Back to skill

Security audit

Agentcad Skill

Security checks for vulnerabilities and agentic risk

Overview

ClawScan verdict: benign, with medium confidence; scanners report a mostly purpose-aligned CAD skill but flag an unpinned installer and additional installation and feedback steps, with some evidence incomplete.

The verdict is benign, not risk-free. Before installing, consider whether you need the extra feedback and diagnostic features, review any upload prompt and data it describes, and note that the installer version is not pinned. Skillspector's scan was partial because referenced paths were not present in the bundle, and the external skill was not inspected; the clean AIG result does not fill those gaps.

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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (4)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The only supplied command installs a skill; it does not execute build123d scripts or generate any declared outputs. Installation could support a larger CAD workflow, but no such implementation is present in this chunk. This assessment is limited to the supplied content, not the uninspected external skill.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
86% confidence
Finding

The supplied commands implement a feedback-reporting workflow rather than CAD modeling or artifact generation. This is a separate user-facing capability, not an implementation detail necessary to execute CAD scripts. The snippet does not establish network transmission or private-data access, so neither is inferred.

Content

No source excerpt is available for this finding.

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

Medium
Category
MCP Rug Pull
Confidence
82% confidence
Finding

The documented npx command can download and execute a mutable release of the skills package, introducing a supply-chain risk if a future release or its dependencies are compromised. This is an installation-time risk rather than an MCP-server configuration issue, and the supplied content does not establish that the package is currently malicious.

Content

No source excerpt is available for this finding.

External feedback and diagnostic uploads exceed the stated CAD capability.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The manifest describes executing CAD scripts and producing geometry, renders, exports, and metrics, but L336–L363 additionally expose automatic feedback transmission and diagnostic uploads to an external recipient. Diagnostics can include command arguments, results, and runtime information; although sharing requires human consent, transmitting tool telemetry is not necessary for the declared CAD purpose.

Content

No source excerpt is available for this finding.