Architecture of kajisho5/ffmpeg-skill
kajisho5/ffmpeg-skill is 300 source files holding 4,798 symbols across 10 modules, written mainly in Python, Shell, Dockerfile. The largest module is scripts, with 61 files, and 3 files are entry points into the dependency graph.
How ffmpeg-skill fits together
The installer at bin/install.js is the package's executable entry point, and it copies the skill folder of instructions, scripts and templates into the skill folders of Claude Code, Cursor and Codex. The core media work happens in the scripts folder, which holds one script per operation such as audio mixing, backgrounds, smart cutting and ASS overlays, and each script runs on its own from the command line. The scripts/render.py script runs a whole project in one pass. The mcp/server.py server lets MCP clients list and run the same ffmpeg tools. The evals/run.py entry point grades how an agent used the skill, using the transcript and files from each saved run. The fuzz folder holds a harness that searches for inputs that crash the JSON parser and the ASR word-timing reader.
Largest modules
- scripts: 61 files and 33,684 lines of code
- tests: 22 files and 41,641 lines of code
- .github: 12 files and 1,224 lines of code
- evals: 7 files and 878 lines of code
- fuzz: 5 files and 106 lines of code
- demos: 3 files and 2,071 lines of code
- .clusterfuzzlite: 2 files and 14 lines of code
- bin: 1 file and 317 lines of code
Languages
By file count: Python 108 files, Shell 5 files, Dockerfile 1 file, JavaScript 1 file.
Documentation
Repowise documents ffmpeg-skill in 339 pages generated from the source, 11 of them written explanations of the repository and its modules. Module guides include:
- fuzz: Fuzzing Harness
- bin: Installer Entry Point
- scripts/_common: Shared Media Helpers
- evals: The Contract Evaluation Suite
- mcp: The MCP Server
More on ffmpeg-skill: summary and FAQ, codebase overview, code health.
Related guide: Best dependency graph tools for monorepos