Back to skill

Security audit

MiniMax Vision Captcha

Security checks for vulnerabilities and agentic risk

Overview

This CAPTCHA/image-analysis skill has legitimate vision-use pieces, but it also enables CAPTCHA bypass and ships an unsafe script that can upload unintended screenshots and allow command injection.

Review carefully before installing. Do not use this for CAPTCHA or slider challenges unless you are certain it is allowed for that site and account. For benign OCR or screenshot analysis, provide only explicit, reviewed image paths, redact sensitive content first, and avoid running the included helper until the command-injection bug and automatic screenshot fallback are fixed.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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

Error
Location
scripts/solve-captcha.js:81
Finding

Shell Command Injection Through Unsanitized Command-Line Arguments

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
scripts/solve-captcha.js:51
Finding

Automatic Disclosure of an Unintended Browser Screenshot

Content
View full analysis
f.endsWith('.png')) .map(f => ({ name: f, time: fs.statSync(path.join(mediaDir, f)).mtime })) .sort((a, b) => b.time - a.time); if (files.length > 0) { imagePath = path.join(mediaDir, files[0].name); console.log('使用最新截图:', files[0].name); } ``` The automatically selected file is subsequently submitted for image analysis: ```js const cmd = `mcporter call minimax-coding-plan.understand_image prompt="${prompt}" image_source="${imagePath}"`; const result = execSync(cmd, { encoding: 'utf-8', timeout: 30000 }); ``` ### Technical Analysis When no image path is supplied, the script enumerates `/root/.openclaw/media/browser`, sorts all PNG files by modification time, and selects the newest one without verifying its origin, purpose, content, or relationship to the current request. The selected screenshot is then supplied to the configured MiniMax vision service through `mcporter`. The latest screenshot may belong to another browsing workflow and may contain authentication pages, messages, personal information, account details, or other sensitive content. Although the documentation describes screenshot analysis, it does not clearly disclose that omitting an image causes the script to inspect a privileged browser-media directory and automatically submit the newest PNG. Merely printing the selected filename does not provide meaningful consent because submission follows immediately without confirmation. ### Attack Path 1. A browser screenshot containing sensitive content is stored in `/root/.openclaw/media/browser`. 2. The Skill is i ...[truncated 1315 chars]
Remediation
View remediation
Vulnerability Patterns
  • 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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (6)

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The script builds a shell command with both prompt and imagePath interpolated directly into a string passed to execSync. Because these values come from command-line arguments and are not safely escaped, an attacker can inject shell metacharacters and execute arbitrary commands on the host, which is far more dangerous than the stated image-recognition purpose.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill description is broad enough to trigger on general image analysis, screenshot extraction, and CAPTCHA-related tasks, which increases the chance the agent will invoke it in contexts involving sensitive visual data or prohibited automation. Broad activation criteria are especially risky here because the skill explicitly covers CAPTCHA solving, making accidental misuse more likely.

Content

No source excerpt is available for this finding.

Ssd 4

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill directly provides a workflow for using vision analysis to solve CAPTCHA and slider challenges, which are access-control and anti-abuse mechanisms. Enabling automated or assisted CAPTCHA solving materially supports bypass of protections intended to prevent scripted access, fraud, and account abuse.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The documentation instructs users to send screenshots or image paths to an external vision model without warning that images may contain sensitive information such as credentials, personal data, or internal application content. This creates a real risk of unintended data exfiltration to a third-party service.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The trigger list includes broad phrases such as '图片识别' and '视觉分析', which can match many unrelated user intents and cause the skill to activate unexpectedly. In this skill’s context, unintended activation is more concerning because it performs captcha and image-analysis tasks, which could route screenshots or sensitive visual content to the skill when the user did not specifically request captcha handling.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The script sends the selected image to an external vision service (mcporter call ... understand_image) without any user-facing notice, consent flow, or minimization of potentially sensitive screenshot contents. Since the tool auto-selects recent browser screenshots, it may exfiltrate credentials, personal data, or session-related information from images users did not explicitly intend to upload.

Content

No source excerpt is available for this finding.