Back to skill

Security audit

Yyds.Auto

Security checks for vulnerabilities and agentic risk

Overview

This skill openly provides powerful Android device control, but its unpinned npm execution and broad privileged actions need careful review before installation.

Install only if you trust the yyds-auto-mcp npm publisher and are comfortable giving an AI agent powerful control over an Android device. Prefer a dedicated test device, pin and verify the npm package version, avoid npx -y auto-fetching, keep remote access limited to trusted networks or localhost forwarding, and require manual review before shell commands, file deletion, APK installation, Python execution, pip changes, or agent_run tasks.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T08 · Insecure Dependencies

Error
Location
SKILL.md:20
Finding

Unpinned External npm Package Is Automatically Retrieved and Executed

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 20–25, 73–76, and 97–101
Vulnerability Type: Unreviewed and unpinned third-party dependency execution
Risk Level: High

Vulnerable Code

yaml
install:
  - kind: node
    package: yyds-auto-mcp
    bins: [yyds-auto-mcp]
bash
npm install -g yyds-auto-mcp
json
{
  "mcpServers": {
    "yyds-auto": {
      "command": "npx",
      "args": ["-y", "yyds-auto-mcp"],
      "env": {
        "YYDS_DEVICE_HOST": "127.0.0.1",
        "YYDS_DEVICE_PORT": "61140"
      }
    }
  }
}

Technical Analysis

The Skill instructs users and compatible runtimes to retrieve and execute yyds-auto-mcp from npm without pinning an exact version or integrity digest. The npx -y configuration also permits package retrieval and execution without an interactive confirmation step.

The audited project contains only SKILL.md; the npm package implementation is not included. Its executable behavior therefore cannot be verified against the documentation during this audit, and its effective payload may change after the Skill has been reviewed. This creates a supply-chain trust boundary in which package compromise, account takeover, or an unexpectedly unsafe future release could introduce arbitrary host-side code.

The package is intended to access a powerful Android automation interface. Documented capabilities include elevated shell execution at line 167, arbitrary Python execution at line 187, file operations, APK installation, and delegated AI-agent execution at line 195. These capabilities amplify the consequences of dependency compromise.

No evidence establishes that the current npm package is malicious. The confirmed issue is the unsafe, mutable, and unreviewed dependency execution mechanism.

Attack Path

  1. A user installs the Skill or configures the documented MCP server.
  2. The installation mechanism runs npm install -g yyds-auto-mcp, or MCP startup invokes npx -y yyds-auto-mcp ...[truncated 1295 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin yyds-auto-mcp to a specific, reviewed version rather than resolving an unconstrained release.
  2. Record and verify package integrity with a lockfile and cryptographic integrity hash.
  3. Remove npx -y from the recommended configuration. Require explicit installation and approval before first execution.
  4. Vendor the executable implementation into the reviewed distribution, or otherwise make its exact source available for security review.
  5. Use a trusted private registry or tightly controlled package mirror where appropriate.
  6. Verify package provenance and signatures during installation, and monitor the publisher account and releases for compromise.
  7. Run the MCP server in a sandbox or dedicated low-privilege account with restricted filesystem, environment-variable, process, and network access.
  8. Disable elevated shell, Python execution, package management, APK installation, and unrestricted file operations unless they are explicitly required.
  9. Apply an allowlist to device operations and require confirmation for destructive or privilege-sensitive actions.
  10. Authenticate and encrypt communications with remote Android devices, particularly when connections extend beyond ADB forwarding or localhost.
  11. Pin the reviewed package version in all installation metadata and example configurations so automated installation cannot silently select a different payload.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (2)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill advertises highly sensitive capabilities including shell execution, file read/write/delete, app install/uninstall, screenshot/OCR, clipboard access, and Python code execution, but provides no user-facing safety boundaries, consent requirements, or warnings about destructive and privacy-invasive actions. In the context of an AI-agent skill, this increases the chance of unintended device takeover, data exfiltration, or destructive actions being triggered through ordinary prompts or prompt injection.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill states that it will automatically establish ADB forwarding and start the on-device engine, which are system-changing actions that enable remote control of the device, yet it does not warn users about trust assumptions, exposed control surface, or the implications of connecting to remote/WiFi devices. In an agent-controlled setting, silent bootstrap behavior reduces user awareness and can normalize privileged connectivity changes that expand attack surface.

Content

No source excerpt is available for this finding.