chat-with-anyone
Pass
Audited by Gen Agent Trust Hub on Mar 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches video and subtitle content from third-party platforms including YouTube, Bilibili, and TikTok to extract reference audio samples.
- [COMMAND_EXECUTION]: The skill invokes several subprocesses to perform its tasks, including
ffmpegfor audio extraction and trimming, and local Python scripts for video downloading and text-to-speech generation. - [PROMPT_INJECTION]: The skill is subject to indirect prompt injection via the processing of untrusted external data.
- Ingestion points: The agent reads content from downloaded subtitle files (.vtt or .srt) in
SKILL.md(Workflow A4). - Boundary markers: There are no explicit delimiters or instructions to ignore embedded prompts within the subtitle files.
- Capability inventory: The skill possesses the ability to execute shell commands (
ffmpeg), run local scripts, and generate synthetic audio content. - Sanitization: No validation or filtering is applied to the subtitle text before the agent parses it for timestamps.
- [DATA_EXFILTRATION]: The script
voice_design.pyaccesses a configuration file at~/.noiz_api_keyto authenticate requests to thenoiz.aiAPI. This is a vendor-specific credential used for intended functionality.
Audit Metadata