Back to skill

Security audit

Pipeworx cataas

Security checks for vulnerabilities and agentic risk

Overview

This skill is a simple cat-image MCP integration with a disclosed remote endpoint and no local code, persistence, or credential handling.

Before installing, understand that requests for cat images and normal connection metadata will go to the disclosed third-party MCP endpoint. Avoid sending sensitive prompts or private data through this skill, but the inspected artifact is otherwise narrow and coherent.

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

Warning
Location
SKILL.md:18
Finding

Unverified Third-Party MCP Service Creates a Supply-Chain Trust Risk

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 18-23
Vulnerability Type: Unverified remote MCP dependency
Risk Level: Medium

json
{
  "mcpServers": {
    "cataas": {
      "url": "https://gateway.pipeworx.io/cataas/mcp"
    }
  }
}

Technical Analysis

The Skill delegates its functionality to an externally hosted MCP service at gateway.pipeworx.io. Although the Skill is presented as a CATAAS integration, the documentation does not establish that this gateway is operated or endorsed by CATAAS. The remote MCP implementation is neither included in the project for review nor pinned to an immutable version.

Consequently, the effective behavior of the Skill depends on a third-party component that can change after installation without corresponding changes to the audited file. The gateway operator—or an attacker who compromises the gateway—could observe requests, alter tool responses, or return misleading content.

The reviewed file does not contain local executable code, credentials, persistence mechanisms, or evidence that the current endpoint is malicious. This finding is therefore a supply-chain trust risk rather than confirmation of active compromise.

Attack Path

  1. A user installs or enables the MCP configuration from SKILL.md.
  2. The Agent sends random_cat, cat_by_tag, or list_tags requests to https://gateway.pipeworx.io/cataas/mcp.
  3. The third-party gateway receives the requests and associated connection metadata.
  4. If the gateway is malicious, compromised, or subsequently modified, it can manipulate MCP responses or record information included in requests.
  5. The Agent or user may trust and act upon the manipulated output because it originates from the configured service.

This project does not demonstrate remote code execution or local privilege escalation through the endpoint. Such effects would require additional behavior not present in the reviewed evidence.

Impact Assessment

The remote servi ...[truncated 477 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace the gateway with a verified first-party CATAAS MCP endpoint or a locally hosted and audited MCP implementation.
  2. Verify and document the ownership, operational responsibility, privacy policy, and security controls of gateway.pipeworx.io.
  3. Pin the remote implementation to a reviewed version or immutable artifact where the MCP deployment mechanism supports it.
  4. Apply an outbound allowlist so the Skill can communicate only with the explicitly approved endpoint.
  5. Avoid transmitting credentials, personal information, internal prompts, filesystem contents, or other sensitive data in requests.
  6. Validate and constrain remote responses before displaying them or using them in subsequent Agent actions.
  7. Monitor endpoint ownership, TLS certificate changes, availability, and unexpected response-schema changes.
  8. Clearly disclose to users that requests and network metadata are sent to a third-party service.
Vulnerability Patterns
  • 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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep