Back to skill

Security audit

Xiaohongshu Assistant

Security checks across malware telemetry and agentic risk

Overview

This skill is advertised as a safe Xiaohongshu copywriting helper, but it also ships callable code for account login, publishing, comments, likes, bookmarks, and local account/draft storage.

Review before installing. The visible assistant flow mostly generates copy and warns users to publish manually, but the package still includes callable automation that could act on a real Xiaohongshu account if invoked and if the external XiaohongshuSkills scripts exist. Prefer a cleaned version that removes xhs_publisher.py and xhs_accounts.py, or one that clearly discloses account automation and requires explicit confirmation for every publish, comment, like, or bookmark action.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (13)

subprocess module call

Medium
Category
Dangerous Code Execution
Content
]
    
    try:
        result = subprocess.run(cmd, capture_output=True, text=True, timeout=30)
        if result.returncode == 0:
            return {"success": True, "message": f"{action} 成功"}
        else:
Confidence
97% confidence
Finding
Unlike the other subprocess cases, this function constructs a subcommand from unvalidated user input via f"note-{action}". While shell injection is still avoided, an attacker can invoke unintended privileged operations implemented by the downstream script if additional note-* subcommands exist, making this a command-surface expansion issue.

Lp3

Medium
Category
MCP Least Privilege
Confidence
85% confidence
Finding
The skill declares only a benign marketing-assistant purpose, yet static analysis detected file read, file write, and shell-capable behavior without corresponding permission declarations. Undeclared execution and filesystem access reduce transparency, can bypass user expectations, and create room for abuse if supporting scripts perform actions beyond simple content generation.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding
This is a strong description-behavior mismatch: the skill presents itself as a copywriting and optimization assistant, but the detected capabilities include account management, login, automated posting, image upload, note interaction, and engagement actions. Hidden automation against a social platform is dangerous because it can lead to unauthorized account actions, credential handling risks, spam/abuse, and user deception about what the skill actually does.

Intent-Code Divergence

Medium
Confidence
95% confidence
Finding
The README repeatedly asserts that auto-publishing and account-management features are removed or disabled, yet the documented repository structure still includes publisher, account-management, and account configuration files. This is a trust and transparency problem: users or downstream agents may rely on the README's safety claims while dangerous automation-capable code may still be present and usable elsewhere in the repo.

Intent-Code Divergence

Medium
Confidence
92% confidence
Finding
The README says auto-publishing has been removed or disabled, but other README sections such as the changelog and file list indicate that one-click publishing and multi-account management still exist in the project history or codebase. Contradictory safety messaging is risky because it can mislead reviewers and users into underestimating the presence of platform-automation functionality that may trigger account abuse, policy violations, or unsafe execution paths.

Description-Behavior Mismatch

High
Confidence
99% confidence
Finding
The manifest describes a content-generation and optimization assistant, but this module performs authentication and automated publishing to a live social-media account. That mismatch is dangerous because operators may grant the skill broader trust than intended, enabling real account actions and content publication under misleading capability disclosure.

Description-Behavior Mismatch

High
Confidence
99% confidence
Finding
This code performs engagement actions such as commenting, liking, and bookmarking, none of which are covered by the stated assistant purpose. Hidden interaction capabilities are especially risky in an agent context because they can manipulate third-party content and reputation using a user's authenticated account without informed expectation.

Context-Inappropriate Capability

Medium
Confidence
94% confidence
Finding
The skill can spawn external login/publishing automation scripts despite being presented as a content assistant. This matters because invoking external automation expands the trust boundary and can trigger powerful account actions not justified by the declared functionality.

Context-Inappropriate Capability

High
Confidence
98% confidence
Finding
Support for account authentication and account addition is outside the expected scope of a copywriting/optimization assistant. In practice, this grants the skill the ability to manage identities and authenticated sessions, increasing the chance of unauthorized account use or operator deception.

Missing User Warnings

Medium
Confidence
82% confidence
Finding
The module writes account names, notes, default-account selection, and login/publish activity history to a local JSON file without any disclosure, consent flow, or protective controls. Even though this is not highly sensitive credential storage, persistent behavioral/account metadata can expose private operational details to other local users, backups, or accidentally committed files.

Missing User Warnings

Medium
Confidence
98% confidence
Finding
The xsec_token is supplied on the command line to a subprocess, where it may be exposed through process listings, audit logs, crash reports, or debugging tools. Because this token authorizes access to note details, leakage could enable unauthorized API operations tied to the user's account.

Missing User Warnings

Medium
Confidence
96% confidence
Finding
This function can post a comment through an authenticated subprocess without any confirmation, policy check, or disclosure to the user. In an agent setting, silent state-changing actions on external platforms can be abused for spam, impersonation, or unwanted account activity.

Missing User Warnings

Low
Confidence
97% confidence
Finding
Like and bookmark actions are executed directly against an authenticated account with no confirmation or policy gate. Even though each action is small, they can be chained for covert engagement manipulation or unauthorized account behavior, and the hidden nature makes the skill context more dangerous.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.