Back to skill

Security audit

立白龙虾购

Security checks for vulnerabilities and agentic risk

Overview

This shopping skill is mostly purpose-aligned, but it asks agents to run an unpinned npm command for login, address, order, and points operations involving sensitive user data.

Review this skill carefully before installing. Its shopping functions are coherent, but it processes sensitive account, address, and order data through an unpinned npm package that may change at runtime. Prefer installation only if the package version is pinned or otherwise governed, the runtime is sandboxed, and users must explicitly confirm login, address access, and order creation.

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:65
Finding

Unpinned npm Package Is Downloaded and Executed at Runtime

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 65–71
Vulnerability Type: Unpinned third-party dependency execution
Risk Level: High

Complete Code Snippet:

markdown
## Tool invocation format

All tools are invoked through the `npx -y @libydic/mall` command:

```text
npx -y @libydic/mall <tool-name> --session-key <session-key> [other parameters]
text

The same unpinned command pattern is subsequently used throughout the tool definitions and workflows.

### Technical Analysis

The Skill directs the Agent to run `npx -y @libydic/mall` without specifying an exact package version. When the package is not already available locally, `npx` can retrieve it from the configured npm registry and execute it immediately. The `-y` option suppresses the normal installation confirmation.

No lockfile, integrity hash, package signature, vendored implementation, or reviewed version is included in the project. Consequently, the code executed at runtime can differ from the code that existed when the Skill was audited. A compromised publisher account, registry incident, malicious maintainer update, or package takeover could therefore turn otherwise legitimate commerce commands into arbitrary code execution.

This package is invoked with sensitive arguments and capabilities, including:

- The Agent session key
- User phone numbers and verification codes
- Recipient names, phone numbers, and delivery addresses
- Product and order identifiers
- Order-creation requests
- Access to account, points, order, and logistics information

### Attack Path

1. An attacker compromises the npm package publisher, the package itself, or its release process.
2. The attacker publishes a malicious version of `@libydic/mall`.
3. A user activates a mall workflow described in `SKILL.md`.
4. The Agent runs `npx -y @libydic/mall ...` without an exact version or integrity constraint.
5. `npx` retrieves and ex
...[truncated 1249 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin @libydic/mall to a specific, audited version rather than resolving the latest release:

    text
    npx -y @libydic/mall@X.Y.Z <tool-name> ...
    
  2. Install the dependency during a controlled build process and commit a lockfile containing registry integrity hashes. Invoke the installed binary without allowing runtime package resolution.

  3. Verify the package's provenance, maintainers, source repository, release process, transitive dependencies, and published artifact contents before deployment.

  4. Use trusted-registry controls, package allowlisting, provenance verification, and continuous dependency monitoring.

  5. Disable dependency lifecycle scripts where compatible with the package and deployment process.

  6. Run the package in a restricted sandbox with:

    • Minimal filesystem permissions
    • An explicit network destination allowlist
    • No access to unrelated credentials or environment variables
    • A dedicated low-privilege operating-system identity
    • Resource and execution-time limits
  7. Avoid exposing sensitive values through command-line arguments, which may be visible in process listings or logs. Pass secrets through a protected input channel and redact them from telemetry.

  8. Scope session keys narrowly, rotate them after authentication-sensitive operations, and prevent the package from receiving data unnecessary for the selected command.

  9. Consider replacing the runtime npm executable with a reviewed, versioned API client whose behavior and permitted endpoints are enforced by the host platform.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • 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
Findings (32)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill description markets shopping and support functions but does not clearly warn that it will process sensitive personal data such as phone numbers, addresses, login state, order history, and payment-related actions. Insufficient upfront disclosure can undermine informed consent and increase privacy risk, especially when combined with broad triggers.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

These trigger terms are broad enough to match ordinary user requests that may not be intended for this commerce skill. Because the skill can lead into login, points lookup, address handling, and ordering, accidental activation increases the chance of unexpected collection or exposure of account and shopping-related data.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The listed triggers overlap heavily with common daily language, making misrouting more likely. In this skill context, that matters because the workflow can progress from casual cleaning questions into account-linked mall operations, increasing privacy and consent risks if invoked unintentionally.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The triggers 立白 and 帮我买东西 are especially broad and can capture many benign mentions or generic shopping requests. That is risky here because the skill is empowered to handle login, orders, addresses, and customer-service flows, so accidental routing can expose users to unexpected transactional behavior.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The skill repeatedly instructs the agent to invoke an npm package via npx -y @libydic/mall without pinning a specific version or integrity hash. This allows supply-chain drift: a future compromised or malicious package release could be fetched and executed automatically at runtime, leading to arbitrary code execution and possible theft of session keys, login credentials, addresses, orders, and other user data.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

Using npx -y @libydic/mall without version pinning causes the agent to execute whatever package version is current at invocation time. In a commerce skill that handles authentication and order operations, a malicious upstream update could execute arbitrary code or alter business logic to exfiltrate user data or manipulate transactions.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The core rule says '登录只有一种方式' and '不能使用其他方式登录', asserting auth-start as the only allowed login path. However, the tool documentation immediately defines a separate login --mobile --verify-code step and the login flow actually requires calling it after auth-start, so the documentation contradicts itself about what the code/tooling does.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The documented command format relies on dynamic package download/execution through npx with no pinned version. That creates a classic supply-chain execution risk where an attacker controlling the package or registry path can run arbitrary code in the agent environment and access sensitive mall-session context.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The document claims the tool list contains only the tools described, yet the workflow relies on undocumented session_status, message, and logout tools. Undocumented tool dependencies undermine reviewability and boundary enforcement, and could let a skill exercise capabilities that platform operators did not intend to expose.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

This unpinned runtime dependency is especially risky because it appears in login-related tooling, where the command may process session identifiers and authentication material. A compromised package update could harvest those values or silently modify authentication flow behavior.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The skill instructs execution of an unversioned npm package for user-profile access, which can expose account state and linked user information to any malicious package update. Because the command is fetched on demand, the attack surface includes registry compromise, maintainer compromise, or typosquat/package-substitution scenarios.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

Fetching and running an unpinned package for goods queries introduces arbitrary code execution risk through the dependency supply chain. Even though the action appears low risk, the same package has access to session context and can act across the broader authenticated workflow.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The skill uses the same unpinned npx pattern for detailed goods lookup, continuing the risk of executing attacker-controlled code if the package changes upstream. Since this skill is tied to shopping and authenticated actions, compromise could affect both confidentiality and transaction integrity.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

Address retrieval uses an unpinned npm package, which is dangerous because the command may process personally identifiable information such as recipient names, phone numbers, and physical addresses. A malicious package update could exfiltrate or tamper with that sensitive data.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The area lookup command still depends on an unpinned package fetched at runtime. While the function seems routine, it keeps the same supply-chain exposure and can be abused as a foothold to execute arbitrary code within the agent environment.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

Saving a shipping address through an unpinned npm package is high risk because it combines code-execution exposure with handling of sensitive PII. If the upstream package is compromised, an attacker could collect, alter, or misuse address and contact data at scale.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

Order creation is one of the most security-sensitive operations in the skill, yet it relies on an unpinned npx package. A malicious release could tamper with order parameters, redirect payments, modify pricing, or exfiltrate user and transaction data while appearing to perform a normal checkout.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

Using an unpinned package for order-list retrieval creates supply-chain risk in a context that exposes purchase history and order metadata. A malicious upstream change could silently harvest order details or stage further fraud by learning account activity.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

Order-detail retrieval via an unpinned package inherits the same arbitrary code execution and data-exfiltration risks, with access to detailed transaction records. In this commerce context, that can reveal identities, addresses, and purchase contents and support follow-on fraud or privacy harm.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

Logistics queries still run through an unversioned npx package, enabling malicious package updates to inspect or alter shipment information. Attackers could track deliveries, infer user location patterns, or manipulate logistics outputs to support social engineering.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

Customer-service retrieval uses the same unpinned package pattern, which still exposes the agent to arbitrary code execution through the dependency chain. The immediate function is less sensitive than order creation, but it can still be abused to serve malicious contact details or exfiltrate session information.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The integral query command processes account-related information through an unpinned runtime dependency. A compromised package could steal identifiers and account metadata or manipulate point balances presented to users, undermining trust and potentially enabling fraud.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

Awarding points via an unpinned package is particularly sensitive because it enables state-changing account operations. A malicious update could grant fraudulent rewards, tamper with loyalty balances, or exfiltrate account data while hidden inside an expected points workflow.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

This repeated unpinned npx reference reinforces a pervasive supply-chain vulnerability across the skill. Because the package underpins multiple authenticated and transactional workflows, compromise of the package would have broad blast radius across login, PII handling, ordering, and logistics.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The workflow instructs the agent to call logout, but that tool is not declared in the documented toolset/capabilities. This capability mismatch is dangerous because hidden or undocumented actions bypass normal review boundaries, making it harder to validate what the skill can actually do and increasing the risk of invoking unintended or unauthorized operations.

Content

No source excerpt is available for this finding.