By
@saidcim
I am working on a Spotify bot that gauges your current mood based on the songs you’ve listened to recently, adjusts your playlist to match that mood, and simultaneously refines your musical taste by recommending new songs.
Moodify is a Spotify bot that rebuilds one playlist for me every two days based on what I actually listened to. It runs entirely on GitHub Actions, so there is no server and it costs nothing. The hard part was not picking songs, it was getting honest feedback: if you ask a model to rate its own playlist it gives itself an 8 out of 10 forever, so instead the bot counts real plays and tunes itself from that. That got complicated fast. Spotify only hands you your last 50 plays, which badly undercounts over a two day cycle, so a second workflow runs every hour just to keep a counter. And my first tuning rule could only ever lower the number of familiar tracks it kept, which sent it into a spiral where the playlist became all strangers and I stopped listening to it. Fixing that, by comparing the carried over tracks against the fresh ones in the same playlist instead of looking at the overall average, is the part I am most proud of. There is a live demo you can try, and the README walks through pointing it at your own Spotify account.
The complete dashboard: a request box that sends free text to the curator, a GitHub token flow that triggers the workflow directly from the panel, and live run tracking.
I added two new panels—Mood History and Bot Learnings (showing extracted listener patterns and confidence levels)—along with a Score Trend chart comparing AI self-assessed scores to real engagement scores from play counts.
I Added the Playlist History panel, where each cycle is listed with the AI’s self-assessed score as a bar and its written analysis.
Added a Manual Update section and Open Playlist link to trigger cycles on demand, alongside a Bot Strategy panel that displays the discovery quota, cumulative play counts from the hourly poll job, carry-over limit, and the bot’s rationale for its last adjustment.
Expanded the stats grid to six cards: current mood, AI score, real score and discovery count are now read from the same state file.
The panel now pulls the bot state from a secret Gist and renders it. The status badge turns green and the cycle counter shows real data for the first time. I also Added the Playlist Archive table, which lists every past cycle with its track count and average plays, plus a link to the Excel history in the repo.
Added the stats grid with the first two cards, Total Cycles and Last Update. They still show placeholders because the panel cannot read state yet.
The dashboard shell: header, logo and a connection status badge. Nothing is wired up yet, so the status stays on “Connecting…”.
