Skills voice-agent-expert

📦

voice-agent-expert

Name: voice-agent-expert
Author: EGAdams

Content revision r1 High Risk 📁 Filesystem access🌐 Network access⚙️ External commands🔑 Env variables

Troubleshoot LiveKit Letta Voice Agents

Voice agent failures are hard to diagnose when audio, WebRTC, tokens, and process state interact. This skill gives focused checks, workflows, and helper scripts for LiveKit and Letta debugging.

Supports: Claude Codex Code(CC)

⚠️ 38 Poor

Stop for confirmation before installing.

Review the plan and obtain explicit user consent before changing files.

Install with my Agent

Copy this request to your Agent. It includes the canonical Skill page and manifest.

Agent request

Review the Skillstore skill "voice-agent-expert" from https://skillstore.io/skills/egadams-voice-agent-expert.md and its manifest at https://skillstore.io/api/skills/egadams-voice-agent-expert/manifest. Verify the artifact. Stop and obtain explicit user consent before installing or changing files.

Your Agent should still show its plan and request any confirmation required by the security policy.

Agent-readable resources

Use these links when an AI agent, crawler, or script needs clean context instead of reading the full page.

Markdown detail GET /skills/egadams-voice-agent-expert.md Signed manifest GET /api/skills/egadams-voice-agent-expert/manifest Signed lockfile GET /api/skills/egadams-voice-agent-expert/lockfile

Test it

Using "voice-agent-expert". My voice agent joins the room but never responds.

Expected outcome:

Probable cause: duplicate voice-agent workers or the wrong LiveKit hook.
Confirm that only one worker is active and that local testing uses DEV mode.
Check logs for the routing marker that proves Letta is receiving messages.

Using "voice-agent-expert". The playground says my token is invalid.

Expected outcome:

Generate a fresh LiveKit token because tokens expire after several hours.
Confirm the LiveKit URL matches the server endpoint used by the browser.
Treat the token as sensitive and avoid sharing logs that contain it.

Using "voice-agent-expert". Audio cuts out after a few seconds.

Expected outcome:

Check for duplicate voice-agent workers before changing model or audio settings.
Restart the worker cleanly after stopping stale processes and clearing stale rooms.
Verify that the active worker is the one receiving current logs.

Security Audit

High Risk

v8 • 7/6/2026 Open versioned report

The skill appears intended for local LiveKit and Letta debugging, not covert exfiltration. Several findings are confirmed because helper scripts read .env files, handle API secrets, spawn shell commands, and start local services with hardcoded paths. Many markdown backtick and localhost URL findings are false positives, but the package needs hardening before broad marketplace distribution.

Files scanned

685

Lines analyzed

Review items

False positives ignored

Confirmed security concerns (8)

High

scripts/diagnose_voice_system.py:68

Environment file access

livekit_env = Path("/home/adamsl/ottomator-agents/livekit-agent/.env")

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/diagnose_voice_system.py:69

Environment file access

planner_env = Path("/home/adamsl/planner/.env")

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/diagnose_voice_system.py:111

Environment file access

print("export $(grep -v '^#' /home/adamsl/ottomator-agents/livekit-agent/.env | xargs)")

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/generate_token.py:9

Environment file access

env_file = Path("/home/adamsl/ottomator-agents/livekit-agent/.env")

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/start_voice_system.py:61

Environment file access

env_file = Path("/home/adamsl/ottomator-agents/livekit-agent/.env")

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/start_voice_system.py:71

Environment file access

planner_env = Path("/home/adamsl/planner/.env")

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

SKILL.md:35-44 QUICK_REFERENCE.md:124-130

Broad Process Termination Instructions

The reset guidance tells users to kill LiveKit and voice-agent processes before restarting services. Broad process matching can disrupt unrelated local work if run without confirmation.

Both cited sections contain process-kill and restart guidance for local services. The commands are operationally plausible and not merely scanner artifacts.

Medium

scripts/generate_token.py:44-53

Connection Token Printed to Terminal

The token generator creates a LiveKit JWT and prints it directly to terminal output. This is useful for setup but can leak session access through copied logs.

The code directly creates a JWT and prints it under a token label. The behavior is intentional, but the token is still sensitive.

Capability review items (32)

These are real local capabilities that may be expected for this skill, so they require review but are not counted as confirmed malicious behavior.

High

QUICK_REFERENCE.md:133

Hidden file in home directory

## 📝 Aliases (Add to ~/.bashrc)

The skill tells users to add aliases to ~/.bashrc. Shell startup files are persistence points, even though the shown aliases appear benign.

High

QUICK_REFERENCE.md:133

Hidden file access

## 📝 Aliases (Add to ~/.bashrc)

The skill tells users to add aliases to ~/.bashrc. Shell startup files are persistence points, even though the shown aliases appear benign.

High

scripts/diagnose_voice_system.py:11

Python subprocess.run

result = subprocess.run(cmd, shell=True, capture_output=True, text=True)

The helper executes shell command strings with shell=True. Current call sites are mostly fixed, but this is still a command-execution surface.

High

scripts/diagnose_voice_system.py:111

Shell command substitution

print("export $(grep -v '^#' /home/adamsl/ottomator-agents/livekit-agent/.env | xargs)")

The printed export command loads every variable from a .env file through shell substitution. That can expose secrets to the shell environment.

High

scripts/diagnose_voice_system.py:68

Hidden file access

livekit_env = Path("/home/adamsl/ottomator-agents/livekit-agent/.env")

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/diagnose_voice_system.py:69

Hidden file access

planner_env = Path("/home/adamsl/planner/.env")

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/diagnose_voice_system.py:111

Hidden file access

print("export $(grep -v '^#' /home/adamsl/ottomator-agents/livekit-agent/.env | xargs)")

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/diagnose_voice_system.py:72

Generic API/secret keys

livekit_vars = ["LIVEKIT_URL", "LIVEKIT_API_KEY", "LIVEKIT_API_SECRET", "DEEPGRAM_API_KEY"]

The code handles named API keys or secrets for LiveKit, Deepgram, or OpenAI. These values are credentials and require explicit handling safeguards.

High

scripts/diagnose_voice_system.py:81

Generic API/secret keys

if check_env_var(planner_env, "OPENAI_API_KEY") or check_env_var(livekit_env, "OPENAI_API_KEY"):

The code handles named API keys or secrets for LiveKit, Deepgram, or OpenAI. These values are credentials and require explicit handling safeguards.

High

scripts/generate_token.py:9

Hidden file access

env_file = Path("/home/adamsl/ottomator-agents/livekit-agent/.env")

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/generate_token.py:16

Python environment access

os.environ.setdefault(key, value)

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/generate_token.py:27

Python environment access

api_key = os.environ.get('LIVEKIT_API_KEY')

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/generate_token.py:28

Python environment access

api_secret = os.environ.get('LIVEKIT_API_SECRET')

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/generate_token.py:27

Generic API/secret keys

api_key = os.environ.get('LIVEKIT_API_KEY')

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/generate_token.py:28

Generic API/secret keys

api_secret = os.environ.get('LIVEKIT_API_SECRET')

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/generate_token.py:31

Generic API/secret keys

if not api_key or not api_secret:

The code handles named API keys or secrets for LiveKit, Deepgram, or OpenAI. These values are credentials and require explicit handling safeguards.

High

scripts/generate_token.py:36

Generic API/secret keys

token = api.AccessToken(api_key, api_secret) \

The code handles named API keys or secrets for LiveKit, Deepgram, or OpenAI. These values are credentials and require explicit handling safeguards.

High

scripts/start_voice_system.py:14

Python subprocess.run

result = subprocess.run(

The startup helper checks ports through shell=True and shell pipelines. The current port values are fixed, but the command-execution pattern remains risky.

High

scripts/start_voice_system.py:95

Python os.exec variants

os.execv(

The script replaces itself with a hardcoded local Python executable and voice-agent file. That starts external code after loading environment secrets.

High

scripts/start_voice_system.py:61

Hidden file access

env_file = Path("/home/adamsl/ottomator-agents/livekit-agent/.env")

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/start_voice_system.py:71

Hidden file access

planner_env = Path("/home/adamsl/planner/.env")

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/start_voice_system.py:68

Python environment access

os.environ[key] = value

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

scripts/start_voice_system.py:78

Python environment access

os.environ.setdefault(key, value)

The code references local .env files that can contain API keys and service credentials. Reading or exporting those files is sensitive behavior.

High

SKILL.md:41

Hardcoded IP address

nohup ./livekit-server --dev --bind 0.0.0.0 > /tmp/livekit.log 2>&1 &

The command binds LiveKit to 0.0.0.0, which listens on all interfaces. That can expose a local dev service beyond localhost.

Medium

QUICK_REFERENCE.md:98

Temp directory access

rm -f /tmp/letta_voice_agent.{pid,lock}

The command deletes predictable files under /tmp that are used for PID and lock state. Removing those files can bypass duplicate-process protection.

Medium

scripts/diagnose_voice_system.py:89

Hidden file access

venv_python = "/home/adamsl/planner/.venv/bin/python3"

The script uses a hardcoded hidden virtualenv path to execute local Python code. This is not portable and can run unexpected local binaries.

Medium

scripts/start_voice_system.py:96

Hidden file access

"/home/adamsl/planner/.venv/bin/python3",

The script uses a hardcoded hidden virtualenv path to execute local Python code. This is not portable and can run unexpected local binaries.

Medium

scripts/start_voice_system.py:97

Hidden file access

["/home/adamsl/planner/.venv/bin/python3", LETTA_VOICE_AGENT_EXE, "start"]

The script uses a hardcoded hidden virtualenv path to execute local Python code. This is not portable and can run unexpected local binaries.

Medium

SKILL.md:41

nohup (background persistent process)

nohup ./livekit-server --dev --bind 0.0.0.0 > /tmp/livekit.log 2>&1 &

The reset workflow starts LiveKit as a persistent background process. This can leave a service running after the user expects debugging to end.

Medium

SKILL.md:44

Hidden file access

/home/adamsl/planner/.venv/bin/python3 letta_voice_agent.py dev > /tmp/letta_voice_agent.log 2>&1 &

The package references hidden local paths in commands or configuration. In a marketplace skill, those paths need explicit user review.

Medium

SKILL.md:41

Temp directory access

nohup ./livekit-server --dev --bind 0.0.0.0 > /tmp/livekit.log 2>&1 &

The workflow writes service logs to predictable files under /tmp. Shared temporary paths can create disclosure or tampering risk.

Medium

SKILL.md:44

Temp directory access

/home/adamsl/planner/.venv/bin/python3 letta_voice_agent.py dev > /tmp/letta_voice_agent.log 2>&1 &

The workflow writes service logs to predictable files under /tmp. Shared temporary paths can create disclosure or tampering risk.

Risk Factors

📁 Filesystem access (32)

🌐 Network access (6)

README.md:48 scripts/generate_token.py:57 scripts/start_voice_system.py:85 SKILL.md:53 SKILL.md:41 SKILL.md:74

⚙️ External commands (22)

🔑 Env variables (16)

Audited by: codex View Audit History →

Share & cite this report

Share the versioned assessment report, neutral badge, embed card, and citations. Skillstore reports evidence without deciding whether this Skill is safe.

Open versioned report

Copy report link

https://skillstore.io/skills/egadams-voice-agent-expert/audits/8?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Markdown badge

[![Skillstore security assessment](https://skillstore.io/badges/skills/egadams-voice-agent-expert/security.svg)](https://skillstore.io/skills/egadams-voice-agent-expert?utm_source=security_passport_badge)

HTML badge

<a href="https://skillstore.io/skills/egadams-voice-agent-expert?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/egadams-voice-agent-expert/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

Embed card

<iframe src="https://skillstore.io/embed/skills/egadams-voice-agent-expert.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>

Academic citations (APA · BibTeX · CFF)

APA citation

EGAdams. (2026). voice-agent-expert security audit report (audit version 8) [Author version unspecified]. Skillstore. https://skillstore.io/skills/egadams-voice-agent-expert/audits/8

BibTeX citation

@techreport{egadams-egadams-voice-agent-expert-2026,
  author = {EGAdams},
  title = {voice-agent-expert security audit report (audit version 8)},
  institution = {Skillstore},
  year = {2026},
  number = {8},
  url = {https://skillstore.io/skills/egadams-voice-agent-expert/audits/8},
  note = {Author version unspecified}
}

CITATION.cff

cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "voice-agent-expert security audit report (audit version 8)"
version: "unspecified"
type: report
authors:
  - name: "EGAdams"
date-released: "2026-07-06"
url: "https://skillstore.io/skills/egadams-voice-agent-expert/audits/8"
identifiers:
  - type: other
    value: "skillstore:egadams-voice-agent-expert:audit:8"
    description: "Skillstore immutable audit report identifier"

Skillstore Score

Why this score Evidence Confidence: Medium

Architecture

Maintainability

Content

Community

Spec Compliance

What You Can Build

Fix Local Voice Chat Setup

Diagnose why a local LiveKit and Letta voice session fails to connect or respond.

Debug Audio Cutoffs

Trace intermittent audio failures to duplicate workers, stale rooms, or wrong agent mode.

Prepare Support Runbooks

Turn repeated LiveKit and Letta failures into repeatable checks and escalation steps.

Try These Prompts

Diagnose a Connection Failure

My LiveKit and Letta voice agent does not connect. Ask for the error, then guide me through the safest checks first.

Choose the Correct Agent Mode

Explain whether I should use START mode or DEV mode for local voice chat, and list the symptoms of choosing the wrong mode.

Investigate Duplicate Workers

Help me determine whether duplicate voice-agent workers or stale LiveKit rooms are causing audio cutoffs.

Design a Safer Startup Workflow

Review my local LiveKit and Letta startup workflow for security risks, secret handling, and safer process management.

Best Practices

Review every command before running it against local services or shell startup files.
Move hardcoded paths into configuration before sharing this workflow across machines.
Treat generated LiveKit tokens and .env contents as sensitive credentials.

Avoid

Do not run broad process-kill commands without confirming the matching processes.
Do not bind local development services to all interfaces without network controls.
Do not paste generated tokens or .env-derived output into public issue reports.