Back to skill

Security audit

Pipeworx dogceo

Security checks for vulnerabilities and agentic risk

Overview

This dog-photo skill appears purpose-aligned, but its MCP setup automatically runs a mutable npm package, which gives more local execution risk than the simple image lookup purpose needs.

Review before installing. Use it only if you are comfortable with requests going through Pipeworx and with npx executing mcp-remote locally. Prefer a pinned, reviewed mcp-remote version, avoid sending secrets or personal data in prompts or tool arguments, and run it in a constrained environment when possible.

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 (2)

T08 · Insecure Dependencies

Error
Location
SKILL.md:43
Finding

Unpinned npm Dependency Is Downloaded and Executed Automatically

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 43-50
Vulnerability Type: Unpinned third-party runtime dependency
Risk Level: High

Vulnerable Code:

json
{
  "mcpServers": {
    "pipeworx-dogceo": {
      "command": "npx",
      "args": ["-y", "mcp-remote@latest", "https://gateway.pipeworx.io/dogceo/mcp"]
    }
  }
}

Technical Analysis

The MCP configuration invokes npx with both -y and the mutable package specifier mcp-remote@latest. This causes npm to download and execute whichever package release is currently identified by the latest distribution tag without interactive confirmation.

The effective code executed by this configuration can therefore change after the Skill has been reviewed. The configuration does not pin an audited version, verify a package integrity hash, or provide a lockfile. If the npm package, maintainer account, publication process, or dependency chain is compromised, malicious code can execute during MCP server startup.

Attack Path

  1. An attacker compromises the mcp-remote package, a transitive dependency, or an authorized publisher account.
  2. The attacker publishes a malicious release and assigns it the npm latest tag.
  3. A user or Agent starts the configured MCP server.
  4. npx -y downloads the malicious release without requesting confirmation.
  5. npm executes the package locally as the user running the Agent.
  6. The malicious package can access resources available to that operating-system account.

Impact Assessment

Successful exploitation provides arbitrary code execution with the privileges of the user running the MCP configuration. Depending on the host environment, the malicious dependency could read accessible files and environment variables, steal credentials, alter project data, invoke network services, or install user-level persistence. The configuration does not itself elevate privileges, so access remains bound ...[truncated 52 chars]

Remediation
View remediation

Remediation Suggestions

  • Replace mcp-remote@latest with an exact, audited package version.
  • Remove -y so unexpected installation or package-resolution changes require explicit approval.
  • Install dependencies through a committed lockfile and use reproducible installation commands such as npm ci.
  • Verify package provenance and integrity through registry signatures, integrity hashes, or an approved internal package mirror.
  • Review the selected package and its transitive dependencies before deployment.
  • Run the MCP process in a restricted environment with minimal filesystem access, sanitized environment variables, and constrained network permissions.
  • Establish an explicit dependency-update process that includes review and security testing before changing the pinned version.

other

Note
Location
SKILL.md:32
Finding

Tool Requests Are Disclosed to an External Pipeworx Gateway

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 32-34 and 43-50
Vulnerability Type: Third-party data disclosure
Risk Level: Low

Relevant Code:

bash
curl -s -X POST https://gateway.pipeworx.io/dogceo/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"breed_images","arguments":{"breed":"labrador","count":3}}}'
json
{
  "mcpServers": {
    "pipeworx-dogceo": {
      "command": "npx",
      "args": ["-y", "mcp-remote@latest", "https://gateway.pipeworx.io/dogceo/mcp"]
    }
  }
}

Technical Analysis

Both the documented API example and MCP configuration send tool calls to gateway.pipeworx.io, an externally operated gateway, rather than directly to the Dog CEO API. Consequently, the gateway can observe request metadata, tool names, arguments such as breed and count, source network information, and response traffic.

The external destination is visible in the documentation, but the Skill does not explain the associated data-processing boundary or caution users against placing secrets or unrelated sensitive information in tool arguments. Normal breed-image requests are not inherently sensitive, making this primarily a transparency and data-minimization concern.

Attack Path

  1. A user asks the Agent to invoke one of the documented dog-image tools.
  2. The Agent constructs an MCP request containing the tool name and arguments.
  3. The request is transmitted to gateway.pipeworx.io.
  4. The gateway can process or log the request, associated metadata, and returned content.
  5. If sensitive or unrelated information is accidentally included in an argument, that information is disclosed to the external service.

Impact Assessment

The external gateway can obtain the contents and metadata of requests sent to it. This does not provide local code-execution privileges or direct access to arbitrary host f ...[truncated 331 chars]

Remediation
View remediation

Remediation Suggestions

  • Clearly disclose that tool requests are processed by the Pipeworx gateway and document the applicable privacy and retention policies.
  • Explicitly prohibit credentials, personal data, confidential prompts, and unrelated sensitive information in tool arguments.
  • Validate and allowlist expected argument formats, such as known breed names and bounded numeric counts.
  • Minimize request contents and avoid forwarding unnecessary conversation context.
  • Prefer direct integration with the official Dog CEO API when the gateway provides no required security or functional benefit.
  • Apply appropriate TLS validation, endpoint allowlisting, request timeouts, and logging controls.
  • Review the gateway operator's data retention, access-control, subprocessors, and incident-response practices before use in sensitive environments.
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 (1)

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 37)May include surrounding context.

Example

bash
curl -s -X POST https://gateway.pipeworx.io/dogceo/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"breed_images","arguments":{"breed":"labrador","count":3}}}'