video-to-article
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data transcribed from user-uploaded videos, which could contain malicious instructions meant to override agent behavior during the rewriting phase.
- Ingestion points: Transcription files (
transcript.json,transcript.txt) generated from video content inSKILL.md(Step 2). - Boundary markers: Absent. The instructions do not specify the use of delimiters or warnings to ignore instructions embedded within the transcribed text.
- Capability inventory: Execution of local scripts (
asr.py,video_ops.py) and file system write access to theoutputs/directory. - Sanitization: Absent. There is no mention of filtering or escaping the transcribed content before it is processed by the LLM.
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands via Python to run local scripts (
skills/shared/scripts/asr.pyandvideo_ops.py). While these are internal scripts, the use of variable-based output paths (e.g.,outputs/主题名/) introduces a potential risk of command injection or path traversal if the 'subject name' is not properly validated by the platform.
Audit Metadata