Voice and Skills: Hands-Free Workflows

+15 Mana ✨
Hermes Skills(DOCUMENTATION)

Introduction

Voice mode and the Hermes Skills system are powerful separately. Together they enable something neither does alone: invoking a whole packaged workflow without touching the keyboard. This lesson is about how the two features compose.

Key Concepts

  • Skill: A packaged, reusable workflow in Hermes, invoked by name (the subject of the Hermes Skills course).
  • Skill invocation: Triggering a skill, conventionally written as /skill-name when typed.
  • Voice invocation: Triggering a skill by speaking, where STT turns your spoken request into the same instruction typing would produce.
  • Composability: The property that two features combine cleanly because neither needs to know about the other.

Real World Context

A developer has a skill that runs their daily review routine: check the build, summarize open issues, list what changed. Normally they type /daily-review. One morning, hands full with coffee, they simply say "run my daily review." STT transcribes it, the agent recognizes the intent, the skill runs, and TTS reads the summary back. A multi-step routine executed without a keystroke. That is voice and skills composing.

Deep Dive

The reason voice and skills compose so cleanly goes back to the pipeline. STT converts speech into ordinary text. The agent then does exactly what it always does with text, including recognizing that a request maps to a skill. Skills require no voice-specific support because, by the time a skill is involved, the input is already text. Voice did its job at the front door and got out of the way.

This composability is the heart of hands-free workflows. A skill packages many steps into one invocation. Voice removes the keyboard from that invocation. Stack them and a single spoken sentence triggers an entire multi-step routine, and TTS can read the result back. The keyboard leaves the loop entirely.

There is a subtlety in how you invoke a skill by voice. When typing, you write a precise token like /daily-review. Slashes and exact slugs do not dictate well. So voice invocation leans on natural phrasing: "run my daily review," "start the daily review skill." You are describing the intent and letting the agent map it to the skill, rather than dictating the exact /slug. This is more forgiving of how speech actually works, and it is why well-named skills (names that match how you would naturally say them) are easier to use by voice.

That insight feeds back into skill design. If you expect to invoke a skill by voice, give it a name you can say comfortably and that STT will transcribe reliably: ordinary words, no cryptic abbreviations, no characters that do not survive dictation. A skill named daily-review is voice-friendly; a skill named dr-v2-x is not.

The larger principle: voice and skills are orthogonal features that multiply. Voice handles input modality. Skills handle workflow packaging. Because each does one job and hands clean text to the next stage, they combine without friction into genuinely hands-free automation.

Common Pitfalls

  1. Trying to dictate the exact /slug: Slashes and cryptic slugs transcribe poorly. Invoke skills by natural phrasing and let the agent map intent to the skill.
  2. Naming skills without voice in mind: A skill you plan to trigger by voice needs a name made of ordinary, easily transcribed words.

Best Practices

  1. Invoke skills by intent when speaking: Say what you want ("run the daily review") rather than dictating punctuation-heavy syntax.
  2. Name voice-invoked skills to be speakable: Choose plain-word names that STT transcribes reliably and that match how you would naturally ask.

Summary

  • Voice and skills compose because STT hands the agent ordinary text, exactly what skill invocation already expects.
  • Stacking them lets one spoken sentence trigger a whole multi-step workflow, fully hands-free.
  • Invoke skills by natural phrasing rather than dictating the exact /slug.
  • Name skills you plan to use by voice with plain, easily transcribed words.

Code Examples

bash
# Typed invocation: a precise slug
/daily-review

# Voice invocation: natural phrasing, STT transcribes it,
# the agent maps the intent to the same skill
# (spoken) "run my daily review"
#
# This is why a voice-invoked skill should be named in plain,
# easily-spoken words rather than a cryptic slug like dr-v2-x.
✓ Completed