Back to skill

Security audit

Ai Intelligent Ar Vr Platform

Security checks for vulnerabilities and agentic risk

Overview

The skill describes an AR/VR platform, but its install instructions ask users to clone and run mutable unaudited remote code without integrity checks.

Review the external GitHub repository before installing, prefer a pinned commit and hash-verified dependencies, use a virtual environment or sandbox, and do not run the commands with administrator/root privileges.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T03 · Remote Payload Retrieval and Execution

Error
Location
SKILL.md:45
Finding

Mutable Remote Payload Retrieved and Executed Without Integrity Verification

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 45–48
Vulnerability Type: Remote payload retrieval and execution
Risk Level: High

Vulnerable Code:

bash
git clone https://github.com/openclaw-skills/ai-intelligent-ar-vr-platform
cd ai-intelligent-ar-vr-platform
pip install -r requirements.txt
python app.py

Technical Analysis

The installation instructions clone a mutable external Git repository without pinning a reviewed commit, validating a cryptographic checksum, or verifying a trusted signature. They subsequently install dependencies from an unaudited requirements.txt file and execute the downloaded app.py.

The audited package contains neither the referenced application nor its dependency manifest. Consequently, the effective code executed by users can differ from the code available when this skill was reviewed. Although unsafe dependencies are also part of the exposure, the best matching classification is remote payload retrieval and execution because the instructions explicitly download and run an externally controlled application.

Python package installation can execute package build or installation hooks before app.py is launched. Therefore, malicious code could run during either pip install or python app.py.

Attack Path

  1. An attacker compromises the referenced repository, its maintainer account, or an upstream dependency source.
  2. The attacker modifies the repository's default branch, requirements.txt, or app.py to include a malicious payload.
  3. A user follows the documented installation commands.
  4. git clone retrieves the attacker-controlled, mutable content without integrity validation.
  5. pip install -r requirements.txt may execute malicious package build hooks or install compromised dependencies.
  6. python app.py directly executes the downloaded application.
  7. The payload operates with the permissions of the user who ran the commands.

...[truncated 683 chars]

Remediation
View remediation

Remediation Suggestions

  • Include the complete application source and dependency manifests inside the auditable skill package.
  • If remote retrieval is unavoidable, pin the repository to an immutable, reviewed commit hash rather than cloning and executing the default branch.
  • Verify downloaded content using a trusted cryptographic signature or a securely distributed SHA-256 digest before installation or execution.
  • Replace unconstrained dependencies with a reviewed lockfile containing exact versions and cryptographic hashes, such as a hash-locked requirements file used with pip install --require-hashes.
  • Review direct and transitive dependencies for provenance, typosquatting, known vulnerabilities, and unexpected build hooks.
  • Run the application in an isolated virtual environment or sandbox with minimal filesystem, network, credential, and operating-system permissions.
  • Document the application's required permissions and explicitly instruct users not to run installation or execution commands as an administrator or root user.
  • Perform security review and automated scanning on the exact pinned application revision before publication.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest description and main README content present the skill in Chinese only, which can impose a specific language/locale on users without offering an alternative or opt-in. The policy explicitly calls out language or locale constraints as a violation unless the skill offers choice or clearly justifies the locale restriction.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The manifest description does not align with the stated purpose of an AR/VR development platform and instead uses a generic 'AI intelligent' label. This weakens transparency and can cause the skill to be invoked in unintended contexts, increasing the chance of misuse, misrouting, or users granting trust under inaccurate pretenses.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The description is too vague to meaningfully constrain invocation behavior, making it harder for systems or users to understand what the skill is supposed to do. In agent ecosystems, overly broad metadata can lead to inappropriate selection or accidental use in unrelated tasks, which expands the attack surface even without explicit malicious content.

Content

No source excerpt is available for this finding.