Back to skill

Security audit

picture-book-maker

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent picture-book workflow, but its packer can be abused through metadata to read local files, contact arbitrary URLs, or inject active content into generated books.

Install only if you trust the book metadata and image sources you will process. Avoid running the packer on untrusted metadata.json files, do not allow arbitrary remote image URLs, and treat generated book.html files as active web content until the script adds path containment, URL restrictions, strict image validation, and HTML escaping.

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 (3)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/pack-book.py:94
Finding

Arbitrary Local File Read and Base64 Disclosure Through Metadata Image Paths

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/pack-book.py:58
Finding

Server-Side Request Forgery Through Unrestricted Remote Image URLs

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/pack-book.py:234
Finding

Stored HTML and JavaScript Injection in Generated Books

Content
View full analysis
{text_cn}' f'
{text_en}
' ) elif text_cn: page_text = f'
{text_cn}
' elif text_en: page_text = f'
{text_en}
' ``` Cover metadata is also interpolated directly: ```python cover_html = f'''

{metadata.get('title', '绘本')}

作者:{metadata.get('author', '未知')}
{f'
{metadata.get("description", "")}
' if metadata.get('description') else ''} 开始阅读
''' ``` The resulting untrusted values are inserted into the final document: ```python
{page['text']}
``` The title is inserted into the document head without contextual escaping: ```python {metadata.get('title', '绘本')} ``` ### Technical Analysis The script treats metadata fields as trusted HTML. Values including `title`, `author`, `description`, `text`, `text_cn`, and `text_en` are inserted directly into the generated document without HTML escaping or sanitization. An attacker can close the surrounding element and inject arbitrary HTML. Depending on browser behavior and the injection context, payloads may include: - Ele ...[truncated 1948 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
Findings (10)

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill is presented as an end-to-end picture-book creation workflow, but the documented operational behavior materially differs: the only concrete implementation implied is packaging/rendering and optional remote image downloading. This mismatch can mislead users and calling systems about what the skill actually does, obscuring network access and file-processing behavior that carries security and privacy implications.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
93% confidence
Finding

The skill clearly describes capabilities that read local reference files, write output files, and fetch remote images, yet it declares no explicit tool scope or permission boundaries. This weakens least-privilege controls and increases the chance the agent can perform unintended file or network actions when the skill is invoked.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The trigger conditions are broad enough to match many ordinary creative-writing or illustration requests, which raises the risk of unintended invocation. In context, unintended invocation matters because this skill can create local files and later trigger packaging or network retrieval steps that the user may not have intended.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill instructs the agent to save files and generate packaged output but does not clearly warn that it will create and modify local directories and files. Lack of transparency around filesystem side effects can lead to unauthorized or surprising local changes, especially if the skill is triggered automatically or by vague prompts.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The packaging flow allows downloading remote image URLs without clearly warning about network activity, privacy exposure, or the trust boundary introduced by external content. This can leak environment metadata to third-party hosts and exposes the system to risks from untrusted remote resources, especially when URL inputs come from metadata.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The format requirements state that 'text_cn' is required while 'text_en' is optional, effectively forcing Chinese as the primary content language. This is a language-policy constraint presented without user opt-in, alternatives, or justification that the skill is region-specific.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

This markdown file is entirely presented in Chinese, starting with the title and continuing throughout the template, with no indication that language selection is optional. Under the stated policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale constraint is explicitly documented and justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The generated document hard-codes lang="zh-CN", which imposes a specific language/locale on all output. The file does not provide any user opt-in, configuration option, or documented region-specific justification for this constraint.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
98% confidence
Finding

L021 写明“本技能使用智能体原生能力,无需安装额外依赖包”,但文件头部 L004-L006 明确列出了 requests>=2.28.0 作为 Python 依赖。这属于文档与实际技能定义直接矛盾,而不是简单的信息缺失。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

This markdown file is entirely written in Chinese and presents the guidance as the default format, but it does not indicate that Chinese is optional or limited to a Chinese-language context. Under the policy rule for language or locale constraints, this can be considered a natural-language policy issue because it implicitly forces one language without user opt-in.

Content

No source excerpt is available for this finding.