cut-talking-head
Pass
Audited by Gen Agent Trust Hub on Jul 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes shell commands to interact with the 'chengfeng-videocut' plugin by dynamically identifying its path via 'codex plugin list'. It executes several Node.js scripts, such as 'ensure-runtime.cjs' and 'videocut-cli.cjs', to manage the project lifecycle, project creation, and cutting operations.
- [EXTERNAL_DOWNLOADS]: As part of its setup routine, the 'ensure-runtime.cjs' script is designed to fetch and install necessary runtime components from GitHub Release if they are not present on the system. This is an expected behavior for maintaining the video processing environment.
- [REMOTE_CODE_EXECUTION]: The skill executes components and scripts that are acquired from the vendor's GitHub repository during the installation or update phase, which is required for the tool's core functionality.
- [DATA_EXFILTRATION]: Audio data from local video files is uploaded to a cloud-based Automatic Speech Recognition (ASR) service to perform word-for-word transcription, which is the primary intended function of the skill.
- [PROMPT_INJECTION]: The skill incorporates a process for ingesting and acting upon user-provided transcripts and video content. It manages this surface by using structured JSON schemas and stable 'wordIds' to communicate edit instructions to the underlying CLI tool, reducing the likelihood of malicious text in a transcript influencing agent behavior.
Audit Metadata