Back to skill

Security audit

PRD-Writer

Security checks for vulnerabilities and agentic risk

Overview

This PRD-writing skill is mostly coherent, but its generated HTML diagram attachments load unpinned third-party JavaScript and use permissive Mermaid rendering, so it should be reviewed before installation.

Install only if you are comfortable with generated HTML attachments fetching and running Mermaid JavaScript from a third-party CDN. Avoid opening or sharing generated HTML diagrams from untrusted requirement text until Mermaid is pinned or bundled, integrity checks are added, and stricter rendering settings are used.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (2)

T08 · Insecure Dependencies

Warning
Location
HTML模板/mermaid_template.html:7
Finding

Unpinned Remote JavaScript Dependency Without Integrity Verification

Content
View full analysis
``` ### Technical Analysis The generated HTML attachment loads executable JavaScript from a third-party CDN whenever a recipient opens it. The dependency uses the floating `@11` major-version selector rather than an exact immutable version and does not provide a Subresource Integrity hash. Consequently, the JavaScript executed by previously generated attachments can change after the Skill package has been reviewed. A compromised CDN, compromised upstream package release, or incompatible future release within major version 11 could cause recipients to execute code that was not present during the audit. This also conflicts with the documentation's claim that the Mermaid library is embedded and that the attachment can render without network access. ### Attack Path 1. The Skill generates an HTML diagram attachment containing the remote script reference. 2. An attacker compromises the relevant CDN or upstream package publication channel, or a later dependency release introduces malicious behavior. 3. A recipient opens the generated attachment while connected to the internet. 4. The browser retrieves the changed JavaScript because the URL is not pinned to an immutable artifact. 5. No integrity hash is available to reject the changed response. 6. The downloaded script executes in the attachment's browser context. ### Impact Assessment Successful exploitation permits attacker-controlled JavaScript to run in the security context assigned to the generated attachment. It could manipulate the document, alter displayed PRD content, initiate network requests, or access browser-exposed data available to that document. The issue does not itself grant operating-system priv ...[truncated 181 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
HTML模板/mermaid_template.html:168
Finding

Permissive Mermaid Rendering Configuration for Requirement-Derived Content

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • 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
Findings (11)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The HTML root declares lang="zh-CN", which fixes the document language to Simplified Chinese. Under the policy, a forced language/locale is a natural-language policy issue unless the skill offers opt-in or clearly documents that it is intentionally region-specific.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The title labels this as a Chinese-standard PRD template, and the document consistently presents the template only in Chinese. Under the policy criteria, forcing a specific language without user opt-in can be a natural-language policy violation when no alternative language option or justification is provided.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The title explicitly labels this as an "English Standard Version," which imposes a language choice in the skill content. There is no accompanying note offering alternative language selection or explaining a justified region-specific requirement, so this appears to violate the language/locale opt-in policy.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The trigger list includes very broad, everyday phrases such as '需求文档' and '做PRD/生成PRD', which can cause the skill to activate unintentionally in normal conversation. In an agent setting, over-broad activation can route unrelated user content into the skill, causing unintended data exposure, confusing task execution, or prompt/context hijacking through accidental invocation.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The trigger phrases include generic action terms such as '写PRD' / '生成PRD' that are ambiguous and could match many ordinary requests. In an agent environment, this broad matching increases the risk of overreach, where the skill hijacks requests that should instead receive a normal conversational response or a different workflow.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The trigger list includes broad phrases like generic PRD/product-document requests that are likely to overlap with normal user conversation. This can cause unintended skill activation, making the assistant follow this workflow when the user did not explicitly intend to invoke it, weakening predictability and consent around tool behavior.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The mode inference rules automatically select a mode when keywords are broad or absent, including defaulting to standard mode. This makes the skill's behavior boundary unclear and can silently enable more expansive questioning, structure generation, or review steps than the user expected.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The HTML example states the attachment works '无需网络' / without network, but it loads Mermaid from a remote jsDelivr CDN. This creates a supply-chain and reliability risk: rendering depends on network availability and trusts third-party JavaScript at open time, which could expose users to script compromise or broken offline behavior.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The language mode table specifies “中文模式” as the default when the user uses Chinese, which imposes a language selection automatically rather than offering an explicit choice first. The policy allows language constraints only when the skill offers user choice or documents a justified locale requirement; here, the default language behavior is automatic rather than opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

该技能说明整体以中文撰写,虽然提到“中英双语:中文和英文模板”,但未说明如何根据用户偏好选择输出语言,也未提供显式的语言选择步骤。若技能默认强制某一语言而无用户选择,可能构成语言/地区策略上的自然语言违规。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

该 Markdown 模板全文仅以中文提供说明与字段命名,没有说明用户可选择其他语言,也没有声明这是仅面向中文团队或区域合规场景的专用模板。按规则,未提供语言/区域选择的强制语言约束属于自然语言层面的策略问题。

Content

No source excerpt is available for this finding.