Back to skill

Security audit

HTML设计工具

Security checks for vulnerabilities and agentic risk

Overview

This HTML/CSS design skill is not malicious, but it asks for broad command, file, and API-related authority that is not clearly scoped to design work.

Install only if you are comfortable granting this skill file write access and potential shell-command authority. Prefer using it in a restricted workspace without secrets, and review any requested API keys or commands before allowing them.

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:16
Finding

Excessive System Command Execution Capability

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 16-19
Vulnerability Type: Excessive tool permissions and violation of least privilege
Risk Level: Medium

yaml
tools:
  - read
  - exec
  - write

Technical Analysis

The skill declares unrestricted access to the exec tool even though its stated primary purpose is generating and optimizing HTML/CSS designs. Normal design operations require reading and writing project files but do not inherently require general system-command execution.

The skill does not define a command allowlist, fixed command templates, argument validation, or a technically enforced sandbox boundary. Although the documentation claims commands run in a secure sandbox, no implementation or configuration in the reviewed project enforces that restriction.

Consequently, untrusted design requirements or file contents processed by the agent could influence it to invoke exec for operations unrelated to HTML/CSS design. The vulnerability is the unnecessary exposure of a privileged execution primitive rather than evidence that the skill itself currently executes a malicious command.

Attack Path

  1. A user or attacker supplies a design request, HTML document, CSS file, or other content containing adversarial instructions.
  2. The agent loads this skill with the declared exec capability available.
  3. The adversarial content persuades the agent to treat an unrelated system command as necessary for the design workflow.
  4. The agent invokes exec without an enforced command allowlist or validated argument policy.
  5. The command runs with the permissions and environmental access granted to the agent process.

Exploitation depends on the hosting agent honoring the declared tool permissions and failing to apply independent sandboxing or approval controls.

Impact Assessment

Successful exploitation could permit commands to execute with the privileges of the agent process ...[truncated 781 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove exec from the declared tool list because it is not necessary for the core HTML/CSS design workflow.
  2. Retain only narrowly required capabilities, such as scoped project-file reading and writing.
  3. If command execution is operationally necessary, replace general-purpose execution with a fixed wrapper exposing only explicitly approved commands and arguments.
  4. Reject shell metacharacters and never concatenate user-controlled content into command strings.
  5. Execute approved commands without an intermediary shell whenever possible.
  6. Enforce runtime sandboxing with a dedicated unprivileged account, a restricted writable directory, minimal environment variables, resource limits, and outbound network access disabled by default.
  7. Require explicit user approval before any command executes and display the exact executable, arguments, working directory, and expected side effects.
  8. Prevent secrets from being inherited by command subprocesses unless a specific approved operation requires them.
  9. Add security tests using adversarial HTML, CSS, and design prompts to verify that embedded instructions cannot trigger command execution.
  10. Do not recommend administrator or root execution as a generic response to permission failures; correct the minimum required filesystem permissions instead.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
Findings (7)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill lacks clear invocation triggers or activation boundaries, so a general-purpose agent may apply it in contexts broader than intended. When combined with read/write/exec tooling, ambiguous activation increases the chance of unnecessary high-privilege behavior being invoked for loosely related requests.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill introduces external API integration and API key configuration despite being described as a local HTML/CSS design tool. This expands the trust boundary to third-party services and creates opportunities for credential mishandling, unintended data egress, or deceptive collection of secrets not needed for the advertised task.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill is presented as an HTML/CSS design tool, but the documentation expands its scope to file processing, API integration, command execution, and information retrieval. This mismatch weakens user expectations and review boundaries, increasing the chance that a seemingly low-risk design skill is granted or uses higher-risk capabilities than necessary.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The documentation describes write, exec, and API behaviors without prominent user warnings about side effects, data exposure, or system impact. In an agent setting, insufficient disclosure can cause users or orchestrators to treat the skill as harmless design assistance while it performs privileged operations.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

Documenting command execution for a design-oriented skill introduces a materially risky capability without clear necessity. If the agent can run shell commands in response to design tasks, prompt injection or user confusion could lead to filesystem changes, data exposure, or execution of unintended programs.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
80% confidence
Finding

The visible display name, summary, and description are presented in Chinese, while no statement offers language selection or indicates that Chinese is optional. This can constitute a language policy issue when users are not given an opt-in or alternative locale.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
74% confidence
Finding

Line L052 frames the skill as unsuitable for complex human-judgment scenarios, suggesting a relatively narrow advisory/design role. Later sections describe it as performing operational actions such as command execution and external/API-driven processing, which materially broadens the implied operational intent and conflicts with the earlier constrained positioning.

Content

No source excerpt is available for this finding.