wolfbud is a Claude Code workflow mod by leopiney.
WolfBud, a voice coworker you call beside the session, hears what Claude does and sends a prompt you agree on. Your voice and that activity go to ElevenLabs.
It runs inside the Claude Code process. The side panel is only what the author's README names. Read the repo before you install, or run claude plugin validate on the plugin folder.
Version
1.2.1
Repo stars
7
Last change
2026-10-08
What it can reach
Draws in the session
Reaches the network or an outside service
Sees tool calls
Calls another program on this machine
WolfBud is a voice coworker that sits beside the session. You need Claude Code 2.1.287 or later, Node 20 or later, Google Chrome, and an ElevenLabs API key. Set the key by exporting ELEVENLABS_API_KEY before you start Claude Code, or save it with /plugin configure wolfbud@elevenlabs-mods. Tested on 2.1.292 on macOS. /wolfbud subscribes this session, opens the pane, and shows the one window. On a Mac with WolfBud.app installed, that app is the face. Everywhere else the face is a Chrome window. The hub listens on 127.0.0.1 port 4747. The first /wolfbud that finds it down starts it with node. The call is an ElevenLabs conversational agent. The key stays in the hub and never reaches the window. The agent hears a snapshot of the session, tool calls, failures, final answers, and permission prompts. When you agree, the voice agent starts a turn, adds a note to the running turn, or queues the prompt. The prompt Claude receives is framed as coming from the wolfbud plugin, and the pane lists each one. It can stop Claude only when you ask. The pane buttons start or end the call, raise the window, or hide the pane. The README says not to expose port 4747 on a public URL. The agent uses deepseek-v41-flash and eleven_v4_turbo.
The marketplace name is elevenlabs-mods. A clone of the repo is a marketplace of the same name, and only one can be added. The prompt it sends is why this page files it under workflow. The wolf is the face of the call. vox draws a waveform above the prompt while vox speaks, from the spectrum vox writes for the utterance, and a short one while it listens. sportscaster speaks play-by-play of the session, with crowd noise, and opens a booth pane. This one is a call you talk on, and the agreed prompt is sent into the session. The call goes to ElevenLabs. The hub on 127.0.0.1 is on this machine. Agreeing on the call is the voice agent's tool, and the pane buttons do not send the prompt, so this page does not list runs-commands. Stopping Claude when you ask is not a refused tool call.
Demo
From the author's repo.
What it hooks into
session.startWhen a session starts
command.runWhen you run its slash command/wolfbud
prompt.submitWhen you send a prompt
turn.startWhen Claude starts a turn
tool.callAround a tool call, where it can hold one
turn.completeWhen a turn ends
classic.NotificationA Claude Code event
session.endWhen a session ends
ui.closeWhen its pane closes
ui.renderDraws into the sessionPane
Read from the on() calls in its hook modules. Open its source under Files to check what each one does.
Slash commands 1
/wolfbudWolfBud: a voice coworker that watches this session and can message Claude
Files
The author's README and source, 4 files. They load when you open them.