Back to skill

Security audit

Ai Video Pro

Security checks across malware telemetry and agentic risk

Overview

The core prompt tool is plausible, but the package also contains under-disclosed installers, credential checks, cloud uploads, and social posting features that need human review.

Install only if you intend to use the full video generation and publishing toolkit, not just prompt optimization. Prefer prompt-only mode, review the scripts before running them, avoid setting social cookies or broad cloud credentials unless needed, and require manual confirmation before dependency installs, uploads, API calls, or public posts.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • 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 (13)

subprocess module call

Medium
Category
Dangerous Code Execution
Content
# 安装依赖
        requirements = target_path / "requirements.txt"
        if requirements.exists():
            subprocess.run(
                [sys.executable, "-m", "pip", "install", "-r", str(requirements)],
                check=True, timeout=600,
            )
Confidence
92% confidence
Finding
subprocess.run( [sys.executable, "-m", "pip", "install", "-r", str(requirements)], check=True, timeout=600, )

Lp3

Medium
Category
MCP Least Privilege
Confidence
92% confidence
Finding
The skill declares no permissions, yet the documented behavior includes access to environment variables, shell commands, local file I/O, and networked backends. This creates a serious transparency and consent problem: users and the hosting platform cannot accurately reason about what the skill may access or do, especially when API keys and local system state are involved.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding
The documented purpose emphasizes zero-dependency prompt optimization, but the observed behavior extends into environment probing, dependency installation, local server startup, external API calls, local persistence, cloud uploads, and even social-media publishing. This mismatch is dangerous because users may invoke what appears to be a harmless prompt tool while actually triggering broad system changes, credential use, third-party data transfer, and publication workflows.

Context-Inappropriate Capability

Medium
Confidence
97% confidence
Finding
The script inspects the presence of unrelated social-media credentials such as WEIBO_ACCESS_TOKEN, XHS_COOKIE, and DOUYIN_ACCESS_TOKEN during generic environment detection. Even though it only checks existence, this still inventories sensitive credentials outside the stated core need of video-generation backend selection and increases privacy and trust risk.

Intent-Code Divergence

Medium
Confidence
98% confidence
Finding
The module claims that all installation actions require user confirmation, but the script performs package-manager, pip, and git installation actions immediately when invoked with --backend. This mismatch is dangerous because users or calling agents may trust the documented safety guarantees and unknowingly trigger system modifications and third-party code installation.

Description-Behavior Mismatch

Medium
Confidence
85% confidence
Finding
The file implements multi-platform publishing and cloud upload behavior that materially exceeds the manifest's stated purpose of prompt decoding/video generation. Scope mismatch is dangerous in agent skills because users may authorize or install the skill expecting local prompt optimization, while the code contains functionality to transmit files to external services.

Context-Inappropriate Capability

Medium
Confidence
89% confidence
Finding
The script imports credential access helpers and later uses them to authenticate to third-party platforms and cloud providers outside the advertised decoder-focused scope. Hidden credential use increases the chance of unintended data exfiltration or account actions under user credentials.

Missing User Warnings

Medium
Confidence
79% confidence
Finding
The README promotes one-click publishing to major Chinese social platforms without any warning about account authorization, privacy exposure, content transmission, or risk of unintended posting. In a skill that may handle generated media and user accounts, omitting these warnings can lead users to connect accounts and publish content without understanding the security and reputational consequences.

Missing User Warnings

Medium
Confidence
84% confidence
Finding
The README instructs users to export multiple API keys but provides no guidance on secure storage, scope limitation, rotation, or the fact that prompts and media may be sent to external providers. In the context of a multi-provider video generation skill, this omission increases the chance of credential mishandling and unintentional disclosure of sensitive content to third parties.

Missing User Warnings

Medium
Confidence
96% confidence
Finding
The code checks for multiple sensitive environment variables without any in-code warning, consent prompt, or disclosure that these secrets will be probed and then included in a JSON report as booleans. Secret discovery without transparency is dangerous because it silently reveals what privileged integrations are configured on the host and may violate user expectations or organizational policy.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The script performs outbound HTTP requests to third-party endpoints to test connectivity without any user-facing disclosure or consent. Silent network egress can leak usage metadata, trigger monitoring controls, and create unnecessary privacy and compliance concerns, especially in restricted or air-gapped environments.

Missing User Warnings

Medium
Confidence
99% confidence
Finding
The script automatically executes package-manager, pip, and git installation operations when called with --backend, without any runtime confirmation or clear explicit opt-in. In an agent skill context, this is more dangerous because toolchains may invoke helper scripts on a user's machine, leading to unexpected privileged changes and installation of third-party code.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The cloud upload flow transmits local video data using retrieved cloud credentials without any explicit confirmation, warning, or summary of what will be uploaded and where. In an agent or skill context, this increases the risk of unintended data disclosure, especially because the overall skill is presented primarily as a prompt decoder rather than a publisher/uploader.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.