Back to skill

Security audit

jarvis-voice

Security checks for vulnerabilities and agentic risk

Overview

ClawScan’s verdict is benign with medium confidence: it found no AIG issues, while Skillspector reported several low-probability findings and flagged a potentially undisclosed deletion of voice, session, and humor configuration files.

The verdict is benign, not a guarantee of safety. Before installing, consider whether you want a skill that can use shell commands and may delete local voice, session, and humor configuration files. Coverage was partial: all six listed components were scanned, but three local path-like references were unresolved, so the evidence is incomplete.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (8)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied code only removes files; it does not establish the advertised persona, voice, or humor. Although the filenames relate to the description, deleting these files is a destructive operation not disclosed by the declared purpose, and the supplied chunk provides no evidence that it is a supporting setup step.

Content

No source excerpt is available for this finding.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · SKILL.md (reported line 104)May include surrounding context.

bash
JARVIS_MUTE=1 jarvis "..."              # silence one call
touch ~/.openclaw/jarvis-voice.mute     # silence permanently
rm ~/.openclaw/jarvis-voice.mute        # audible again

Where it stays quiet on its own. If the caller sets JARVIS_CHANNEL to a messaging

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · SKILL.md (reported line 261)May include surrounding context.

Uninstall — removes every persistent change, immediately:

bash
rm -f ~/.openclaw/workspace/{VOICE,SESSION,HUMOR}.md

Precedence, so there are no surprises. Your instructions beat these files, every time. Ask

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · bin/jarvis (reported line 30)May include surrounding context.

text
# --- OFF SWITCH (checked first, before anything runs) -----------------------
# Silence the voice permanently:  touch ~/.openclaw/jarvis-voice.mute
# Silence it for one command:     JARVIS_MUTE=1 jarvis "..."
# Re-enable:                      rm ~/.openclaw/jarvis-voice.mute
MUTE_FILE="${JARVIS_MUTE_FILE:-$HOME/.openclaw/jarvis-voice.mute}"
if [ -n "${JARVIS_MUTE:-}" ] || [ -f "$MUTE_FILE" ]; then
  exit 0

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · templates/VOICE.md (reported line 1)May include surrounding context.

md
# VOICE.md — Jarvis voice output rules

> **Optional file.** Copying this into your workspace root changes agent behaviour in every
> future session. Delete it to stop, immediately and completely. Your explicit instructions

Context Leakage

High
Category
Data Exfiltration
Confidence
85% confidence
Finding

Code or instructions that leak agent conversation context to external services, potentially exposing sensitive user interactions.

Content

Scanner excerpt · templates/VOICE.md (reported line 43)May include surrounding context.

md
overheard and the transcript is visible to anyone reading the chat later. Summarise instead:
  "the credentials are in the reply below."
- **Metadata.** Channel gating uses only the channel name the runtime provides. Do not read,
  parse or log session identifiers for this purpose.
</consent_and_silence>

<rules>

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 138)May include surrounding context.

md
3. Keep spoken text ≤ 1500 characters; sherpa-onnx truncates above that.
4. One `jarvis` audio call per response — stacked calls fight over the audio device.
5. The bundled Alan voice is **English only**. Do NOT silently translate or condense someone's
   content into English just to make it speakable — that changes their meaning without asking.
   If the reply is not in English, say so and offer the choice: keep it text-only, speak an
   English summary if they want one, or install a voice for their language. Text-only is the
   default when in doubt.