Back to skill

Security audit

image-to-3d

Security checks for vulnerabilities and agentic risk

Overview

The image-to-3D workflow is coherent, but mandatory troubleshooting uploads lack clear user control over what gets shared.

Make diagnostic reporting optional and review sanitized payloads before submission. Use images suitable for hosted processing and durable links, retain spending limits, and prefer reviewed package versions and workspace-scoped installation. The supplied artifact contains guidance only; the downloaded CLI and hosted service were not audited.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (11)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied chunk is solely a skill-installation command. It neither accepts an image nor performs signup, mesh generation, asset retrieval, or receipt reporting. Installation could support a larger workflow, but no such workflow is present here, and the installed skill's behavior cannot be established from this command alone.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The supplied chunk is a skill-update command, not an image-to-3D implementation. It contains no image submission, account provisioning, conversion request, or result retrieval. Updating could support a larger workflow, but no such workflow is present here, and the behavior of any downloaded replacement cannot be inferred from this chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The supplied code only installs another skill; it does not accept an image, perform conversion, register a service account, or return a model URL, job ID, or receipt. Installation could be a prerequisite for the described workflow, but it is the entirety of the supplied behavior, not an implementation of that workflow. The downloaded skill's behavior cannot be verified from this chunk.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y luxin-cli'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The first command executes luxin-cli@latest through npx with automatic confirmation, allowing a mutable package release to run with the agent's local permissions. This is a CLI rather than an MCP server, but the underlying supply-chain risk is real: a compromised release could access local files and credentials even though the requested operation is described as no-spend.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y luxin-cli'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The main conversion command executes the mutable luxin-cli@latest release rather than a reviewed version. The spending limit constrains the intended API operation, not the permissions of the downloaded executable, which could access local data if compromised.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The installation command invokes an unpinned skills package and installs skill content from a repository without an immutable revision. A compromised installer or repository update could introduce executable behavior or persistent adversarial instructions into the agent's globally available skills.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The canonical-skill installation also executes an unpinned skills CLI and retrieves mutable repository content. Its global scope can expose later agent sessions to compromised skill instructions, although installation itself is a legitimate setup activity.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The writable-home fallback repeats the unpinned installer and mutable-source installation. Redirecting HOME changes the installation location but does not sandbox the installer or prevent it from accessing other files available to the process.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y luxin-cli'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

Fetching guidance executes luxin-cli@latest, so reading documentation requires running mutable third-party code. The instruction to rely on current CLI guidance also introduces changing instructions that have not necessarily received the same review as this file.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The update command invokes an unpinned skills CLI and can replace previously reviewed skill instructions with newer content. Requiring an explicit refresh reduces silent-update risk, but the revision environment variable is not an integrity check or an immutable update target.

Content

No source excerpt is available for this finding.

Mandatory hosted feedback submission lacks a privacy warning or redaction guidance.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The fallback instruction requires hosted feedback containing the attempted command and expected and actual behavior, without requiring redaction or permission to disclose that information. These fields can contain private image URLs, local paths, task details, or sensitive error output; image-processing authorization does not necessarily authorize uploading troubleshooting context.

Content

No source excerpt is available for this finding.