lov-publish-wechat-channels
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes a local Python script (
scripts/check_video.py) to perform a pre-upload validation check on video files. This script safeley invokes theffprobeutility usingsubprocess.runwith a list of arguments to extract metadata without modifying the source files. - Evidence: Found in the execution flow instructions in
SKILL.mdand therun_ffprobefunction withinscripts/check_video.py. - [INDIRECT_PROMPT_INJECTION]: As the skill processes and displays user-provided video metadata (titles, descriptions, and tags) within a browser automation context, there is a theoretical surface for indirect prompt injection. The skill includes explicit instructions for the agent to verify all data against the user's intent through a 'read back' process before final submission, significantly reducing this risk.
- Ingestion points:
SKILL.md(User input section for video metadata). - Boundary markers: Mandatory回读 (read-back) verification steps in
SKILL.md. - Capability inventory: Browser automation and execution of local validation scripts.
- Sanitization: The workflow relies on logical verification against confirmed session history rather than automated string sanitization.
Audit Metadata