Extract every visible song from music library screenshots and find the best matching YouTube page.
This is an open-source experiment for learning how to build reliable Codex agent workflows.
The agent:
- reads every visible music row
- extracts the song title and artist
- searches for a matching YouTube page
- creates one CSV row per song
- checks whether the final result is complete
- Place your screenshots inside
samples/images/. - Start Codex:
codex --search- Ask Codex:
Run Music Finder
- The complete pipeline will execute automatically.
Music Finder is built as a modular AI agent pipeline.
The workflow consists of four stages:
- Extract songs
- Search YouTube
- Verify matches
- Audit the final CSV
The pipeline is orchestrated through prompts/run-all.md.
The individual stage prompts are implementation details and normally do not need to be run manually.
AGENTS.md defines the project rules and behavior used throughout the pipeline.
music-finder/
├── .gitignore
├── AGENTS.md
├── LICENSE
├── README.md
├── prompts/
│ ├── run-all.md
│ ├── 01-extract-songs.md
│ ├── 02-search-youtube.md
│ ├── 03-verify-results.md
│ └── 04-audit-csv.md
├── samples/
│ └── images/
├── examples/
│ └── example-results.csv
└── results/