← ClaudeAtlas

megafanlisted

Fill in metadata for concert and live-music discs in a Plex library that runs no metadata agent, where the filename is the only thing Plex knows and every field must be written by hand and locked. Researches the disc's catalogue number across MusicBrainz, Discogs, CDJapan and Wikipedia, finds and visually checks cover art, then applies release year, release date, summary, setlist and poster. Also names chapters after songs, fills chapter thumbnails, and handles English subtitles for a disc that ships none. Works for live releases from any band and any country, and is built around the awkward cases: catalogue numbers written inconsistently across sources, databases that disagree about a release date, and imported discs that ship no English subtitle track. Handles one disc or a whole queue, and a RESCAN — re-checking a disc already in the library against better scans of the object. Runs as a COORDINATOR that spawns one worker agent per stage, so research, cover hunting, subtitles and chapter work proceed in par
vinnybouchard/Megafan · ★ 0 · Data & Documents · score 67
Install: claude install-skill vinnybouchard/Megafan
## Megafan — concert-disc metadata (research → apply → LOOK) For the Plex **Concerts** library (section 5, `/mnt/media/Concerts`), which runs `tv.plex.agents.none`. No agent means **the filename is the only metadata Plex has** and it fetches no artwork, so everything below is written by hand and `.locked=1` — an unlocked field gets re-derived on the next scan, and with nothing to derive it from Plex stamps the scan date. That is how a 2013 release came up as 2026. **You do not rip anything.** The rip is a `.bat` you or someone else double-clicks on Windows. Your job starts once the file is in the library. Every rule here was learned from a disc that broke it, and the measurement is kept beside the rule rather than summarised away — which is why so many carry a date and a catalogue number. **Read the reasoning before working around a rule**, and don't compress it out: rediscovering it is the expensive part. ### How this skill runs: one coordinator, parallel workers One session used to do all of this in one context. It ended up carrying 160 booklet tiles it no longer needed while making the run's hardest judgement calls — does this Discogs entity describe THIS artist, did a page change who it is — and it took hours, serially, on stages that share nothing. So this file is the **coordinator**: it does the cheap serial work (§0–§2.5), writes `job.json`, and spawns one **fresh worker agent per stage**. Each stage's procedure lives in `stages/<stage>.md` with its section number