Back to skill

Security audit

视频转操作指南

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly does what it claims, but normal use can send a license bearer token to an external service and auto-install unpinned dependencies without clear upfront disclosure.

Install only if you are comfortable with a third-party licensing endpoint receiving XIAPING_TOKEN, automatic unpinned dependency installation, and possible external processing of video-derived audio, screenshots, or transcripts. Prefer reviewing and manually installing pinned dependencies in an isolated environment, using environment variables instead of command-line API keys, and avoiding confidential videos unless the external services and mirror are acceptable under your policy.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (4)

other

Error
Location
video_to_doc.py:38
Finding

Undisclosed Transmission of a Bearer Token to an External Licensing Service

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
video_to_doc.py:123
Finding

Implicit Installation of Unpinned Python and npm Dependencies

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
scripts/transcribe_audio.py:52
Finding

Whisper Model Downloads Are Redirected to an Unverified Third-Party Mirror

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Note
Location
scripts/transcribe_audio.py:107
Finding

OpenAI API Keys Can Be Supplied Through Process Command-Line Arguments

Content
View full analysis
[method] [parameters]") print("Methods: local(default), openai, alibaba, xfyun") print("") print("Examples:") print(" python transcribe_audio.py audio.wav local") print(" python transcribe_audio.py audio.wav openai --api-key YOUR_KEY") print(" python transcribe_audio.py audio.wav local --model small") sys.exit(1) audio_path = sys.argv[1] method = sys.argv[2] if len(sys.argv) > 2 else "local" # Parse additional parameters kwargs = {} args = sys.argv[3:] for i in range(0, len(args), 2): key = args[i].replace("--", "").replace("-", "_") value = args[i + 1] if i + 1 < len(args) else None if value: kwargs[key] = value ``` The parsed key is passed to the OpenAI client: ```python client = openai.OpenAI(api_key=api_key or os.environ.get("OPENAI_API_KEY")) ``` ### Technical Analysis The documented command interface encourages users to pass an API key using `--api-key`. Command-line arguments are commonly stored in shell history and can be exposed through local process inspection, diagnostic tooling, audit logs, crash reports, and job-management systems. The script already supports `OPENAI_API_KEY`, making command-line secret handling unnecessary. The issue is local secret exposure rather than command injection: arguments are parsed directly and are not passed to a shell. ### Attack Path 1. A user follows the documented example and runs the transcription command with a real API key. 2. The shell records the complete command in its history file. 3. While the process is running, the key may also be visible through process-list or monitoring interfa ...[truncated 633 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
Findings (57)

Tainted flow: 'user_token' from os.environ.get (line 44, credential/environment) → requests.post (network output)

Critical
Category
Data Flow
Confidence
96% confidence
Finding

The skill reads a bearer token from an environment variable or CLI-provided token and transmits it to an external service unrelated to the core video-to-document transformation. This creates credential exposure and covert data-flow risk: users must trust a third-party endpoint with a reusable secret, and the skill metadata does not justify needing that credential to perform local conversion.

Content

Scanner excerpt · video_to_doc.py (reported line 50)May include surrounding context.

python
return {"valid": False, "message": "请设置 XIAPING_TOKEN 环境变量"}
    
    try:
        response = requests.post(
            "https://xiaping.coze.site/api/skills/verify",
            headers={"Authorization": f"Bearer {user_token}"},
            json={"skill_id": SKILL_ID},

Tainted flow: 'user_token' from os.environ.get (line 44, credential/environment) → requests.post (network output)

Critical
Category
Data Flow
Confidence
95% confidence
Finding

This sends the same bearer token to a remote consume endpoint to meter usage, again outside the core document-conversion function. Repeated transmission of a reusable environment credential increases the chance of leakage, misuse, or unexpected account actions if the service or endpoint is compromised.

Content

Scanner excerpt · video_to_doc.py (reported line 85)May include surrounding context.

python
return False
    
    try:
        response = requests.post(
            "https://xiaping.coze.site/api/skills/consume",
            headers={"Authorization": f"Bearer {user_token}"},
            json={"skill_id": SKILL_ID},

Tp4

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

If the skill truly performs online license verification, remote API communication, local license caching, usage metering, and automatic dependency installation while presenting itself as a simple offline document-conversion tool, that is a meaningful security and trust issue. Hidden network activity and auto-install behavior can expose data, introduce supply-chain risk, and bypass user expectations about what the skill will execute.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

If the skill truly performs online license verification, remote API communication, local license caching, usage metering, and automatic dependency installation while presenting itself as a simple offline document-conversion tool, that is a meaningful security and trust issue. Hidden network activity and auto-install behavior can expose data, introduce supply-chain risk, and bypass user expectations about what the skill will execute.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

If the skill truly performs online license verification, remote API communication, local license caching, usage metering, and automatic dependency installation while presenting itself as a simple offline document-conversion tool, that is a meaningful security and trust issue. Hidden network activity and auto-install behavior can expose data, introduce supply-chain risk, and bypass user expectations about what the skill will execute.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

If the skill truly performs online license verification, remote API communication, local license caching, usage metering, and automatic dependency installation while presenting itself as a simple offline document-conversion tool, that is a meaningful security and trust issue. Hidden network activity and auto-install behavior can expose data, introduce supply-chain risk, and bypass user expectations about what the skill will execute.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

If the skill truly performs online license verification, remote API communication, local license caching, usage metering, and automatic dependency installation while presenting itself as a simple offline document-conversion tool, that is a meaningful security and trust issue. Hidden network activity and auto-install behavior can expose data, introduce supply-chain risk, and bypass user expectations about what the skill will execute.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

If the skill truly performs online license verification, remote API communication, local license caching, usage metering, and automatic dependency installation while presenting itself as a simple offline document-conversion tool, that is a meaningful security and trust issue. Hidden network activity and auto-install behavior can expose data, introduce supply-chain risk, and bypass user expectations about what the skill will execute.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

If the skill truly performs online license verification, remote API communication, local license caching, usage metering, and automatic dependency installation while presenting itself as a simple offline document-conversion tool, that is a meaningful security and trust issue. Hidden network activity and auto-install behavior can expose data, introduce supply-chain risk, and bypass user expectations about what the skill will execute.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

If the skill truly performs online license verification, remote API communication, local license caching, usage metering, and automatic dependency installation while presenting itself as a simple offline document-conversion tool, that is a meaningful security and trust issue. Hidden network activity and auto-install behavior can expose data, introduce supply-chain risk, and bypass user expectations about what the skill will execute.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

If the skill truly performs online license verification, remote API communication, local license caching, usage metering, and automatic dependency installation while presenting itself as a simple offline document-conversion tool, that is a meaningful security and trust issue. Hidden network activity and auto-install behavior can expose data, introduce supply-chain risk, and bypass user expectations about what the skill will execute.

Content

No source excerpt is available for this finding.

Chaining Abuse

High
Category
Tool Misuse
Confidence
75% confidence
Finding

Tool calls are chained to bypass individual safety checks or escalate capabilities beyond what any single tool call would allow.

Content

Scanner excerpt · scripts/install_deps.sh (reported line 24)May include surrounding context.

sh
else
    echo "正在安装 ffmpeg..."
    if [ "$OS" == "linux" ]; then
        sudo apt update && sudo apt install -y ffmpeg
    elif [ "$OS" == "macos" ]; then
        brew install ffmpeg
    fi

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The README states that speech transcription can use an online API, but it does not warn users that video audio may be transmitted to a third-party service. Because these videos may contain sensitive business processes, credentials spoken aloud, or personal data, the omission can lead to unintended data disclosure and privacy/compliance issues.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
89% confidence
Finding

The skill describes operations that inherently require reading local files, writing outputs, invoking shell commands, and potentially using networked transcription services, yet it declares no explicit tool scope or permissions boundaries. This increases the attack surface because an agent may be granted broader capabilities than users expect, with no least-privilege constraint documented in the skill manifest.

Content

No source excerpt is available for this finding.

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 32)May include surrounding context.

bash
# 安装 ffmpeg(视频处理)
# Ubuntu/Debian
sudo apt install ffmpeg

# macOS
brew install ffmpeg

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill processes screenshots and audio transcripts that may contain sensitive UI data, credentials, personal information, or confidential spoken content, and it explicitly routes screenshots to platform image analysis and may use external transcription APIs. Without a clear privacy warning and data-handling notice, users may unknowingly expose sensitive information to third-party or platform services.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The guide explicitly recommends using the OpenAI Whisper API to transcribe extracted audio, but it does not warn users that audio contents may be sent to a third-party service and may contain sensitive speech, credentials, customer data, or internal business information. In a video-to-document workflow, uploaded audio is especially likely to originate from enterprise training or system-demo recordings, so omission of privacy and data-handling guidance creates a meaningful data exposure risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The ffmpeg command uses the '-y' flag, which forces overwriting output files in the target directory. Although the script prints progress messages, it does not clearly disclose to the user that existing matching JPEG files may be replaced without confirmation.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
96% confidence
Finding

The script invokes sudo to run apt update and apt install, which requests elevated privileges and executes package management actions as root. While common for dependency installation, embedding privileged operations in a convenience script increases the blast radius if the script is modified, misread, or run in an unexpected environment.

Content

Scanner excerpt · scripts/install_deps.sh (reported line 24)May include surrounding context.

sh
else
    echo "正在安装 ffmpeg..."
    if [ "$OS" == "linux" ]; then
        sudo apt update && sudo apt install -y ffmpeg
    elif [ "$OS" == "macos" ]; then
        brew install ffmpeg
    fi

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The script performs system-level package installation with sudo and non-interactive flags, which changes the host environment without an explicit confirmation step or warning. In a skill installer context, this is risky because users may execute it expecting a harmless setup helper, but it can modify system packages and trigger privileged operations immediately.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The script installs Python packages directly into the active environment without checking whether it is a virtual environment or obtaining confirmation. This can overwrite or contaminate the user's global Python setup, creating dependency conflicts or unexpected package execution paths.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This code file contains natural-language documentation and runtime messages entirely in Chinese, including the module description and usage text. The file does not offer a language choice or explain that it is intentionally limited to a Chinese-speaking or region-specific environment, which can violate a language/locale policy requiring opt-in or justification.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The script sends raw transcript text to third-party LLM providers when API keys are present, but provides no explicit consent flow, privacy notice, redaction, or control over what content is shared. In this skill’s context, transcripts may contain sensitive spoken operational details, personal data, account identifiers, or business information, so external transmission creates a real confidentiality and compliance risk.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The file-level documentation says the script automatically chooses among dynamic interval, scene detection, and voice alignment strategies. However, the implemented logic only inspects whether audio exists and never uses audio timing, transcription, or alignment to drive extraction, so the documentation actively overstates what the code does.

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/smart_extract.py (reported line 20)May include surrounding context.

python
"-show_format", "-show_streams",
        video_path
    ]
    result = subprocess.run(cmd, capture_output=True, text=True)
    info = json.loads(result.stdout)
    
    duration = float(info["format"]["duration"])