Skip to main content
AI/MLWaterplanAI

ac-tools-voice-user

Use the say tool for short one-way spoken alerts to the user. Use when you finish a meaningful task, need attention, confirmation, approval, or input, or when the user explicitly asks for voice output. Keep spoken text brief and put the full details in writing.

Stars
31
Source
WaterplanAI/agentic-config
Updated
2026-05-25
Slug
WaterplanAI--agentic-config--ac-tools-voice-user
View on GitHubRaw SKILL.md

// install — copy + paste into any project

mkdir -p .claude/skills && curl -fsSL https://raw.githubusercontent.com/WaterplanAI/agentic-config/HEAD/packages/pi-ac-tools/skills/ac-tools-voice-user/SKILL.md -o .claude/skills/ac-tools-voice-user.md

Drops the SKILL.md into .claude/skills/ac-tools-voice-user.md. Works with Claude Code, Cursor, and any agent that loads SKILL.md files from .claude/skills/.

Voice User

Use the say tool for concise one-way spoken communication with the user.

When to use

  • You finished a meaningful task and want to alert the user.
  • You need the user's attention, confirmation, approval, choice, or missing input.
  • The user explicitly asks for spoken output.
  • Only in the top-level agent runtime. Subagents cannot use say.

How to use

  • Call the say tool with a short spoken message.
  • Prefer under 50 words when asking for attention or confirmation.
  • Use this shape when possible:
    • <project> - <what you need>
  • After the spoken alert, give the full explanation in the normal written response.

Keep spoken output short

Do not speak:

  • long explanations
  • code
  • secrets, tokens, or credentials
  • stack traces
  • large diffs or long file paths

Tool shape

{
  "text": "play - say extension updated; please reload.",
  "voice": "Samantha",
  "rate": 190
}

Only text is required. If voice or rate is omitted, the extension may use the saved defaults.

Notes

  • The extension may configure automatic voice as off, always, or long.
  • Even when automatic voice is off, use the say tool when the user explicitly wants spoken output.
  • Subagents do not have access to say. If a subagent finishes work or needs attention, it should report that in writing and let the parent agent decide whether to speak.
  • Long explanations stay in writing; voice is only the concise alert.