Back to skill

Security audit

Pixshop MCP — 28+ AI Creative Tools

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a disclosed Pixshop MCP integration, but it requests broad local Bash and file-read authority that is not needed for its remote image/video workflow.

Review this before installing. Use it only if you are comfortable sending prompts, image URLs, portraits, product images, and publishable media to Pixshop and any backend providers it uses. Avoid confidential or regulated content unless Pixshop's terms and privacy practices fit your needs, and prefer a version that removes unnecessary Bash and local file-read permissions.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:4
Finding
Unnecessary Shell and Local File Access Permissions## Vulnerability Details **File Location**: `SKILL.md`, line 4 **Vulnerability Type**: Excessive tool permissions violating the principle of least privilege **Risk Level**: Medium ```yaml allowed-tools: Bash, Read ``` ### Technical Analysis The skill declares access to both `Bash` and `Read`, although its documented purpose is to connect to the Pixshop remote MCP service and invoke image and video processing tools. The documented workflows do not require arbitrary shell command execution or unrestricted access to local files. Granting these capabilities expands the skill's security boundary beyond its stated functional requirements. If malicious instructions are introduced through compromised remote MCP responses, indirect prompt injection, or a future modification to the skill, the agent could be induced to read local files or execute shell commands. This finding concerns the excessive permissions declared by the skill; the audited file does not itself contain commands that actively exploit those permissions. ### Attack Path 1. A user loads the skill, giving the agent access to the declared `Bash` and `Read` tools. 2. The agent communicates with the external Pixshop MCP endpoint as directed by the skill. 3. A compromised service response, attacker-controlled content processed during the workflow, or a later malicious instruction attempts to influence the agent. 4. Because the skill has unnecessary local capabilities, the influenced agent may read local files or execute shell commands unrelated to image or video processing. 5. Data obtained through those capabilities could potentially be exposed in subsequent tool requests or used to affect the local environment. ### Impact Assessment Successful exploitation could provide access to files readable by the user running the agent and allow command execution with that user's operating-system privileges. The practical scope depends on the host application's authorization con ...[truncated 407 chars]
Remediation
## Remediation Suggestions - Remove `Bash` and unrestricted `Read` from the `allowed-tools` declaration. - Grant access only to the specific Pixshop MCP tools required for the documented image and video workflows. - If local configuration access is genuinely necessary, restrict file operations to the exact MCP configuration file and use narrowly scoped read or write capabilities. - Require explicit user confirmation before any local file access, shell execution, paid credit-consuming operation, biometric-image processing, or public publishing action. - Treat responses and resources returned by the remote MCP service as untrusted data and prevent them from authorizing additional tool calls. - Apply host-level sandboxing and outbound-data controls to limit the consequences of indirect prompt injection or remote-service compromise.
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)

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The skill instructs users to connect Claude/Cursor directly to a remote MCP endpoint at https://pixshop.app/api/mcp and extensively describes sending prompts, image URLs, and generated media to that service, but it does not clearly warn that user content will leave the local client and be processed by a third party. This can lead users to unknowingly transmit sensitive images, personal data, or proprietary prompts to an external provider, increasing privacy, compliance, and data-handling risk.

Static analysis

No suspicious patterns detected.