ClawHub
Back to skill

Security audit

Text to Audio GPT

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed text-to-audio workflow that uses browser-based TTS services and local FFmpeg merging, with privacy cautions and no evidence of hidden persistence or exfiltration.

Install only if you are comfortable using browser-based TTS services and FFmpeg. Do not provide private, confidential, or credential-like text for remote narration unless you explicitly approve sending it to the selected TTS provider; prefer a local TTS path for sensitive content.

A.I.G

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

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (9)

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The description claims an end-to-end text/document narration skill that speaks content in an expressive voice and produces a checked MP3. The code shown does only the final assembly/validation portion: it accepts input MP3 parts, probes metadata with ffprobe, concatenates them with ffmpeg using stream copy, validates duration, and runs a decode check. There is no text ingestion, document handling, speech generation, voice selection, or browser/GPT control logic. While the validation/MP3 assembly aspect aligns partially with the description, the primary declared purpose is materially broader and different from the actual code behavior.

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
89% confidence
Finding

The skill instructs use of shell/FFmpeg and file creation/download handling, but it does not declare an explicit tool scope such as allowed-tools or permissions. In agents that honor capability declarations, this mismatch can lead to overbroad tool access, weaker sandboxing, or ambiguous enforcement, increasing the chance of unintended file or command execution.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The file is entirely written and scoped around Russian output, including a section explicitly titled 'Проверенный русский вариант' and service guidance specific to Russian voices. There is no statement that the skill is intentionally region- or language-specific, nor any opt-in mechanism for users who may prefer another language.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The manifest’s user-facing description and default prompt are entirely Russian-language text. This can violate language/locale policy expectations because the skill appears to force a specific language for all users without opt-in, alternative locale support, or justification that the skill is region-specific.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The manifest advertises the skill for API and Atlas even though the surrounding skill context says it depends on GPT browser control and external media tooling. This mismatch can cause unsupported deployment into environments without the required isolation, controls, or runtime assumptions, leading to failed execution, unsafe fallback behavior, or accidental exposure of capabilities beyond the intended platform.

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/merge_audio.py (reported line 17)May include surrounding context.

python
def run(args: list[str], *, cwd: Path | None = None) -> subprocess.CompletedProcess[str]:
    return subprocess.run(args, cwd=cwd, capture_output=True, text=True, check=False)


def probe(path: Path) -> dict[str, object]:

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The file contains multiple natural-language strings shown to users in Russian, beginning with the runtime error messages at L036 and L040 and continuing throughout the script. This forces a specific language/locale without user opt-in or documentation that the tool is intentionally region-specific, which matches the language/locale policy violation criteria.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

The description omits key operational constraints such as dependence on browser control and FFmpeg/FFprobe, which can mislead operators or orchestration systems about how the skill runs. In security-sensitive environments, undisclosed external-tool or browser dependencies increase the chance of unsafe invocation patterns, broken sandbox assumptions, and improper trust decisions.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.