Back to skill

Security audit

蛋糕烘焙培训老师

Security checks for vulnerabilities and agentic risk

Overview

This is a cake-teaching skill with a disclosed report generator, but the report script can write to arbitrary paths and embeds report content into HTML without escaping it.

Review before installing. Use this only if you trust the report content and can ensure reports are written to a safe directory. Avoid feeding untrusted text into report fields, do not run it with elevated privileges, and treat generated HTML files as potentially active browser content until the script escapes HTML and restricts output paths.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/generate_report.py:42
Finding

Stored HTML and JavaScript Injection in Generated Reports

Content
View full analysis
{title} ``` ```python
🎂

{title}

{description or 'Professional cake baking instruction'}

{category}
``` ```python for group_name, items in recipe["ingredients"].items(): html += f"""

{group_name}

""" for item in items: name = item.get("name", "") amount = item.get("amount", "") html += f"""
{name} {amount}
""" ``` ```python for i, step in enumerate(steps, 1): step_time = step.get('time', '') step_alert = step.get('alert', '') html += f"""
Step {i}: {step.get(' ...[truncated 3913 chars]
Remediation
View remediation
` elements. - Event handlers such as ``. - Closing tags and malformed markup. - Quotes, ampersands, and angle brackets. - Payloads in every supported report field. 5. If reports are hosted, serve them from an isolated origin with no authenticated application state. Apply a restrictive Content Security Policy, for example one that disallows inline scripts and remote resources. 6. Remove the inline `onclick` handlers from the generated document so a strict Content Security Policy can be adopted without allowing `unsafe-inline`. ]]>

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/generate_report.py:581
Finding

Unrestricted Output Path Permits Arbitrary File Creation and Overwrite

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • 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 (7)

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill instructs the agent to invoke a local script that writes an HTML report, but the manifest does not declare any explicit tool scope or permission boundary. This creates a capability mismatch where file-writing behavior may occur without clear authorization, making unintended file creation or path misuse harder to govern and audit.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The description embeds many broad trigger phrases such as '蛋糕怎么做' and '烘焙入门', which can cause the skill to activate on generic conversation about cooking or learning. Over-broad invocation raises the risk of unintended routing into a skill that can write files and potentially use networked behavior, expanding exposure beyond deliberate user intent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The description states the skill is an "AI 蛋糕烘焙培训老师" and all instructional content is framed in Chinese, but the file does not offer a language choice or note that responses can follow the user's preferred language. This can violate language/locale policy when users invoke the skill with English triggers like "cake teacher" or "baking tutorial" yet are implicitly forced into Chinese output.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The manifest lists numerous ambiguous triggers, including short or generic phrases, without scope guards or disambiguation rules. This can lead to accidental invocation in unrelated contexts and unnecessarily expose the skill's file-generation and possible web-search workflows.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The workflow includes 'LLM 推理 + 联网搜索补充' for '其他' requests, which expands the skill beyond the stated cake-teaching function into open-ended web-connected behavior. Scope drift increases the chance of the agent handling unrelated prompts, retrieving untrusted content, or invoking broader capabilities than users would reasonably expect from the manifest.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill’s natural-language description and output are explicitly fixed to Chinese-language usage, and the generated HTML also declares a Chinese locale. There is no indication that users may choose another language or opt in to this locale, which conflicts with the policy against forcing a specific language without user choice.

Content

No source excerpt is available for this finding.

Overly Broad Trigger

Low
Category
Trigger Abuse
Confidence
86% confidence
Finding

The single-word trigger '裱花' is extremely short and context-poor, making accidental matching likely in broader conversations. Even though the domain is benign, unintended activation still matters because the skill can generate files and contains wider workflow behavior than a simple text-only assistant.

Content

No source excerpt is available for this finding.