Back to skill

Security audit

爆款广告变量拆解|AI-HIVE

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly coherent for Chinese ad-analysis and AI-HIVE media generation, but its API client can send a stored bearer key to an unrestricted endpoint and can write downloaded results using unsafe server-controlled path data.

Review before installing. Use only the default AI-HIVE endpoint unless you fully trust a custom endpoint, avoid setting AI_HIVE_BASE_URL from untrusted instructions, keep the API key scoped and revocable, and run downloads in a low-risk output directory. Confirm before billable generation and avoid using this skill for unauthorized ad copying or unsupported product claims.

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 (3)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/videogen.py:91
Finding

Bearer API Key Can Be Sent to an Arbitrary Configured Endpoint

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/videogen.py:397
Finding

Server-Controlled Task Type Can Escape the Download Directory

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Note
Location
scripts/videogen.py:7
Finding

Runtime Dependency Installation Is Unpinned and Non-Reproducible

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • 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
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (14)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

There is a strong description-behavior mismatch. The description presents a marketing/ad-tech skill focused on analyzing authorized reference ads and producing structured creative-variable breakdowns, with optional API-based asset generation workflows. The actual code contains none of that logic. It is a standalone Python script that shells out to ffmpeg/ffprobe to inspect and edit video files locally. Its commands are limited to probe, trim, aspect conversion, mute, loudness normalization, and concatenation. While video editing could conceivably support ad workflows in a broad sense, this code’s primary function is generic media processing, not the declared ad-analysis or AI-HIVE orchestration capability. Therefore the description does not accurately represent the supplied code chunk.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The declared description emphasizes a marketing-analysis skill that learns transferable variables from authorized ads and outputs structured ad breakdown artifacts. However, the supplied code does not analyze ads or transform business inputs into variable frameworks. Instead, it is an API client/CLI for AI Hive OpenAPI with commands for init, upload, task polling, chat, image generation, and especially video generation through fixed model IDs. The only ad-related elements are SKILL_CONFIG metadata strings such as the title, keywords, and example prompt; these do not correspond to actual analytical behavior. The code’s real function is to submit media-generation jobs and download outputs, not to perform the claimed decomposition workflow. Therefore the description materially misrepresents the code’s primary purpose and capabilities.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
92% confidence
Finding

The skill advertises and demonstrates capabilities that imply access to environment variables, local files, networking, and shell execution, but it does not declare an explicit tool scope or permission boundary. That makes the operational trust model unclear and increases the risk of unintended secret exposure, filesystem access, or command execution if the runtime grants broad defaults.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The activation description and 'when to use' section are entirely defined through Chinese phrases such as '爆款广告拆解', '广告变量', and related Chinese-language workflow descriptions. This creates a language-specific invocation and operating policy without stating that other languages are supported or that the Chinese-only scope is intentional and justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The prompt explicitly instructs the skill to produce a production-ready Chinese workflow, which imposes a language choice regardless of user preference. This is a natural-language policy issue because the file does not offer language selection or explain why Chinese is required.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill enables implicit invocation with no visible trigger constraints, exclusions, or narrowing conditions. Because this skill can produce runnable AI-HIVE commands and workflow outputs for ad generation and creative replication, broad automatic activation increases the chance it is invoked in unrelated or insufficiently authorized contexts, leading to unintended content generation or policy-bypassing assistance.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

This code embeds Chinese-only natural-language identifiers and descriptions in the docstring, argument parser description, skill name, aliases, and artifact labels. Under the policy, forcing a specific language without offering a language/locale choice can be a natural-language policy violation.

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

python
def run(args: list[str]) -> None:
    print(" ".join(args))
    subprocess.run(args, check=True)


def main() -> None:

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The file is packaged as a specialized ad-variable-decomposition/video skill, but the codebase is a general-purpose AI Hive client exposing broader capabilities than the manifest suggests. This creates a scope mismatch that can enable unintended use paths, weaken least-privilege expectations, and make policy enforcement or review based on the manifest unreliable.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The module documentation explicitly describes the tool as a general AI capability wrapper, which contradicts the narrow manifest description. This discrepancy is dangerous because it signals that the reviewed artifact may offer materially broader behavior than users, operators, or automated governance expect.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The top-level docstring is entirely in Chinese and presents the tool as a general-purpose AI skill, but it does not state that the skill is Chinese-only or offer an opt-in language/locale choice. For a general-purpose skill, this creates a natural-language policy concern because the user-facing instructions implicitly force a specific language without documented justification.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The unrestricted chat interface allows arbitrary text and multimodal interaction unrelated to the declared specialized skill. In a skill expected to perform bounded ad-analysis/video tasks, this broadens functionality into a generic AI access channel, increasing the chance of policy bypass, unintended data handling, or misuse beyond the reviewed purpose.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The manifest centers this skill on analyzing authorized reference ads and producing variable tables and test hypotheses, with AI-HIVE used for material generation when needed. This script emits a blueprint that explicitly lists deliverables such as '差异化复刻方案', '逐镜脚本', and '生成提示词和验收清单', which extend from analysis into concrete production-planning/generation outputs not described as this script's purpose in its docstring.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The user-info endpoint exposes account details and wallet/balance information that are not necessary for ad-variable decomposition or video generation. While not a direct exploit by itself, unnecessary access to account metadata expands the available data surface and can leak operational or billing information if invoked inappropriately.

Content

No source excerpt is available for this finding.