chubbyskills: Open-Source Codex Skill for Video Subtitle & Script Extraction

For digital content creators, researchers, and developers analyzing short-form video content, converting speech into verbatim text transcripts is a critical data gathering workflow. The open-source Codex skill 'chubbyskills' (chubbyguan/chubbyskills) has recently surfaced as a streamlined utility capable of transcribing and downloading video audio scripts across major media platforms including Xiaohongshu, Bilibili, Douyin, and WeChat Channels directly within an AI agent environment.

Image source: @Eejoylove via X

Multi-Platform Coverage and Verbatim Transcription Capabilities

chubbyskills functions as a versatile transcription bridge tailored for Chinese video and social ecosystems, addressing platforms that frequently present non-trivial media delivery pipelines. The tool is designed to extract audio transcripts from videos across platforms including Xiaohongshu, Bilibili, WeChat Channels (视频号), and Douyin.

Rather than managing distinct platform extractors and download workflows individually, users can pass video URLs into a single operational interface to extract and download verbatim text transcripts. Additionally, in community thread discussions, individual user feedback noted that the tool could also handle transcription for videos on external platforms such as YouTube, X (formerly Twitter), and TikTok.

Seamless Codex Agent Integration and Direct Installation

A defining characteristic of chubbyskills is its native alignment with the Codex AI agent environment, deliberately minimizing complex compilation and manual environment configuration.

Users can equip their Codex setup simply by providing the GitHub repository link (chubbyguan/chubbyskills) directly to the agent in conversational prompts. Once installed, extracting transcripts requires only passing a target video URL and instructing the agent in natural language to download and transcribe the audio, substantially lowering the friction typically associated with local scraping scripts and transcription pipelines.

Handling Background Music and AI Transcript Refinement

In community thread discussions, users raised questions regarding transcription accuracy on vertical videos featuring prominent background music (BGM). In response, the original post author noted that while they had not yet encountered recognition failures firsthand, users can instruct their connected AI agent like Codex to review and adjust transcripts whenever necessary.

Leveraging the language model's contextual understanding to refine raw transcripts helps streamline post-processing and improves text readability without requiring tedious manual line-by-line edits.

Sources

• GitHub Repository: chubbyguan/chubbyskills
• X Post: 文子🔆 (@Eejoylove)