Back to skill

Security audit

电商TVC广告片工作室|AI-HIVE

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly matches its AI-HIVE video-production purpose, but its API client can send a bearer API key to any configured base URL, which users should review before installing.

Install only if you trust AI-HIVE and can protect the API key. Avoid using --base-url, AI_HIVE_BASE_URL, or edited ~/.ai-hive/config.json values unless you fully trust the endpoint, because the bearer key will be sent there. Expect local media reads/uploads only for files you explicitly pass, downloads to the configured output directory, and ffmpeg execution for local video edits.

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 (2)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/videogen.py:92
Finding

Bearer API Key Can Be Transmitted to an Arbitrary Configured Endpoint

Content
View full analysis
Remediation
View remediation

other

Note
Location
scripts/videogen.py:615
Finding

Initialization Opens a Vendor URL with Undisclosed Skill Attribution

Content
View full analysis
`. The code comment expressly identifies these parameters as reserved for backend attribution. Opening the API-key page is relevant to initialization, but transmitting source-attribution data is not necessary to configure the API credential. The Skill documentation does not clearly disclose this attribution event or provide an opt-out before opening the browser. The transmitted data is limited: the reviewed code does not put an API key, local file contents, or personal data directly into this URL. Nevertheless, the request allows the vendor to associate a browser visit with use of this CLI Skill, in addition to ordinary browser metadata such as IP address and user-agent information. ### Attack Path 1. The user runs `python3 scripts/videogen.py init`. 2. The program constructs the vendor URL with fixed `from=cli-skill` attribution and the configured Skill name. 3. `webbrowser.open()` launches the user's browser without a separate attribution-consent prompt. 4. The browser requests the URL from AI-HIVE. 5. The vendor can record the source and Skill identifier alongside normal web-request metadata. No third-party attacker takeover or credential disclosure is demonstrated by this behavior; the risk concerns transparency and unnecessary vendor attribution. ### Impact Assessment The vendor can learn that a browser ...[truncated 320 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • 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
Findings (9)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill description promises a structured AI-HIVE advertising-production workflow, but the detected behavior appears to center on generic local ffmpeg media processing and lacks the declared orchestration, intake, and API-integration features. This mismatch is dangerous because users and agents may grant trust or permissions based on the stated purpose while the actual behavior operates differently, weakening reviewability and increasing the chance of unintended command execution or misuse.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill advertises and demonstrates code paths that can use environment variables, filesystem access, network calls, and shell execution, but it declares no explicit tool scope or permission boundaries. In an agent environment, this can lead to overbroad capability exposure, making it easier for downstream prompts or modified scripts to access secrets, invoke external services, or run local commands without clear policy enforcement.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The default prompt hardcodes Chinese output ('production-ready Chinese workflow') regardless of the user's preferred language. This can override user intent, reduce transparency, and increase the chance that users approve or execute generated workflow/code they do not fully understand, especially in a skill that emits runnable commands and production deliverables.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

Implicit invocation is enabled without narrow trigger constraints, so the skill may activate automatically on broad ad/video-related requests and begin producing executable AI-HIVE commands or workflow outputs without clear user consent. In this skill's context, that is more dangerous because it interfaces with external APIs, uploads reference assets, performs pricing/routing decisions, and generates runnable operational steps that could lead to unintended actions, cost exposure, or misuse.

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
93% confidence
Finding

The module docstring and the code expose general-purpose commands for text chat, image generation, model listing, user-info retrieval, and generic media upload (for example L170-L219, L221-L232, L421-L499, L665-L727), while the manifest describes a specialized ecommerce TVC ad studio focused on concepting, storyboards, AI previsualization, video generation workflow, and related delivery steps. Those extra generic AI operations are not obviously required by the stated skill purpose and make the implementation materially broader than the manifest suggests.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This code file contains natural-language instructions, help text, and descriptions that are presented only in Chinese, including the module docstring and later CLI help strings. Under the policy, forcing a specific language without user choice is a natural-language policy violation unless the locale restriction is explicitly justified, which it is not here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The file embeds the skill description and name entirely in Chinese in user-facing natural-language strings. Under the policy rule for language/locale constraints, this can be a violation because the skill presents a fixed language without any opt-in or alternative locale handling.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The manifest centers on producing ecommerce TVC advertising outputs via AI-HIVE model configuration, upload, pricing, routing, polling, and download. Exposing a dedicated user-info endpoint and CLI command for querying user information and wallet balance (L170-L171, L421-L423, L665-L667) adds an account-inspection capability unrelated to creative video production itself.

Content

No source excerpt is available for this finding.