Video Subtitles

Generate SRT subtitles from video/audio with translation support. Transcribes Hebrew (ivrit.ai) and English (whisper), translates between languages, burns subtitles into video. Use for creating captions, transcripts, or hardcoded subtitles for WhatsApp/social media.

Security Scan

The skill's description (transcribe via ivrit.ai and Whisper, burn subtitles) is plausible, but the instructions and assets omit key operational details (no declared credentials, unclear install method, platform-specific tooling) that don't line up with the stated behavior.

Purpose & Capability

The skill claims to use an external Hebrew service (ivrit.ai) and OpenAI Whisper. Neither the SKILL.md nor the registry metadata declare any required API keys or credentials for external services. The README also asks the user to install 'ffmpeg-full' via brew (macOS) while the skill has no OS restriction — this is inconsistent. The 'uv' package manager is nonstandard and not justified.

Instruction Scope

Runtime instructions are limited to running scripts/generate_srt.py with flags (transcribe/translate/burn). That's appropriate for the stated purpose. However SKILL.md mentions automatic model downloads and external models (ivrit.ai, Whisper) without describing where data is sent, whether audio/video is uploaded to remote APIs, or what credentials/tokens are required. That gap could lead to unintended network access or data exfiltration by the script.

Install Mechanism

There is no install spec in the registry (instruction-only), but a Python script is included that will run and reportedly auto-download ~3GB models at runtime. Auto-downloading large model files at runtime from unspecified sources is a risk. The use of 'uv' as the package manager and a brew-only ffmpeg instruction are both unusual and not cross-platform; there is no guidance for Linux/Windows or verified trusted download hosts for models.

Credentials

The skill declares no required environment variables or primary credential, yet it references third-party services (ivrit.ai, OpenAI Whisper) which commonly require API keys. The absence of declared credentials is a mismatch: either the script uses unauthenticated remote services (privacy concern) or it expects keys but fails to document them (operational/permissions concern).

Persistence & Privilege

The skill does not set always:true and does not disable model invocation, so it behaves like a normal user-invoked skill. That is reasonable, but because it will read user-supplied video/audio files and can auto-download large models, you should be aware it may store sizable files on disk and perform network transfers when invoked. There's no explicit higher privilege requested.

What to consider before installing

This skill could be legitimate, but the manifest and README leave important questions unanswered. Before installing or running it: 1) Inspect scripts/generate_srt.py to confirm whether it uploads audio/video to remote APIs (ivrit.ai or OpenAI) and where it downloads model weights from. 2) Ask the author which environment variables or API keys are required and whether keys are ever transmitted or logged. 3) Verify the source URLs for model downloads and ffmpeg instructions; prefer well-known release hosts (GitHub releases, official project domains). 4) Clarify what 'uv' is and get cross-platform install instructions (Linux/Windows). 5) Run the tool in a sandbox/VM/container first to limit disk usage and network access. 6) If you cannot verify these points, do not provide sensitive credentials and consider this skill suspicious.

Generate movie-style subtitles from video or audio files. Supports transcription, translation, and burning subtitles directly into video.

Features

• Hebrew: ivrit.ai fine-tuned model (best Hebrew transcription)
• English: OpenAI Whisper large-v3
• Auto-detect: Automatically detects language and selects best model
• Translation: Translate Hebrew → English
• Burn-in: Hardcode subtitles into video (visible everywhere, including WhatsApp)
• Movie-style: Natural subtitle breaks (42 chars/line, 1-7s duration)

Quick Start

# Plain transcript
./scripts/generate_srt.py video.mp4

# Generate SRT file
./scripts/generate_srt.py video.mp4 --srt

# Burn subtitles into video (always visible)
./scripts/generate_srt.py video.mp4 --srt --burn

# Translate to English + burn in
./scripts/generate_srt.py video.mp4 --srt --burn --translate en

# Force language
./scripts/generate_srt.py video.mp4 --lang he    # Hebrew
./scripts/generate_srt.py video.mp4 --lang en    # English

Options

Flag	Description
--srt	Generate SRT subtitle file
--burn	Burn subtitles into video (hardcoded, always visible)
--embed	Embed soft subtitles (toggle in player)
--translate en	Translate to English
--lang he/en	Force input language
-o FILE	Custom output path

Output

• Default: Plain text transcript to stdout
• With --srt: Creates video.srt alongside input
• With --burn: Creates video_subtitled.mp4 with hardcoded subs

Requirements

• uv: Python package manager (auto-installs dependencies)
• ffmpeg-full: For burning subtitles (brew install ffmpeg-full)
• Models: ~3GB each, auto-downloaded on first use

Subtitle Style

• Font size 12, white text with black outline
• Bottom-aligned, movie-style positioning
• Max 42 chars/line, 2 lines max
• Natural breaks at punctuation and pauses