# AGENTS.md

This file provides guidance to agents when working with code in this repository.

## Project Overview
Speaker diarization viewer for WhisperX transcription output. Static web app (no build system/package.json).

## Commands
- **Run**: `npx serve -p 5000` then open http://localhost:5000
- **Generate waveform**: `node scripts/generate-waveform.js <audio> [output.json] [columns]` (requires ffmpeg)
- **Transcribe**: `python whisperX-nic.py <audio>` (requires GPU, HuggingFace token, model approvals)

## Non-Obvious Patterns
- **Config naming**: [`config.js`](config.js) uses `snake_case` (Pythonic style), JavaScript code uses `camelCase`
- **Speaker merges**: Diarization errors fixed via `speaker_merges` in config - all merged IDs transparently resolve via [`resolveSpeaker()`](app.js:52)
- **Path computation**: Audio/transcript paths derived from `meeting_name` at bottom of [`config.js`](config.js:88)
- **Waveform required**: Must be pre-generated (browser can't decode long opus files) - place in same dir as transcript JSON
- **No tests/linting**: Project has no test framework or linter configuration
- Edit [`config.js`](../../config.js) for settings, not [`app.js`](../../app.js) - paths, labels, merges go in config