Security audit

Ai Content Polish

Security checks for vulnerabilities and agentic risk

Overview

The skill is transparent and non-executable, but its main purpose is to help hide AI authorship and evade detection, so it belongs in Review.

Review carefully before installing. Use only for legitimate readability, tone, and clarity editing where AI assistance disclosure rules are respected; do not use it to bypass AI detectors, watermark/provenance checks, academic integrity rules, or platform disclosure requirements.

A.I.G

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

SkillSpector

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (9)

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 97% confidence
Finding: The README explicitly frames the skill as making AI-written content look unlike AI-written content and references '反检测' and removing 'AI watermark' style signals, but provides no guardrails. This materially enables deception by helping users conceal AI authorship or bypass AI-detection expectations in contexts like schoolwork, reviews, marketing, or compliance-sensitive publishing.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The invocation examples are broad, generic user phrases that can cause the skill to activate in loosely related contexts without clear boundaries or exclusions. In this case, the skill’s stated purpose is to make AI-generated Chinese text appear more human and less detectable, so over-broad triggers increase the chance of the capability being applied to deceptive or policy-sensitive use cases such as academic dishonesty, spam, or evasion of AI-origin scrutiny.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The description specifies '中文 AI 内容去痕与润色工具', which constrains the skill to Chinese by default. Under the policy, a language or locale restriction should either offer explicit user opt-in/choice or be clearly documented as a justified region- or language-specific tool.
Content: No source excerpt is available for this finding.

Ssd 2

Medium

Category: Not specified by scanner
Confidence: 97% confidence
Finding: The skill explicitly advertises use cases such as '去AI痕迹', '反检测', 'AI watermark', and 'AI content detection', which amount to rewriting content to evade AI-origin detection. This is dangerous because it operationalizes deception: users can mask machine-generated text to bypass academic, platform, workplace, or editorial integrity controls.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 89% confidence
Finding: The activation criteria are broad enough to trigger on common writing and editing requests, which increases the chance the skill will be invoked outside its narrowly described purpose. In this case that matters because the skill is not just polishing text: it explicitly supports concealment of AI authorship, so over-broad routing expands exposure to misuse.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 84% confidence
Finding: The skill processes user-submitted text in detail, quotes passages back, scores it, and may substantially rewrite it, but it does not clearly warn users about that level of transformation. That creates a transparency and consent problem, especially for sensitive, proprietary, or personally identifying text pasted for review.
Content: No source excerpt is available for this finding.

Ssd 2

Medium

Category: Not specified by scanner
Confidence: 98% confidence
Finding: The tagline '让 AI 写的东西看起来不像 AI 写的' and the described core behavior directly instruct covert transformation of AI text to hide its origin. The explicitness of this framing makes the risk higher, because it signals intended misuse rather than incidental dual use.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 95% confidence
Finding: The title and the entire rule library are written exclusively in Chinese, indicating a fixed language/locale choice. There is no accompanying note that this skill is Chinese-only by design, no user language choice, and no justification for the locale constraint, which matches the policy-violation criterion for forced language without opt-in.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 88% confidence
Finding: The entire test case file is written only in Chinese and the expected outputs are defined exclusively in that language, with no indication that users can choose another language or that the skill is intentionally limited to a Chinese-language context. Under the policy, imposing a single language without opt-in or documented justification is a natural-language policy concern.
Content: No source excerpt is available for this finding.