Back to skill

Security audit

Jarvis Vocal

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed local voice-synthesis helper with optional Android audio delivery, with setup and privacy cautions but no artifact-backed malicious behavior.

Install only if you are comfortable installing external TTS tooling and downloading model files without pinned checksums. Avoid synthesizing sensitive text when using Android push or streaming, because the audio can be transferred to the connected device. Review or create the missing jarvis-tts/jarvis-speak wrapper scripts before relying on device playback.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:35
Finding

Unpinned Third-Party Dependencies and Mutable Model Artifacts

Content
View full analysis

Vulnerability Details

File Locations:

  • SKILL.md:35-48
  • README.md:18-31
  • package.json:22-26

Vulnerability Type: Supply-chain exposure caused by unpinned software dependencies and model artifacts
Risk Level: Medium

Vulnerable Code

SKILL.md:35-48:

bash
pipx install piper-tts
sudo apt install ffmpeg  # or equivalent
bash
# Create voice directory
mkdir -p ~/.local/share/piper/voices/en_GB

# Download models via HuggingFace CLI
cd ~/.local/share/piper/voices/en_GB
hf download jgkawell/jarvis en/en_GB/jarvis/high/jarvis-high.onnx --local-dir .
hf download jgkawell/jarvis en/en_GB/jarvis/high/jarvis-high.onnx.json --local-dir .
# Optional: medium quality model
hf download jgkawell/jarvis en/en_GB/jarvis/medium/jarvis-medium.onnx --local-dir .
hf download jgkawell/jarvis en/en_GB/jarvis/medium/jarvis-medium.onnx.json --local-dir .

README.md:18-31:

bash
pipx install piper-tts
sudo apt install ffmpeg  # or equivalent
bash
# Create voice directory
mkdir -p ~/.local/share/piper/voices/en_GB

# Download models (via HuggingFace)
cd ~/.local/share/piper/voices/en_GB
hf download jgkawell/jarvis en/en_GB/jarvis/high/jarvis-high.onnx --local-dir .
hf download jgkawell/jarvis en/en_GB/jarvis/high/jarvis-high.onnx.json --local-dir .
hf download jgkawell/jarvis en/en_GB/jarvis/medium/jarvis-medium.onnx --local-dir .
hf download jgkawell/jarvis en/en_GB/jarvis/medium/jarvis-medium.onnx.json --local-dir .

package.json:22-26:

json
"peerDependencies": {
  "piper-tts": ">=1.0.0",
  "ffmpeg": ">=4.0"
},

Technical Analysis

The installation instructions do not pin piper-tts to an exact version. The package metadata similarly permits any future version satisfying >=1.0.0, while FFmpeg permits any version satisfying >=4.0. The Hugging Face downloads identify repository paths but do not specify an immut ...[truncated 2639 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin piper-tts to a reviewed exact version rather than installing the latest available release.
  2. Replace broad package.json ranges with exact, tested versions or document that these entries are informational system prerequisites rather than installable npm peer dependencies.
  3. Pin every Hugging Face download to an immutable repository commit using the CLI revision option.
  4. Publish expected SHA-256 checksums for every ONNX and JSON artifact and require verification before use.
  5. Maintain a versioned manifest containing artifact source, immutable revision, size, and digest.
  6. Use trusted package indexes and operating-system repositories explicitly, and avoid adding unverified mirrors.
  7. Test upgrades in an isolated environment before changing pinned versions or model revisions.
  8. Document a controlled update process that includes provenance review, checksum updates, vulnerability scanning, and rollback instructions.
  9. Run TTS processing with least privilege and restrict network, filesystem, and device access where practical.
Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Rogue AgentSelf-Modification, Session Persistence
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (6)

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · README.md (reported line 18)May include surrounding context.

Prerequisites

bash
pipx install piper-tts
sudo apt install ffmpeg  # or equivalent

Install Voice Model

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 39)May include surrounding context.

Prerequisites

bash
pipx install piper-tts
sudo apt install ffmpeg  # or equivalent

Install Voice Model

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · README.md (reported line 23)May include surrounding context.

Install Voice Model

bash
# Create voice directory
mkdir -p ~/.local/share/piper/voices/en_GB

# Download models (via HuggingFace)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The documentation explicitly encourages streaming generated audio and pushing it to an Android device via ADB, which moves user-provided content off the local machine without any privacy or consent warning. If users synthesize sensitive text, this can unintentionally transfer confidential data to another endpoint or persist it on the device.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 44)May include surrounding context.

Install Voice Model

bash
# Create voice directory
mkdir -p ~/.local/share/piper/voices/en_GB

# Download models via HuggingFace CLI

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The manifest explicitly advertises the ability to push voice announcements to Android devices via ADB over Tailscale, which is network- and device-affecting behavior, but it provides no warning, consent language, or usage constraints. In an agent/skill ecosystem, understated remote device interaction increases the risk of unexpected actions on connected endpoints and can mislead users about the operational scope of the skill.

Content

No source excerpt is available for this finding.