Back to skill

Security audit

Frontend Design 3

Security checks for vulnerabilities and agentic risk

Overview

This design skill is not an active malware payload, but it asks for broad command and file access without enough limits for a frontend-design helper.

Review before installing. Use this only in a sandboxed workspace, require explicit approval for every command, do not run the agent as administrator, and avoid exposing sensitive API keys or private project data unless the host agent enforces tight file, network, and command limits.

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:15
Finding

Excessive Command-Execution Capability Violates Least Privilege

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 15–18
Vulnerability Type: Excessive tool permissions
Risk Level: Medium

Vulnerable Code

yaml
tools:
- read
- exec
- write

The document also states at line 356 that the Skill can execute system commands and collect their results. Troubleshooting guidance at lines 368 and 390 recommends running with administrator privileges when permissions are insufficient.

Technical Analysis

The Skill is intended to create frontend designs and generate HTML, CSS, JavaScript, React, or Vue source code. These activities ordinarily require constrained file-reading and file-writing capabilities, but not unrestricted system-command execution.

Declaring the exec tool gives the Skill access to a code-execution channel beyond its stated functional requirements. Although line 347 recommends command allowlisting and sandboxing, the package contains no enforceable policy, wrapper, or implementation that restricts commands, arguments, paths, network access, or shell interpolation.

The administrator-execution troubleshooting advice further conflicts with least-privilege principles. If followed, it could increase the effect of an unsafe command from workspace-level access to system-level modification.

No command that directly exploits this capability is embedded in the reviewed file. The vulnerability is therefore an excessive-permission and unsafe-configuration issue rather than evidence of an active malicious payload.

Attack Path

  1. An attacker supplies a hostile or manipulated frontend-design request to an Agent using this Skill.
  2. The request induces the Agent to treat a shell command as necessary for generating, testing, converting, or troubleshooting frontend assets.
  3. The Skill's declared exec permission allows the Agent to invoke the command.
  4. In the absence of an enforced command allowlist and argument validation, the command may access ...[truncated 1041 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove exec from the declared tool list and retain only the minimum read and write capabilities needed for frontend code generation.
  2. Scope file access to a designated project workspace and reject absolute paths, traversal sequences, symbolic-link escapes, and writes outside that workspace.
  3. If command execution is genuinely required for optional build or test operations, expose a dedicated wrapper instead of a general shell.
  4. Enforce a small allowlist of executable names and fixed argument patterns. Do not pass user-controlled strings through a shell interpreter.
  5. Disable network access, privilege escalation, interactive shells, subprocess chaining, and access to sensitive environment variables inside the execution sandbox.
  6. Run commands under an unprivileged account with filesystem, CPU, memory, and execution-time limits.
  7. Require explicit user approval before each command and display the exact executable, arguments, working directory, and affected files.
  8. Remove the recommendation to run as administrator. Replace it with guidance to correct workspace ownership or grant narrowly scoped project permissions.
  9. Add automated policy tests confirming that commands outside the allowlist, workspace escapes, shell metacharacters, and elevation attempts are rejected.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (5)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The invocation description is overly broad and mixes multiple creative and UI-related contexts without precise trigger conditions. In an agent environment, vague invocation criteria can cause the skill to activate unexpectedly and access read/write/exec tooling when a narrower, safer skill would have sufficed.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill is presented as a frontend design/code-generation tool, but later expands into generic API integration and system command execution. That mismatch can cause an agent or user to authorize far broader capabilities than expected, increasing the chance of unsafe exec usage, over-privileged invocation, and misuse outside the intended design domain.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The applicability section broadens the skill from frontend/UI design into generic development, operations, and content-creation tasks. This weakens security boundaries by encouraging invocation in contexts where exec, file, or network-related capabilities may be inappropriate or more dangerous than users expect.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The applicable scope is described so broadly that the skill could be routed into unrelated development, ops, or content workflows. Because the skill advertises read/write/exec capabilities, this broad scope materially raises the risk of unnecessary tool access and unintended actions.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

The display name, summary, and primary description are presented in Chinese, and the skill does not indicate that users may choose another language or locale. That can amount to a language policy issue when a skill implicitly constrains interaction language without opt-in or justification.

Content

No source excerpt is available for this finding.