Back to skill

Security audit

MiniMax Vision

Security checks for vulnerabilities and agentic risk

Overview

This image-analysis skill appears purpose-aligned, but it can automatically send image paths and prompts to MiniMax through a broad local command without enough consent or path validation.

Review before installing. Use this only if you are comfortable with images and prompts being processed by the configured MiniMax MCP service. Avoid using it on sensitive photos, IDs, documents, private screenshots, or confidential business images unless the service configuration and retention policy are acceptable. The publisher should narrow activation, add explicit user confirmation or disclosure, validate image paths to the inbound media directory, and remove the hard-coded PATH modification.

Vulnerability Patterns
  • Tool Hijacking and SpoofingModifies or replaces tools so legitimate-looking calls execute attacker logic
  • 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 (2)

T09 · Insecure Skill Coding Practices

Warning
Location
minimax_vision.py:13
Finding

Unrestricted Local File Path Submission to the Vision Service

Content
View full analysis
[问题]") sys.exit(1) image_path = sys.argv[1] prompt = sys.argv[2] if len(sys.argv) > 2 else "描述这张图片的内容" print(f"识别图片: {image_path}") print(f"问题: {prompt}") print("-" * 30) result = call_mcp(image_path, prompt) ``` ### Technical Analysis The script accepts an arbitrary path from `sys.argv[1]` and forwards it to `minimax.understand_image` as `image_source`. It does not canonicalize the path or verify that it: - Resides within the documented `~/.openclaw/media/inbound/` directory. - Refers to a regular file. - Is not a symbolic link. - Has an approved image extension. - Contains a valid supported image format. The restriction described in `SKILL.md` is therefore informational rather than enforced. An attacker who can influence the script invocation can supply an absolute path, directory-traversal path, or symbolic link targeting another readable local file. Whether the target file's contents are successfully processed or disclosed depends on how the configured MCP implementation interprets `image_source`. Nevertheless, the wrapper unnecessarily allows arbitrary local paths to cross the MCP trust boundary. ### Attack Path 1. An attacker obtains influence over the image-path argument, such as through an automated agent invocation or another component that passes user-controlled paths to this script. 2. The attacker supplies a sensitive readable path, a traversal path, or a symlink rather than a file in the approved inbound-media ...[truncated 850 chars]
Remediation
View remediation

T07 · Tool Hijacking and Spoofing

Warning
Location
minimax_vision.py:22
Finding

Hard-Coded Search Path Can Select an Untrusted mcporter Executable

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (7)

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · minimax_vision.py (reported line 24)May include surrounding context.

python
]
    
    # 设置环境变量
    env = os.environ.copy()
    env["PATH"] = "/Users/js/.local/bin:" + env.get("PATH", "")
    
    try:

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill invokes an external command via mcporter call ... but does not declare any permissions or allowed-tools scope. That creates an implicit shell/tool execution capability that may bypass least-privilege controls and makes it harder for the host to constrain what the skill can execute or access.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The trigger phrases are broad, common conversational phrases like '帮我看看' and '这张是什么', which can cause accidental activation during normal chat. That can lead to unintended processing of user content and unexpected transmission of images to an external service without sufficiently clear user intent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill describes image recognition but does not warn users that their images are sent to an external MiniMax/MCP service for analysis. Because images may contain sensitive personal or confidential information, omission of this disclosure undermines informed consent and can expose private data to third-party processing.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The workflow says to auto-call the tool whenever a user sends an image or asks for analysis, but it does not define clear consent, ambiguity handling, or safety boundaries. In context, this means routine image uploads in media/inbound/ may be sent to an external MCP service automatically, increasing privacy and data-handling risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill sends a user-supplied image and prompt to an external MCP service without any explicit notice, consent flow, or privacy disclosure. In an image-understanding skill, this is especially relevant because images may contain sensitive personal, biometric, document, or location data, creating a real risk of unintended third-party data exposure.

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 · minimax_vision.py (reported line 28)May include surrounding context.

python
env["PATH"] = "/Users/js/.local/bin:" + env.get("PATH", "")
    
    try:
        result = subprocess.run(
            cmd,
            capture_output=True,
            text=True,