视频字幕自动生成器——免费的才是最好的

• Outcome: Pass
• Latest audit: May 29, 2026 · 8:28 PM UTC
• Version: 2.1.1

Overview

This is a local video subtitle tool whose risks are mainly ordinary media-processing, privacy, and reliability issues rather than hidden or malicious behavior.

Use this in a private working directory, process only recordings you are allowed to transcribe, and review/delete generated audio, transcript, subtitle, and video files when done. Install faster-whisper if you need real transcription, verify the output before publishing, and avoid untrusted subtitle style files or unusual filenames until FFmpeg filter escaping is hardened.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Output Handling: Unvalidated Output Injection, Cross-Context Output, Unbounded Output
• Behavioral AST: exec() Call, eval() Call, Dynamic Import
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands

Findings (9)

subprocess module call

Medium

• Category: Dangerous Code Execution

Content

]
    
    try:
        result = subprocess.run(cmd, capture_output=True, text=True, timeout=600, encoding='utf-8')
        if result.returncode == 0:
            file_size = Path(output_path).stat().st_size / 1024 / 1024
            print(f"       ✅ 视频已生成：{output_path} ({file_size:.2f} MB)")

• Confidence: 79% confidence
• Finding: While the subprocess itself avoids shell injection, the code constructs an ffmpeg `subtitles` filter string from a file path with only partial escaping. FFmpeg filter parsing has its own metacharacters and protocol behaviors, so a crafted subtitle path could break filter syntax or trigger unintended file/protocol access when this skill processes attacker-controlled paths.

Intent-Code Divergence

Medium

• Confidence: 97% confidence
• Finding: If Whisper is unavailable, the tool silently generates fabricated transcript content that looks like legitimate speech recognition output. In a workflow that users may trust for subtitles, summaries, or downstream decisions, this can cause integrity failures, misinformation, or accidental publication of false content without clear warning.

Missing User Warnings

Medium

• Confidence: 84% confidence
• Finding: The skill processes spoken audio into transcripts and subtitles, which can expose sensitive personal, corporate, or meeting information if users do not realize speech content is being extracted and persisted as text. In a video-processing workflow, this omission is more significant because transcription materially changes the sensitivity and searchability of the data and may create compliance or privacy issues.

Missing User Warnings

Low

• Confidence: 80% confidence
• Finding: The README lists output files but does not clearly warn users up front that the tool writes multiple derived artifacts, including extracted audio, transcripts, subtitles, metadata, and a rendered video, to disk. This can lead to unintended data persistence, disk exposure, or accidental sharing of sensitive content, especially on shared systems or synced folders.

Missing User Warnings

Medium

• Confidence: 88% confidence
• Finding: The skill description presents automated transcription, subtitle generation, and derivative video creation as normal behavior but does not warn that it will create and store transcripts, subtitle files, and modified media outputs. This can expose sensitive spoken content, personal data, or confidential media in local output folders or downstream systems without informed user consent.

Missing User Warnings

Medium

• Confidence: 86% confidence
• Finding: The installation/dependency section references speech recognition and optional external tools such as Whisper, Google/Azure engines, yt-dlp, and other media tooling without warning that video/audio content may be processed by third-party software or services. In practice, users may route sensitive media through cloud-backed or externally integrated tools without understanding the privacy, retention, or compliance implications.

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: The markdown provides copy-pastable destructive PowerShell commands (`Remove-Item ... -Force` and recursive directory deletion) that permanently delete files and folders without any explicit warning, confirmation step, or backup guidance. In a skill-packaging guide, users may execute these commands directly, and if paths or filenames are wrong or adapted unsafely, this can cause unintended data loss on the local system.

Missing User Warnings

Low

• Confidence: 78% confidence
• Finding: The document instructs users to regenerate subtitles and reburn the video, but it does not warn that output paths may overwrite prior generated files or replace existing media artifacts. In a media-processing skill this is not code execution, but it can still cause unintended data loss or confusion if users reuse output directories or filenames.

Unvalidated Output Injection

High

• Category: Output Handling

Content

]
    
    try:
        result = subprocess.run(
            cmd, 
            capture_output=True, 
            timeout=600

• Confidence: 72% confidence
• Finding: User-controlled subtitle paths and style configuration are interpolated into the ffmpeg -vf subtitles filter string with only partial escaping. ffmpeg filter syntax is sensitive to quotes, brackets, and special characters, so a crafted subtitle path or style value could break out of the intended filter context, alter ffmpeg processing behavior, or cause unintended file access/processing within ffmpeg.

VirusTotal

67/67 vendors flagged this skill as clean.

Static analysis

No suspicious patterns detected.