bridge-wang/creator-skills

creator-skills is a toolbox of AI skills built by a content creator named Bridge, based on habits picked up while making videos for social media. It packages repetitive, error prone steps of the creation workflow, like topic selection, filming, editing, and subtitling, into tasks you can trigger with a single natural language request. Each skill works on its own and does not depend on the others. Right now the project ships one skill, with more planned. The included skill, called bri video srt, handles the full path from raw footage to finished subtitles. Give it a phone recorded voiceover video and it removes filler pauses without touching picture quality, transcribes the audio locally, corrects the transcript, and formats it into a subtitle file ready to import into editing software. Give it an already edited video or audio file and it transcribes and corrects automatically. Give it an existing subtitle file and it fixes typos and proper nouns, merges or resplits lines by meaning, and adjusts timestamps. All audio and video processing happens on your own machine using whisper.cpp, with nothing uploaded to the cloud. Installation is meant to happen through an AI agent such as Claude Code or Codex: you simply ask the agent to install the skill from its GitHub URL, and it clones the repository, copies the skill files into the right folder, and runs a setup check script that reports any missing dependencies. Alternatives include a dedicated skills command line tool or the Claude Code plugin marketplace. Required dependencies are ffmpeg, whisper.cpp, python3, and a roughly 1.6 gigabyte Whisper speech model, an optional tool called auto editor is only needed for the filler pause removal step. The project targets macOS with Apple Silicon for best performance, though transcription and correction also work on Linux and Windows. Users can customize a glossary of proper nouns and names that keep getting misheard, protect certain phrases from being split across subtitle lines, and adjust how aggressively filler pauses get removed. The project is released under the MIT license, so it can be used, modified, and shared freely.