Back to skill

Security audit

voice-output

Security checks for vulnerabilities and agentic risk

Overview

This voice-output skill is mostly coherent for cloud text-to-speech, but it ships a plaintext service token and uses under-scoped invocation instructions that warrant review before installation.

Before installing, require the publisher to remove and rotate the embedded Doubao token, document that response text is sent to a third-party TTS service and played aloud, narrow the trigger phrases, and update invocation guidance so speech text is passed as a literal argument or stdin rather than interpolated into a shell command. Avoid using it for secrets or private content until those issues are fixed.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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 (3)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/voice_speak.py:24
Finding

Hard-Coded Doubao TTS Access Token

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/voice_speak.py:38
Finding

Symlink-Unsafe and Non-Atomic Lock File in Shared Temporary Directory

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:20
Finding

Unsafe Shell Command Template for Agent-Generated Speech Text

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (11)

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill description omits materially important behaviors: outbound transmission of response text to an external TTS provider, local audio file creation, and reliance on local playback rather than a guaranteed specific device. This mismatch can cause sensitive content to be exfiltrated or played aloud unexpectedly because reviewers and users are not given an accurate picture of what the skill actually does.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill also fails to warn that generated audio will be played aloud on a connected speaker, which can expose confidential content to bystanders. In the context of an assistant skill, this increases the risk of inadvertent disclosure in shared homes, offices, or public spaces.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The script embeds a live APPID and access token directly in source code, which exposes secrets to anyone who can read the repository, logs, backups, or deployed files. An attacker could reuse these credentials to access the TTS service, incur charges, abuse the account, or pivot into related cloud resources depending on provider permissions.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
90% confidence
Finding

The skill declares shell, network, and file-writing capabilities implicitly through its documented invocation, but it does not declare any tool scope or permissions. This weakens reviewability and enforcement, making it easier for an agent to use sensitive capabilities without explicit authorization boundaries.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill also fails to warn that generated audio will be played aloud on a connected speaker, which can expose confidential content to bystanders. In the context of an assistant skill, this increases the risk of inadvertent disclosure in shared homes, offices, or public spaces.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The single-word trigger '语音' is too broad and likely to occur in ordinary dialogue, increasing the chance of accidental invocation. Because this skill externalizes and vocalizes content, false triggers can create privacy leaks and confusing or disruptive behavior.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The catch-all trigger phrase 'Or any similar request' is overly vague and can cause accidental activation from normal conversation. In this skill, unintended activation is risky because it may send arbitrary response text to an external TTS service and play it aloud to nearby listeners.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The function sends arbitrary spoken content to an external TTS provider, which can disclose sensitive user data, internal prompts, credentials, or personal information if passed into the script. In a voice-output skill, this is contextually more dangerous because users may assume speech is generated locally and may not realize their text is being transmitted to a third party.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/voice_speak.py (reported line 192)May include surrounding context.

python
f.write(audio_data)

        # 阻塞等待播放完成,带超时保护
        result = subprocess.run(
            ["afplay", "-q", "1", tmp_file],  # -q 1 = 高质量,阻塞
            capture_output=True,
            timeout=AFPLAY_TIMEOUT,

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

SQP-3 applies to all file types and includes language or locale policy violations. This markdown reference forces a single language presentation without any opt-in, alternative language guidance, or justification that the skill is region-specific.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The natural-language interface, examples, and defaults are entirely presented in Chinese, and the script's usage suggests operation around Chinese text and a Chinese voice model. There is no indication that users may choose another language or locale, which can violate a language-choice policy when not explicitly justified.

Content

No source excerpt is available for this finding.