im done mf

Voice layer for Claude Code. Speaks when your shit's done.

npm install -g imdone-mf
imdone

What it does

You run imdone instead of claude. It wraps Claude Code and listens for lifecycle events via Claude Code's native HTTP hooks. Today it reacts to Stop and Notification hooks: it speaks, then (on Stop) listens and injects what you said into Claude Code. No paste required.

macOS only. Requires Claude Code.

Install

npm install -g imdone-mf

That's it. No compilation. Pre-built binaries for arm64 and x86_64 are downloaded automatically.

Requirements

• macOS 12+
• Node.js 18+
• Claude Code installed (claude on your PATH)

Usage

imdone                        # starts Claude Code with voice layer
imdone "build me a todo app"  # pass a task directly
imdone --diagnose             # check all system dependencies

On first run, imdone auto-writes .claude/settings.json in your project directory with the HTTP hook config. You never touch that file manually.

"Listening..." appears immediately when TTS finishes. Speech is committed after 2 seconds of silence — mid-sentence pauses are fine. Max recording: 60 seconds.

How it works

You run: imdone [task]

imdone starts:
  1. HTTP server on :51234 (receives Claude Code hook events)
  2. Claude Code via PTY (your terminal works exactly as normal)

When Claude Code fires a Stop or Notification hook:
  → imdone speaks the phrase out loud (three-tier TTS, see below)
On Stop only (after TTS):
  → mic opens (SFSpeechRecognizer, on-device)
  → prints "I heard: [transcript]" and injects it into Claude Code (`\r` = Enter)
Notifications do not run the listen/inject step.

TTS priority order:

1. Local audio file — any .mp3/.wav/.aiff/.m4a in ~/.imdone/audio/stop/ or ~/.imdone/audio/notification/
2. ElevenLabs — if ELEVENLABS_API_KEY is set in your environment
3. macOS say — always-available fallback, zero config

Customize phrases

Edit ~/.imdone/phrases.json (created on first run):

{
  "Stop": [
    "yo your shit's done motherfucker",
    "aye, task complete, what's next"
  ],
  "Notification": [
    "aye, claude needs you"
  ]
}

To use a specific say voice, add a "voice" field with any say-compatible name. Run say -v ? to list installed voices.

Custom audio files

Drop your own clips into ~/.imdone/audio/ and they take priority over everything else:

~/.imdone/audio/
  stop/          ← played on Claude Code Stop events
    clip1.mp3
    clip2.mp3
  notification/  ← played on Notification events
    clip1.mp3

Supported formats: .mp3, .wav, .aiff, .m4a. A random clip is picked each time. If the folder is empty or missing, imdone falls through to ElevenLabs or say.

Troubleshoot

imdone --diagnose

Checks: claude on PATH, say on PATH, afplay on PATH, local audio files (optional), ElevenLabs API key (optional), .claude/settings.json hook URL, phrases.json valid, imdone-listen binary present, port 51234 available.

Port conflict:

IMDONE_PORT=51235 imdone

Microphone permission denied: System Settings → Privacy & Security → Microphone → enable Terminal (or your terminal app).

License

MIT