tts

TTS (one-off)

/tts reads one thing aloud through the aiTTS Mac app and exits. The app does all synthesis and playback (local Kokoro voices by default; engine and voice are chosen in the app's Settings). This skill only forwards a request to the app's local socket. Continuous narration of your sessions is owned by the app itself; no daemon, no hooks.

If the app is not installed, the script prints a pointer to https://aitts.dev and exits. There is no fallback synthesis path; do not try to install one.

Arguments: $ARGUMENTS (optional). With no argument, read the last assistant reply of the current session.

Phase 0: resolve the shim

if [ -n "${CLAUDE_PLUGIN_ROOT:-}" ] && [ -f "${CLAUDE_PLUGIN_ROOT}/tts.js" ]; then
  SKILL_DIR="${CLAUDE_PLUGIN_ROOT}"
elif [ -d "${HOME}/.claude/skills/tts" ] && [ -f "${HOME}/.claude/skills/tts/tts.js" ]; then
  SKILL_DIR="${HOME}/.claude/skills/tts"
else
  echo "ERROR: cannot locate tts.js. Install the plugin: claude plugin marketplace add wickdninja/aitts-plugin && claude plugin install tts@aitts" >&2
  exit 1
fi

Retired flags: --on, --off, --status, --mute, --skip, --pause, --resume, --gemini, --voices, --voice, --speed. If $ARGUMENTS contains any of these, tell the user: continuous narration and voice choice live in the aiTTS menu-bar app now (Settings -> Voice / Agents), then read any remaining content normally without the flag.

Phase 1: route by input type

Decide what $ARGUMENTS is and send it. You (Claude) do the gathering; the shim only forwards.

• No argument -> take your own last assistant reply in this session, strip code fences and tool noise so it reads well aloud, and pipe it:

  printf '%s' "<the cleaned reply text>" | node "$SKILL_DIR/tts.js"
  

• Plain text (quoted words, a sentence, pasted prose) -> pass it through:

  node "$SKILL_DIR/tts.js" "<the text>"
  

• Local file (PDF, image, audio, or any document; absolute or relative path that exists) -> let the APP extract it natively (PDFKit / Vision OCR beat anything you could do here):

  node "$SKILL_DIR/tts.js" --file "<absolute path>"
  

  For a plain-text/markdown/source file you may instead read it yourself, trim boilerplate that reads badly aloud (badges, tables, license headers), and pipe the cleaned text. Prefer that when the user asked for a summary or "explain", not a verbatim read.

• URL -> fetch the page yourself (WebFetch), extract the readable article text, and pipe it as text. Do NOT pass URLs to the shim; it has no fetcher on purpose.

• Multiple items -> combine flags freely. --file and --clipboard items play in the order given; all bare text is joined into ONE item that plays last, e.g. node "$SKILL_DIR/tts.js" --file "/a.pdf" "and then this remark". If something must play before a file, send it as a separate invocation.

Phase 2: report

The shim prints sent to aiTTS on success. Tell the user it is playing in aiTTS (the floating caption window and menu bar controls handle pause/skip/replay). If the shim printed the install pointer, relay it: the aiTTS Mac app from https://aitts.dev is required, with a 14-day free trial.