norman-fig2ppt

Warn

Audited by Gen Agent Trust Hub on Oct 14, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill ships with a complete bundled Python runtime for multiple architectures (macOS arm64/x86_64) and uses "author bytecode" rather than source code for its core engine. The Windows version is a Nuitka-recompiled standalone binary. This prevents static auditing of the engine's internal logic.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and reconstruct scientific figures, screenshots, and diagrams. It explicitly instructs the agent to "Preserve the source image's text" and "distinguish editable text," which creates a vector for instructions embedded within processed images to potentially influence the agent's behavior without sanitization or explicit boundary markers.
  • [COMMAND_EXECUTION]: The instructions in SKILL.md require the agent to use shell tools (PowerShell on Windows and /bin/sh on macOS) to invoke the bundled binaries with various arguments, including path names and JSON metadata. This grants the skill direct execution capabilities on the host system.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 14, 2026, 06:21 AM