Agent skill · Audited from source

voicemode

skill · mbailey/voicemodeMIT licenceAudited 14 Oct 2026 · source a40d930a5d03

★ 1,399pushed 8 days ago
!CAUTIONGrade B · 89/100 · 5 findings · a40d930
1,399repo stars · whole mbailey/voicemode

Is it safe? With care — CAUTION, grade B (89/100). We read this skill's source at commit a40d930a5d03 and found no filesystem, network or shell access and no hard-coded secrets. The audit raised 5 findings, 0 critical or high. What it does: Talk with the user by voice through the converse tool: speak, listen, run a short survey, pick or clone a voice, share the channel with other agents, recover when voice drops. Use when the user mentions voice, speaking,…

Ask an AI about this pageGChatGPTCClaudePPerplexityGeGeminiXGrok
More
Cite

Cite this page

ModelsAtlas. “voicemode (mbailey/voicemode)”. Retrieved 14 Oct 2026. https://modelsatlas.com/skills/mbailey/voicemode

Trust audit

Audit 0.4.1 · source read 14 Oct 2026

Trust score89/100!CAUTION B
0 critical0 high2 medium3 low0 info

CAUTION means the audit found things worth knowing before you trust this skill. Grade B is shared by 13,985 of 14,816 audited skills.

What the source does

  • FilesystemNone observedNo reads or writes in the source
  • NetworkNone observedNo outbound calls in the source
  • ShellNone observedNo commands run in the source
  • DependenciesPinnedNothing floats
  • SecretsNone foundNo keys in source

Five audit layers

LayerWhat we checkResultFindings
L0 ProvenanceRepo owner, licence, archive status, activityWARN3
L1 Static sourceSecrets, path traversal, supply chain, hidden codeNA1
L2 InstructionsPrompt injection, authority framing, hidden directivesPASS1
L3 InventoryThe files and declarations match the sourcePASS0
L4 SandboxRun in isolation and watch behaviourSKIPPED why0

Gates applied: no behavioural pass.

Findings (5)

LOWNetwork egressnet.raw_ip · references/impressions.md:30
F-04 Network egressCWE-200 · CWE-319Layer L1
| `VOICEMODE_MLX_AUDIO_BASE_URL` | `http://127.0.0.1:8890/v1` | the mlx-audio endpoint |
Why
No further detail recorded for this rule.
Fix
—
LOWPrompt injectionprompt.transfer_instruction · references/controls.md:9
F-01 Prompt injectionCWE-94 · CWE-1427Layer L2
`voicemode control pause | resume | stop | skip-forward | skip-back` drives a live converse turn: pause it, barge in (cut the utterance and take the mic), finish answering early, hear the question aga
Why
an instruction to move sensitive data to an outside destination
Fix
remove; a skill never needs the user's secrets off the machine

Audit 0.4.1 · source sha a40d930a5d03 · 14 Oct 2026Report an issue or request a re-scan

Provenance

Who published it, and what the verdict is pinned to

  • Publishermbailey GitHub userOwner of the repository this skill is read from
  • Source foldermbailey/voicemode/.claude/skills/voicemodeThe exact folder we audited
  • LicenceMITAs the repository states it
  • Commit pinned for this verdicta40d930a5d03A new audit runs when the source changes
  • Repository statusNot archivedLast push 8 days ago
  • Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

What it tells the agent

Full source on GitHub ↗

Its description: Talk with the user by voice through the converse tool: speak, listen, run a short survey, pick or clone a voice, share the channel with other agents, recover when voice drops. Use when the user mentions voice, speaking,…

---
name: voicemode
description: Talk with the user by voice through the converse tool: speak, listen, run a short survey, pick or clone a voice, share the channel with other agents, recover when voice drops. Use when the user mentions voice, speaking, talking, listening, a voice name, an impression, or background music.
---

# VoiceMode

You speak with the `converse` tool. Its description and schema are the contract; this page is how to use it well. No `converse` in your tool list? See [recovery].

## How a conversation goes

- `converse("...")` speaks, then listens: the user's reply comes back as the result, and they never have to address you. The turn stays with you for as long as you keep listening: [listening].
- Narrate with `wait_for_response=false` and fire the work in the same response (Bash, Agent, Read...), so speech and action overlap. Then listen with the next call: a speak-only call leaves nobody listening.
- Short utterances, one question per turn. More than one thing to ask? `turns` runs them as a survey: [surveys].
- Leave `listen_duration_max` unset. The user owns that ceiling, and silence detection ends the turn for you.
- Leave `voice` unset unless asked; a name is lowercase with its prefix (`bm_daniel`, `af_sky`). Personas, cloning, impressions: [voices].
- Other agents on the channel: `hold_conch=true` when you will continue the thread; `wait_for_conch` to queue: [conch].
- "Hang on" or "wait" at the end of a reply pauses and re-listens on its own. For minutes, run `sleep N` as a background Bash call; converse again when it finishes.

## When something is off

- Tool missing, `-32000 Connection closed`, permission prompts, an empty reply after they spoke, misheard names: [recovery].
- Pause, barge in, replay or skip during a turn; media keys and Stream Deck: [controls].
- Background music: [music].

## References

Every bracketed name above is a file in `references/`; this list is here so the links render.

[conch]: references/conch.md
[controls]: references/controls.md
[listening]: references/listening.md
[music]: references/music.md
[recovery]: references/recovery.md
[surveys]: references/surveys.md
[voices]: references/voices.md

Excerpt, first 2,191 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

Install

Pinned to the audited commit a40d930

Download folder ↗
git clone https://github.com/mbailey/voicemode
git -C voicemode checkout a40d930a5d03

The author documents no install command, so this uses the repository directly. Our verdict covers commit a40d930a5d03 only.

Works in which agents?

What its own text says, not a test

AgentStatusMore
Claude CodeMentioned in its text, not testedSkills for Claude Code · About Claude Code
“Mentioned” means the skill’s text names the agent. We have not run it there.

Audit history

Every audit, every source change

  1. !CAUTION B 89 / 100

    first audit · source a40d930a5d03

README badge

For maintainers · shows the latest audit

ModelsAtlas: CAUTION BMODELSATLASCAUTION · B
[![ModelsAtlas: CAUTION B](https://modelsatlas.com/skills/mbailey/voicemode/badge.svg)](https://modelsatlas.com/skills/mbailey/voicemode/)

The badge shows our verdict and grade for this skill, and links back to this page. It changes when the audit does.

Similar skills

find-docs✓SAFE

upstash/context7

Retrieves up-to-date documentation, API references, and code examples for any developer technology. Use this skill whenever the user asks about a specific library, framework, SDK, CLI tool, or cloud service — even for…

★ 64kGrade B
context7-mcp✓SAFE

upstash/context7

This skill should be used when the user asks about libraries, frameworks, API references, or needs code examples. Activates for setup questions, code generation involving libraries, or mentions of specific frameworks…

★ 63kGrade B
context7-cli✓SAFE

upstash/context7

Use the ctx7 CLI to fetch library documentation and configure Context7 MCP. Activate when the user mentions "ctx7" or "context7", needs current docs for any library, or needs to set up Context7 for their AI coding agent.

★ 63kGrade B
troubleshooting✓SAFE

chromedevtools/chrome-devtools-mcp

Uses Chrome DevTools MCP and documentation to troubleshoot connection and target issues. Trigger this skill when list_pages, new_page, or navigate_page fail, or when the server initialization fails.

★ 54kGrade B
memory-leak-debugging✓SAFE

chromedevtools/chrome-devtools-mcp

Diagnoses and resolves memory leaks in JavaScript/Node.js applications. Use when a user reports high memory usage, OOM errors, or wants to capture, compare, or inspect heap snapshots with Chrome DevTools MCP memory…

★ 54kGrade B
debug-optimize-lcp✓SAFE

chromedevtools/chrome-devtools-mcp

Guides debugging and optimizing Largest Contentful Paint (LCP) using Chrome DevTools MCP tools. Use this skill whenever the user asks about LCP performance, slow page loads, Core Web Vitals optimization, or wants to…

★ 54kGrade B

Add this badge to your README

[![ModelsAtlas: CAUTION B](https://modelsatlas.com/skills/mbailey/voicemode/badge.svg)](https://modelsatlas.com/skills/mbailey/voicemode/)

Questions about voicemode

What does the voicemode skill do?
Natural voice conversations with Claude Code
Is voicemode safe to install?
With care. The audit graded it B (89/100) and found 5 things worth knowing before you trust this skill, listed below with the exact line each was found on.
What can voicemode access on my machine?
The audit observed no filesystem, network or shell use at all in its source.
Which assistants does voicemode work with?
Its documentation mentions claude-code. That is what the text claims, not a compatibility test we ran.
How current is this page?
The grade is for one exact copy of the source (a40d930a5d03), read on 2026-10-14. The repository is watched, and a new audit runs when it changes — this is the first audit.

Sources

Spotted something wrong? Report a correction · Methodology

Keep going in your AI

Opens with this page as the source. We add nothing else to the prompt.