Back to skill

Security audit

Volcengine Digital Human Video Generator

Security checks for vulnerabilities and agentic risk

Overview

The skill largely performs the advertised digital-human video workflow, but it ships cloud credentials and publicly uploads user media in ways that require review before installation.

Review before installing. The publisher should remove and rotate the bundled Volcengine credentials, replace config.json with a placeholder, require explicit confirmation before each public upload, prefer private expiring storage, avoid auto-selecting unrelated inbound images, and isolate temporary files per run. Do not use this skill with private photos, sensitive speech, minors' images, customer data, or regulated content unless those issues are fixed and public hosting is acceptable.

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
config.json:2
Finding

Bundled Volcengine Access Credentials

Content
View full analysis
配置文件""" ak = os.environ.get('VOLC_AK', '') sk = os.environ.get('VOLC_SK', '') if ak and sk: return ak, sk config_path = SKILL_DIR / 'config.json' if config_path.exists(): try: with open(config_path, 'r') as f: config = json.load(f) ak = config.get('ak', '') sk = config.get('sk', '') if ak and sk: return ak, sk except Exception as e: print(f"[警告] 读取配置文件失败: {e}") raise Exception( "❌ 未配置火山引擎凭证!请通过以下方式之一配置:\n" " 1. 设置环境变量: export VOLC_AK='your_ak' VOLC_SK='your_sk'\n" " 2. 创建配置文件: ~/.openclaw/extensions/volc-digital-human/config.json\n" " 内容: {\"ak\": \"your_ak\", \"sk\": \"your_sk\"}\n" " 3. 获取 AK/SK: https://console.volcengine.com/iam/keymanage/" ) AK, SK = load_credentials() ``` ### Technical Analysis The distributed configuration contains an access-key identifier and an encoded secret-key value. The secret is not protected by encryption: Base64 decoding reveals another Base64 value, which decodes to a 32-character hexadecimal secret. Encoding therefore provides no confidentiality. The script automatically loads these values and uses them as signing credentials for Volcengine API requests. This contradicts the documentation stating that users must configure their own credentials and must not commit AK/SK values. The code does not app ...[truncated 1597 chars]
Remediation
View remediation

other

Warning
Location
scripts/volc_digital_human.py:67
Finding

Public Disclosure of User Images and Generated Speech Through Anonymous File Hosts

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/volc_digital_human.py:56
Finding

Predictable Shared Temporary Files Permit Local Disclosure and Symlink Attacks

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • 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
Findings (15)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The documented purpose understates several sensitive behaviors: uploading user media and generated audio to public third-party hosts, reading cloud credentials from env/config, and automatically scanning a fixed inbound directory for the latest image. In this context, that mismatch is dangerous because users may believe they are providing a direct private image-to-video workflow when the skill actually publicizes content and broadens data access beyond the described scope.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The code uploads user images and generated audio to public third-party file-hosting services (catbox.moe and 0x0.st) to obtain URLs, making sensitive content accessible outside the declared service flow. Because these are public hosting endpoints unrelated to the primary vendor, users may unknowingly expose portraits and speech content to broad unauthorized access, retention, or redistribution.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill transfers user-provided images and generated audio to public hosts without any visible warning, consent flow, or minimization controls. In the context of a digital-human generator, these files are likely personally identifying and potentially sensitive, so silent publication to public URLs materially increases privacy and data leakage risk.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill declares capabilities that include environment access, file reads, network access, and shell execution, but it does not define any explicit tool scope or allowed-tools boundary. This increases the chance that the agent can invoke broader-than-expected actions, including reading secrets and accessing local files, without clear user-visible restrictions.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The trigger terms include broad everyday phrases such as '视频合成' and '数字人视频', which can cause the skill to activate in situations where the user did not intend to authorize image upload, TTS generation, or third-party sharing. Because this skill performs network uploads and handles user media, accidental invocation raises meaningful privacy and consent risks.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The voice-matching rules hard-code only zh-CN voices and the manual override examples are centered on Chinese-language gender terms, but the document does not state that the skill is limited to Chinese content or offer users a language choice. This can violate language/locale policy because the skill imposes a specific locale by default without explicit opt-in or justification.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The documentation instructs operators to upload generated audio to a third-party public hosting service unrelated to the core cloud API flow. That creates an unnecessary data exfiltration path for user-provided or synthesized voice content, exposes potentially sensitive media to public access, and adds dependency on an untrusted external service.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The markdown tells users to upload audio to a free public hosting service without any privacy, consent, or retention warning. In this skill context, audio may contain user-supplied speech or generated voice content tied to a personal photo, so public hosting materially increases confidentiality and misuse risks.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
81% confidence
Finding

The natural-language descriptions, status messages, defaults, and sample text are entirely Chinese, and the voice configuration is hardcoded to zh-CN voices. There is no indication that users can opt into another language or that the skill is intentionally limited to a China-specific audience for policy or compliance reasons.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
83% confidence
Finding

The script silently reads VOLC_AK and VOLC_SK from environment variables or a local config file to authenticate API calls. While comments explain configuration for developers, there is no runtime notice to the user that sensitive credentials will be accessed and used.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill performs gender inference on user-supplied photos using face-analysis tooling, which is a biometric-like sensitive attribute inference not required to fulfill the core function of generating a talking-head video. This creates privacy risk, may violate user expectations, and can expose the operator to regulatory or policy issues, especially because it runs automatically without explicit consent.

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/volc_digital_human.py (reported line 249)May include surrounding context.

python
def generate_thumb(video_path, output_path, timestamp="00:00:02"):
    """截取视频缩略图"""
    subprocess.run([
        'ffmpeg', '-i', video_path, '-ss', timestamp,
        '-vframes', '1', '-q:v', '2', output_path, '-y'
    ], capture_output=True, timeout=15)

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/volc_digital_human.py (reported line 284)May include surrounding context.

python
voice = select_voice(gender, is_cartoon)
    print(f"[Step2] 生成TTS,音色={voice},文本: {text[:30]}...")
    
    result = subprocess.run([
        'edge-tts',
        '--voice', voice,
        '--text', text,

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
73% confidence
Finding

The manifest says the skill generates digital-human videos and performs TTS dubbing, but this file specifies invoking a server-installed command-line utility to do so. Using an external executable is a broader operational capability than the manifest communicates, and it is not obviously required if TTS could be handled via an API or library within the described workflow.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The examples and recommended voices are all fixed to zh-CN variants, which imposes a specific language/locale in the skill guidance. The file does not indicate that users may choose other languages or that the Chinese-only scope is a documented, justified regional requirement.

Content

No source excerpt is available for this finding.