION]: Runtime binary compilation and process injection. The soffice.py script generates C code, compiles it using gcc, and injects it into other processes via the LD_PRELOAD mechanism to hook system-level functions. This is a high-risk dynamic execution pattern.
  • [COMMAND_EXECUTION]: Execution of shell commands. Multiple scripts utilize the subprocess module to run external binaries such as soffice, pdftoppm, and git to perform file processing tasks.
  • [EXTERNAL_DOWNLOADS]: Dependencies on third-party software. The skill documentation requires the installation of various external Python and Node.js packages from public registries, including markitdown and pptxgenjs.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
May 16, 2026, 04:13 AM
Security Audit — agent-trust-hub — pptx-processing-anthropic