Back to skill

Security audit

Ai Intelligent Churn Prediction

Security checks for vulnerabilities and agentic risk

Overview

The skill describes a customer churn prediction tool, but its install instructions run unpinned code and dependencies from a mutable GitHub repository outside the reviewed package.

Review the remote GitHub repository and dependency list before installing. Prefer a pinned commit, verified checksum or signed release, and an isolated environment with no sensitive credentials before running python app.py.

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:41
Finding

Mutable Remote Payload Retrieval and Execution

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 41–44
Vulnerability Type: Remote code retrieval and insecure dependency installation
Risk Level: High

Vulnerable Code

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

Technical Analysis

The distributed skill does not include the application source code or its dependency manifest. Instead, its installation instructions tell users to clone a mutable remote repository, install all packages listed in an unaudited remote requirements.txt, and execute the remote app.py.

The Git repository is not pinned to a reviewed commit, release tag, or cryptographically verified artifact. Consequently, the code retrieved when a user follows these instructions may differ from the code available at audit time. The dependency versions and integrity hashes are also unavailable in the audited package, preventing verification of the software supply chain.

This creates two related risks:

  1. The remote repository owner or an attacker who compromises the repository could alter app.py or other imported source files to execute an arbitrary payload.
  2. The remote requirements.txt could reference malicious, compromised, unpinned, or substitution-prone packages whose installation scripts execute code during pip install.

No malicious payload is embedded in the two audited files, and the instructions do not automatically execute without user action. Nevertheless, following the documented installation procedure crosses the audit boundary and runs mutable, unaudited content.

Attack Path

  1. An attacker compromises the referenced repository, its maintainer account, or a dependency used by its remote requirements.txt.
  2. The attacker inserts malicious code into app.py, an imported module, or a package installation hook.
  3. A user follows ...[truncated 1114 chars]
Remediation
View remediation

Remediation Suggestions

  1. Include the complete reviewed application source and dependency manifests in the skill package so the executed implementation is within the audit boundary.
  2. If remote retrieval is unavoidable, pin the repository to an immutable full commit hash rather than cloning the mutable default branch.
  3. Verify the downloaded source against an authenticated checksum or cryptographic signature before installation or execution.
  4. Pin every Python dependency to an exact reviewed version and use a lock file containing package hashes. Install with hash enforcement, such as pip install --require-hashes.
  5. Use trusted package indexes explicitly and prohibit unreviewed direct URL, Git, local-path, and alternative-index dependencies.
  6. Audit package metadata and build scripts because Python packages can execute code during source builds or installation.
  7. Run the application in an isolated, least-privilege environment with restricted filesystem access, no unnecessary credentials, and limited network connectivity.
  8. Establish a release process in which source revisions and dependencies are reviewed, signed, and reproducibly associated with the declared skill version.
Vulnerability Patterns
  • 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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The provided manifest description at L4 is a generic placeholder ('AI intelligent ai-intelligent-churn-prediction') rather than describing the skill's stated purpose of customer churn prediction, risk identification, and retention strategy. This creates a semantic mismatch between what the skill is supposed to do in context and what the manifest actually claims.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

The visible user-facing description and headings are presented only in Chinese, which can impose a language choice on users without opt-in. The policy allows locale constraints when clearly documented and justified, but this file does not state that the skill is region- or language-specific.

Content

No source excerpt is available for this finding.