Back to skill

Security audit

Eric's Engineering Wisdom

Security checks for vulnerabilities and agentic risk

Overview

This is a Chinese-language engineering advice and routing skill with broad presentation guidance, but no executable code, external retrieval, credential use, or hidden destructive behavior was found.

Installers should expect this skill to answer mostly in Chinese, route broadly across engineering topics, and include Eric-style personal-experience framing. Avoid feeding it confidential engineering history unless you intentionally want that material summarized or organized.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (68)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The activation description is very broad, covering generic engineering questions, product development, diagnostics, and even ambiguous cases. In a skill-routing system, this can cause the skill to trigger too often, intercepting requests better handled by narrower or safer skills, leading to incorrect routing, policy bypass by scope confusion, or degraded answer quality.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The description and operational instructions are presented in Chinese and frame the skill behavior in that language, but there is no indication that the user can choose another language or locale. This can constitute a language-policy issue when a skill implicitly forces a specific language without opt-in.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

Describing this skill as the 'first-entry router' without precise boundaries creates ambiguity about when it should activate instead of other engineering-related skills. That ambiguity increases the chance of unintended takeover of conversations, especially in multi-skill environments where activation precedence affects which instructions and references are applied.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill description defines itself as a broad 'first entry' for engineering questions and includes vague triggers like engineering problems, product development, technical diagnosis, and ambiguous descriptions. This can cause the skill to activate for loosely related requests, leading to incorrect routing, unnecessary context loading, or overriding a more appropriate specialized skill.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The trigger table uses ambiguous keywords such as 'innovation', 'strategy', 'growth', and general engineering terms without boundary rules or disambiguation. In a routing skill, this increases the chance of false positives, accidental invocation on normal conversation, and loading unrelated subskills that can degrade safety and reliability.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The output template hardcodes Chinese section headers and structure without honoring the user's language, which can conflict with user intent and platform expectations. In practice, this can cause response integrity issues, reduce usability, and make the agent ignore explicit user language preferences.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill’s trigger description is broad enough to match many generic engineering-related requests, making accidental invocation likely. Over-broad routing can hijack user intent, cause inappropriate context loading, and increase the chance the agent answers outside its intended scope.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
83% confidence
Finding

The skill content is written as if Chinese is the required interaction language, without specifying language negotiation or honoring the user’s locale. This can lead to unsafe misunderstanding of technical guidance, especially when users ask in another language and receive mixed-language or untranslated engineering advice.

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The trigger keywords are broad and context-light, so ordinary engineering terms like '电机', '噪声', or '产品开发' may activate this orchestration skill even when the user's intent is narrower or belongs to another domain. In a routing skill, overbroad matching can cause unintended loading of additional domain skills, increasing the chance of prompt scope drift, irrelevant guidance, or exposure of internal skill behavior across contexts.

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The document instructs the parent skill to automatically include Eric's personal perspective in responses, creating a persona/voice override without explicit user opt-in. This can misrepresent subjective judgment as authoritative advice and may cause inconsistent language or persona behavior, especially when users expect neutral, domain-only guidance.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The file is written entirely in Chinese and says this layer is automatically loaded whenever the corresponding domain is involved. That creates a natural-language policy concern because it effectively forces a specific language context on users without any opt-in or alternative language choice being documented.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The routing protocol directs the skill to identify domains from keywords and load additional reference files without defining strong boundaries, exclusions, or validation steps. In an agent setting, broad activation logic can be exploited by adversarial or ambiguous prompts to trigger unintended skills, expand the instruction surface, and introduce irrelevant or conflicting guidance that affects downstream responses.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill declares very broad activation conditions such as being the 'first entry point' for engineering questions and triggering on vague descriptions of engineering problems. This can cause over-activation, routing the agent into this skill for loosely related conversations and allowing its strong output-formatting and routing instructions to override more appropriate behavior or user intent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The instruction says the output should always include the '⭐ Eric的工程判断' section, which forces a specific presentation style regardless of user preference. While not directly a security exploit, this reduces user control and can inject persona-driven authority cues that may make speculative advice seem more definitive than warranted.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill content is written entirely in Chinese and prescribes Chinese-language output patterns without offering a user-language check or opt-in. This can cause the agent to ignore or override the user's preferred language, degrading usability, transparency, and potentially the user's ability to safely understand technical guidance.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The activation scope is extremely broad: it claims to be the 'first entry point' for engineering problems, product development, technical diagnosis, fault analysis, and even vague requests. In an agent system, that can cause this router skill to trigger on many unrelated or only partially related prompts, overriding more appropriate skills and increasing the chance of incorrect, overconfident, or unsafe guidance in technical contexts.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The skill content is written as if responses should be produced in Chinese, without an explicit rule to preserve or ask for the user's preferred language. This can degrade usability, obscure safety-critical technical advice for users expecting another language, and create misunderstanding in engineering diagnostics where terminology precision matters.

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The trigger keywords are very broad generic engineering terms, so this routing skill may activate for loosely related queries and pull in unnecessary sub-skills. In a top-level router, over-broad activation increases the chance of misrouting, overcollection of context, and unintended propagation of instructions or data across domains, which can degrade safety and reliability even if no explicit malicious content is present.

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This markdown file presents all natural-language content in a single forced locale without offering any user choice or documenting that the skill is region- or language-specific. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The file states that this judgment layer is automatically carried whenever the engineering-expert skill is invoked, which broadens activation beyond a narrowly scoped, user-consented context. In a router skill that already serves as a first entry point for vague engineering questions, this can cause unintended loading and use of personal-experience guidance, increasing the chance of over-collection, misrouting, or hidden prompt influence across many requests.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The phrase inviting bulk ingestion of experience whenever the user says something like 'I’ll send you my experience' overlaps heavily with ordinary conversation and can be triggered accidentally or by prompt injection in adjacent content. Because this file describes persistent structuring and storage of supplied material, an overly broad trigger can lead to unauthorized ingestion of sensitive proprietary engineering knowledge, confidential project details, or poisoned memory entries.

Content

No source excerpt is available for this finding.