Back to skill

Security audit

gesture-control-generator

Security checks for vulnerabilities and agentic risk

Overview

This skill matches its gesture-control purpose, but generated pages can request and start webcam use on page load while running external CDN scripts.

Review before installing. The skill is not showing evidence of theft or sabotage, but generated scenes may ask for webcam access immediately when opened and rely on third-party scripts. Use it only if you are comfortable with browser camera prompts and CDN-loaded code, prefer mouse-only mode when possible, and avoid opening generated files in contexts where camera access would be sensitive.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (2)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
assets/gesture-scene.js:95
Finding

Camera Access Is Requested and Activated Automatically

Content
View full analysis
{ this._tryStartCamera(); }); ``` ```javascript async _tryStartCamera() { if (!navigator.mediaDevices || !navigator.mediaDevices.getUserMedia) { this.statusEl.textContent = '鼠标控制中(浏览器不支持摄像头)'; return; } try { const stream = await navigator.mediaDevices.getUserMedia({ video: true }); stream.getTracks().forEach(track => track.stop()); await this.startCamera(); } catch (error) { console.warn('摄像头权限检测失败:', error); this._handleCameraError(error); } } ``` ### Technical Analysis Creating a `GestureScene` automatically invokes `_tryStartCamera()` after loading MediaPipe. That function immediately calls `getUserMedia({ video: true })`, causing a browser permission request without a preceding, explicit in-page user action. If permission is granted, the temporary stream is stopped and `startCamera()` opens the camera again for gesture processing. Camera access is relevant to the optional gesture feature, and browser permission controls prevent access without user consent. However, requesting and activating this privacy-sensitive capability automatically violates least-privilege and user-intent principles. It also conflicts with the documented workflow in `README.md`, which states that users enable gesture control by clicking a button. No code was found that explicitly transmits camera frames to an application-controlled server. Frames are passed to the MediaPipe API for hand-landmark processing. ### Attack Path 1. A user opens a generated or reference HTML scene. 2. `GestureScene` initializes automatically. 3. MediaPipe dependencies are loaded from the CDN. 4. `_tryStartCamera()` invokes `navigator.mediaDevices.getUserMedia({ video: true })`. 5. ...[truncated 799 chars]
Remediation
View remediation

T03 · Remote Payload Retrieval and Execution

Warning
Location
assets/gesture-scene.js:391
Finding

Externally Hosted JavaScript Is Retrieved and Executed Without Integrity Verification

Content
View full analysis
{ const script = document.createElement('script'); script.src = src; script.crossOrigin = 'anonymous'; script.onload = resolve; script.onerror = reject; document.head.appendChild(script); }); } ``` The generated template and all reference scenes also execute Three.js directly from the CDN: ```html ``` ### Technical Analysis The application dynamically inserts script elements whose sources point to external CDN resources. Appending these elements to the document causes the downloaded content to execute with the privileges of the page origin. Although dependency versions are pinned and jsDelivr is a recognized CDN, the runtime does not cryptographically verify the returned JavaScript. The effective code executed by the application can therefore differ from the code reviewed in this repository if a package artifact, CDN response, upstream publishing account, DNS path, or delivery infrastructure is compromised. The static Three.js tags likewise omit Subresource Integrity metadata. The `cross ...[truncated 1566 chars]
Remediation
View remediation
``` 3. Replace unrestricted dynamic script insertion with a fixed, audited dependency manifest. Do not allow caller-controlled URLs. 4. Locally host and version-pin MediaPipe runtime, WebAssembly, and model assets rather than resolving them dynamically from a CDN. 5. Add a restrictive Content Security Policy that allows scripts and connections only from required origins and avoids `unsafe-eval` and `unsafe-inline` where feasible. 6. Record dependency hashes in release artifacts and verify them during build or deployment. 7. Establish a dependency update process that includes provenance validation, security review, and regression testing. ]]>
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (16)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The README instructs users to enable camera-based gesture control but does not disclose that webcam access is required or explain the associated privacy implications. In a skill that explicitly uses hand-gesture input via camera, omission of this warning can lead users to grant sensitive permissions without informed consent, increasing privacy and trust risks.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill centers on camera-based gesture control but does not instruct the agent to warn the user that camera access will be requested, what data is processed, or that camera usage is optional. In this context, the generated artifact encourages webcam activation in a browser, so lack of notice meaningfully increases privacy risk and reduces informed consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The workflow directs the agent to copy assets/gesture-scene.js and write a new HTML file, but it does not require disclosure that files will be created or potentially overwritten. In an agent setting with filesystem access, silent file operations can lead to accidental clobbering of existing files or unexpected persistence of executable web content.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The framework initializes by loading MediaPipe support and then attempting to start the camera automatically, which triggers privacy-sensitive behavior before a clear user action specifically enabling gesture mode. In this skill context, camera use is related to the feature, but automatic activation exceeds least-privilege expectations and can surprise users or violate platform consent requirements.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Camera-related functionality is started on initialization without a prior in-app warning or opt-in flow, which is unsafe for privacy-sensitive capabilities. In a gesture-control skill, camera use is expected only after the user knowingly chooses that mode, not as a side effect of page load.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The code dynamically injects executable scripts from a third-party CDN at runtime, creating a supply-chain and script-integrity risk. If the CDN, package, or network path is compromised, arbitrary code would execute in the page origin with access to DOM, camera workflow, and user data.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The scene probes and requests webcam access automatically during startup through _tryStartCamera, rather than waiting for a deliberate user request to use gesture control. Even if the browser prompts for permission, unsolicited access attempts are a privacy and trust issue and can normalize overbroad permission behavior.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The document declares lang="zh", and the visible title/comments are also Chinese-oriented, which indicates a fixed language choice. The policy requires flagging language or locale constraints when the skill forces a specific language without user opt-in or a documented justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The skill documentation appears to force a specific language for all user-facing instructions, and there is no indication that the skill is intentionally limited to Chinese-speaking users or that another language option is available. The policy calls for flagging language or locale constraints when they are imposed without user opt-in or documented justification.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The page fetches third-party scripts from a CDN without any user-facing notice that network requests to external domains will be made. While not as severe as direct code injection by itself, the lack of disclosure reduces transparency and can conflict with privacy and platform expectations.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The document declares lang="zh", and the visible UI text throughout the page is exclusively Chinese. Under the policy, forcing a specific language without user opt-in or a documented regional justification is a natural-language locale violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The HTML root sets lang="zh", and the visible title text is also exclusively in Chinese, which indicates a fixed language choice. The policy for this audit flags locale constraints when a skill forces a specific language without user opt-in or a clearly documented region-specific reason.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

The manifest description says the skill will output a '完整的 HTML 文件' for mouse/gesture dual-mode input. This file is not complete on its own because it requires a remote Three.js CDN script and a separate local '../assets/gesture-scene.js' dependency to function.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The document declares lang="zh", and the visible title text is also exclusively Chinese, which indicates the skill is fixed to a specific language. Under the stated policy, forcing a language or locale without user opt-in or documented justification is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The document declares lang="zh", and the visible title text is also Chinese, which indicates the skill is fixed to a specific language/locale. The provided file does not offer any user opt-in, language selection, or justification that this is a region-specific tool, so it conflicts with the language/locale policy criteria.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The file imports executable JavaScript from a public CDN at runtime, which creates a supply-chain and integrity risk: if the CDN, package, or delivery path is compromised, arbitrary code will run in the user's browser. In this skill's context, the external dependency is not strictly necessary because the scene could vendor or pin the library locally, so the network-loaded script increases attack surface beyond the stated purpose.

Content

No source excerpt is available for this finding.