How do I install the Skill Speaker skill?

Run npx skillmds@latest add eugene-e0a80fd8080ff8e/skill-speaker in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Skill Speaker safe to use?

This skill has not completed SkillMD's automated safety review yet. Capability flags: executes scripts, makes network calls. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Skill Speaker?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Skill Speaker free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Skill Speaker?

Eugene-E0a80fd8080ff8e (@eugene-e0a80fd8080ff8e) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Skill Speaker by eugene-e0a80fd8080ff8e

Name: Skill Speaker
Rating: 4.2 (5 reviews)
Author: Eugene-E0a80fd8080ff8e

Skill Speaker

Run a local microphone↔speaker voice dialog using OpenAI Realtime audio models (Node.js). Uses the `mic` and `speaker` npm packages. Pass a detailed prompt that includes all dialog context + what must be learned from the user.

File contents

Local Voice Dialog (mic ↔ OpenAI Realtime ↔ speaker)

This skill runs a two-way voice conversation:

Your microphone streams audio to an OpenAI Realtime audio model.
The model streams audio back to your speakers.
Optional live transcripts are printed to stdout (useful for Codex to capture what was learned).

Prompt requirement (important)

During the call, the Realtime model does not have access to most of the context that Codex has (repo files, prior reasoning, etc). Put everything the voice agent needs into the startup prompt, including:

Relevant background/context for the conversation
What the agent must learn from the user (explicit questions or a checklist)
Any constraints (tone, length, allowed topics, how to summarize learned info)

One-time setup

Install the skill into Codex:
- Copy skill-speaker/ into $CODEX_HOME/skills (default ~/.codex/skills/)
Install Node deps:
- cd scripts && npm install
Hardcode your OpenAI key:
- Edit scripts/config.js and set OPENAI_API_KEY
- Avoid committing the key to a public repo.

Run

Prompt from an argument:
- node scripts/voice-dialog.js --prompt "You are a voice interviewer. Context: ... Learn: ..."
Prompt from a file (best for long prompts):
- node scripts/voice-dialog.js --prompt-file /path/to/prompt.txt
Prompt from stdin:
- cat /path/to/prompt.txt | node scripts/voice-dialog.js --prompt-stdin
Test microphone level (no OpenAI connection):
- node scripts/voice-dialog.js --mictest

While running:

Press q or space to end the conversation and print RESULT_JSON_*.
Saying “goodbye” (or similar) should cause the model to call end_conversation and end automatically.

Notes / troubleshooting

Requires network access (OpenAI Realtime).
In network-restricted Codex runs, request approval to allow the connection.
Uses the modern Realtime WebSocket endpoint (wss://api.openai.com/v1/realtime) via openai/realtime/ws (not the deprecated beta module path).
Use headphones to avoid speaker→mic feedback loops.
mic typically shells out to arecord (Linux) or sox; install those if startup fails.
If you see silence in --mictest, set the capture device (ALSA) with --mic-device (default is default), e.g. --mic-device plughw:0,0. Use arecord -L to list devices.
speaker is a native module; you may need build tooling for npm install.
If the model name changes or you lack access, edit scripts/config.js (REALTIME_MODEL).

---name: skill-speakerdescription: Run a local microphone↔speaker voice dialog using OpenAI Realtime audio models (Node.js). Uses the `mic` and `speaker` npm packages. Pass a detailed prompt that includes all dialog context + what must be learned from the user.---# Local Voice Dialog (mic ↔ OpenAI Realtime ↔ speaker)This skill runs a two-way voice conversation:- Your microphone streams audio to an OpenAI Realtime audio model.- The model streams audio back to your speakers.- Optional live transcripts are printed to stdout (useful for Codex to capture what was learned).## Prompt requirement (important)During the call, the Realtime model does **not** have access to most of the context that Codex has (repo files, prior reasoning, etc). Put *everything the voice agent needs* into the startup prompt, including:- Relevant background/context for the conversation- What the agent must learn from the user (explicit questions or a checklist)- Any constraints (tone, length, allowed topics, how to summarize learned info)## One-time setup1. Install the skill into Codex: - Copy `skill-speaker/` into `$CODEX_HOME/skills` (default `~/.codex/skills/`)2. Install Node deps: - `cd scripts && npm install`3. Hardcode your OpenAI key: - Edit `scripts/config.js` and set `OPENAI_API_KEY` - Avoid committing the key to a public repo.## Run- Prompt from an argument: - `node scripts/voice-dialog.js --prompt "You are a voice interviewer. Context: ... Learn: ..."`- Prompt from a file (best for long prompts): - `node scripts/voice-dialog.js --prompt-file /path/to/prompt.txt`- Prompt from stdin: - `cat /path/to/prompt.txt | node scripts/voice-dialog.js --prompt-stdin`- Test microphone level (no OpenAI connection): - `node scripts/voice-dialog.js --mictest` While running:- Press `q` or `space` to end the conversation and print `RESULT_JSON_*`.- Saying “goodbye” (or similar) should cause the model to call `end_conversation` and end automatically.## Notes / troubleshooting- Requires network access (OpenAI Realtime).- In network-restricted Codex runs, request approval to allow the connection.- Uses the modern Realtime WebSocket endpoint (`wss://api.openai.com/v1/realtime`) via `openai/realtime/ws` (not the deprecated beta module path).- Use headphones to avoid speaker→mic feedback loops.- `mic` typically shells out to `arecord` (Linux) or `sox`; install those if startup fails.- If you see silence in `--mictest`, set the capture device (ALSA) with `--mic-device` (default is `default`), e.g. `--mic-device plughw:0,0`. Use `arecord -L` to list devices.- `speaker` is a native module; you may need build tooling for `npm install`.- If the model name changes or you lack access, edit `scripts/config.js` (`REALTIME_MODEL`).

Frequently asked questions