Initial commit: /vibe-sing skill

A Claude Code skill that ends a session with a song. Reads the current
session transcript via CLAUDE_CODE_SESSION_ID, asks Gemini for a
user-centric music prompt (no corny tech references), calls Google Lyria
to compose with vocals, and auto-plays via afplay.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: harajlim

.env.example

@@ -0,0 +1,3 @@
+GOOGLE_API_KEY=your-google-ai-studio-key-here
+# Optional: override the Gemini model (defaults to gemini-3-flash-preview)
+# VIBE_SING_GEMINI_MODEL=gemini-3-pro-preview

.gitignore

@@ -0,0 +1,6 @@
+.env
+.venv
+generations/
+__pycache__/
+*.pyc
+.DS_Store

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 MH (@harajlim)
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -0,0 +1,67 @@
+# /vibe-sing
+
+A Claude Code skill that ends your session with a song.
+
+Pipeline: **Claude Code session transcript → Gemini (mood translation) → Google Lyria 3 (music) → auto-play.**
+
+- `/vibe-sing` — 30-second clip (default)
+- `/vibe-sing pro` — ~2-minute full song
+
+## Why it isn't corny
+
+Both the transcript filter and the Gemini system prompt explicitly forbid references to programming, files, libraries, bugs, or anything technical. The output is a *cinematic mood prompt* — genre, instrumentation, tempo, feel — not a song about your session. A listener should never guess what you were working on.
+
+## Install
+
+Clone directly into Claude Code's skills directory:
+
+```bash
+git clone https://github.com/harajlim/vibe-sing.git ~/.claude/skills/vibe-sing
+cd ~/.claude/skills/vibe-sing
+
+# Python deps
+python3 -m venv .venv
+.venv/bin/pip install -r requirements.txt
+
+# API key — get one at https://aistudio.google.com/apikey
+cp .env.example .env
+$EDITOR .env  # paste your GOOGLE_API_KEY
+```
+
+Then in any Claude Code session, type `/vibe-sing`.
+
+> **Hacking on the source?** Clone anywhere and symlink instead:
+> `ln -s "$(pwd)" ~/.claude/skills/vibe-sing` — edits go live immediately.
+
+## Configuration
+
+Env vars (set in `.env` or shell):
+
+- `GOOGLE_API_KEY` — required.
+- `VIBE_SING_GEMINI_MODEL` — defaults to `gemini-flash-latest` (auto-tracks newest Flash). Pin a version like `gemini-2.5-flash` if you want.
+
+Output mp3s land in `./generations/` (gitignored).
+
+## How it works
+
+1. Finds the JSONL transcript of the current session at `~/.claude/projects/<encoded-cwd>/<session>.jsonl` (picks the most recently modified — i.e. the live session).
+2. Extracts user prompts and assistant prose. Skips tool calls, tool results, thinking blocks, and system reminders. Up to ~100k tokens, tail-biased.
+3. Sends to Gemini with strict instructions: cinematic mood prompt only, no technical references, no specifics, no corniness.
+4. Sends Gemini's prompt to Lyria 3 (clip or pro).
+5. Saves the mp3 and `open`s it (macOS default audio player).
+
+## Files
+
+```
+vibe-sing/
+├── SKILL.md           # instructions Claude follows when /vibe-sing fires
+├── run.sh             # launcher (picks .venv/bin/python or system python3)
+├── vibe_sing.py       # pipeline
+├── requirements.txt   # google-genai, python-dotenv
+├── .env.example       # template for your GOOGLE_API_KEY
+└── generations/       # output mp3s (gitignored)
+```
+
+## Platform
+
+macOS (uses `open` to auto-play). Linux users: swap `open` for `xdg-open` in `vibe_sing.py`.

SKILL.md

@@ -0,0 +1,37 @@
+---
+name: vibe-sing
+description: Generate a song that captures the vibe of the current Claude Code session. Reads the session transcript, asks Gemini for a non-corny mood prompt, calls Google Lyria, and auto-plays the result. Default is a 30s clip; pass `pro` for a ~2 minute version.
+disable-model-invocation: false
+argument-hint: [pro]
+allowed-tools: Bash
+---
+
+When the user invokes `/vibe-sing`, run the pipeline script. It distills the current session's emotional vibe into a Lyria music prompt (via Gemini) and composes a song.
+
+## What to do
+
+1. Run the skill's launcher, passing through the user's argument verbatim:
+
+   ```bash
+   ~/.claude/skills/vibe-sing/run.sh <arg-if-any>
+   ```
+
+   The only meaningful argument is `pro` — generates a ~2 minute song instead of the default 30s clip.
+
+2. The script handles everything: finding the current session's transcript JSONL, slicing the recent vibe-relevant text (skipping tool calls/results), calling Gemini for a mood prompt, calling Lyria, saving the mp3 to `~/.claude/skills/vibe-sing/generations/`, and `open`-ing the file so it plays.
+
+3. After the script returns, the **last line of stdout** is a JSON object with `mood_prompt`, `audio_file`, `model`, and `gemini_model`. Report back to the user in **at most two sentences**:
+   - One sentence quoting the mood prompt Gemini chose.
+   - One sentence with the file path.
+
+   Do **not** describe what the song is "about" — let the audio speak. Do not add commentary about whether the vibe seems right. The user will hear it and decide.
+
+## Errors
+
+If the script fails (missing `GOOGLE_API_KEY`, no transcript, API error), surface the stderr message verbatim and stop — do not retry.
+
+## Notes
+
+- Requires `GOOGLE_API_KEY` in `.env` inside the skill dir (or in the environment).
+- The transcript reader explicitly skips tool calls and tool results to keep the vibe extraction focused on conversational tone.
+- Both the transcript filter and the Gemini system prompt are designed to prevent corny "song about debugging React" output — the mood prompt should never reference programming, files, or specific topics.

requirements.txt

@@ -0,0 +1,2 @@
+google-genai>=0.3.0
+python-dotenv>=1.0.0

run.sh

@@ -0,0 +1,12 @@
+#!/usr/bin/env bash
+set -euo pipefail
+DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PY="$DIR/.venv/bin/python"
+if [ ! -x "$PY" ]; then
+  PY="$(command -v python3 || true)"
+fi
+if [ -z "$PY" ] || [ ! -x "$PY" ]; then
+  echo "vibe-sing: no python interpreter found. Create $DIR/.venv (see README)." >&2
+  exit 1
+fi
+exec "$PY" "$DIR/vibe_sing.py" "$@"