| name | call-me |
| description | Voice-call the user out loud through their speakers when you need their attention. Invoke with `/call-me <message>` or `/call-me` (no args - then ping with a generic "your turn"). Also triggers when the user says "call me", "ping me out loud", "voice ping", "alert me on speakers", "say it out loud", or any equivalent - or proactively when you finish a long-running task, hit a question only the user can answer, fail a build/deploy, or have been blocked >30s waiting on the user. Plays a short musical chime then a TTS sentence saying their name, which Windows desktop the calling window is on, and a one-line message. |
call-me
You have access to a CLI installed on this machine: call-me-skill.
When this file is installed as ~/.claude/skills/call-me/SKILL.md, Claude Code exposes it as the /call-me slash command.
Slash-command usage
When the user types /call-me <message>:
- Use
<message> verbatim as the spoken text. Do NOT rephrase, expand, or add a name prefix - the CLI already prepends the user's name and desktop number.
- Run:
call-me-skill speak "<message>" (quote properly, escape inner quotes).
When the user types /call-me with no args:
- Ping with a short generic line based on session state. Examples: "your turn", "ready when you are", "blocked, need you". 5-8 words max.
- Run:
call-me-skill speak "<your generic line>".
Desktop-number accuracy (v0.4.0+)
As of v0.4.0 the CLI auto-detects the caller's desktop, not the foreground desktop. When invoked from a VS Code / Cursor / Antigravity integrated terminal, it reads VSCODE_PID / CURSOR_PID / ANTIGRAVITY_PID from env, finds that process's main-window HWND, and reports the desktop where THAT window actually lives. So if you (the agent) live on desktop 2 and the human is currently looking at desktop 5, the message correctly says "calling from desktop 2" - the human's eyes can find you.
You don't need to do anything to enable this - it just works when the env var is set (which is always, when invoked from Claude Code's Bash tool). To force a specific PID, set CALL_ME_PID=<pid> in the env before calling.
If no PID env var is set (rare - interactive shell outside an IDE terminal), the CLI falls back to the legacy foreground-desktop lookup.
How to invoke (programmatic / proactive)
Run via shell:
call-me-skill speak "<one short sentence>"
The CLI will:
- Play the user's chosen intro chime (~2-3s)
- Speak a sentence in this shape:
"{name}, {address}, calling from desktop {N}, {your message}."
You only supply the {your message} part. Everything else (name, address, desktop number, voice, tone) is configured by the user during setup and should not be overridden.
When to call
Good triggers:
- User explicitly asks: "call me", "use the call me skill", "ping me", "alert me", "say it out loud"
- You finished a long-running task and have results to show
- You need a decision or input you can't make on your own
- A build/test/deploy failed in a way that needs the user to look
- You've been blocked for more than ~30 seconds waiting on something the user controls
Bad triggers:
- Routine progress updates ("step 3 of 7 done")
- Things you can verify and fix yourself
- Anything happening in the foreground window the user is already watching
Pair voice ping with window-foreground (recommended)
When you ping the user to DO something (type a password, click a 2FA, accept an OAuth consent), bring the relevant window to the foreground at the same time. A voice ping that says "go to the consent dialog" is useless if the user has to alt-tab through 20 windows to find it. Foreground the target window first, then voice-ping referencing what's now visible.
Message style
Keep it short. 5-8 words is the sweet spot. The CLI prepends the addressing
boilerplate, so don't repeat the user's name yourself.
Good:
call-me-skill speak "render finished, 2 minutes saved"call-me-skill speak "build broke on line 47"call-me-skill speak "need your call on the migration"
Bad:
call-me-skill speak "Hi {name}, I just wanted to let you know that..." (too long, repeats name)call-me-skill speak "done" (too vague)call-me-skill speak "Step 3 of 7 complete, moving to step 4" (routine progress)
Options
--length short|medium|long Override the configured sentence length for this call
--no-intro Skip the chime (use sparingly - the chime is the cue)
--focus <pattern|alias> Window the daemon hotkey will foreground when the user
presses Win+Ctrl+J (aliases: vscode, cursor, chrome,
edge, firefox, terminal; or any window-title substring)
Subcommands (for reference)
call-me-skill setup - first-run wizard (8 questions)call-me-skill speak "<msg>" - main verb (above)call-me-skill jump <N> - switch Windows to virtual desktop N (1-based)call-me-skill daemon start|stop|status|install|uninstall - hotkey daemon (Win+Ctrl+J jumps to last call's desktop+window)call-me-skill mode on|off - voice-mode toggle for Stop hook (v0.3)
If the CLI is missing
If call-me-skill is not installed, fall back to a normal text response and
mention to the user: "Heads up - call-me-skill isn't installed. Run
npm install -g @kivimedia/call-me-skill && call-me-skill setup if you want
voice alerts."
Do not try to install it yourself.
