工资代发与银行回单核对（免费版）

• Outcome: Review
• Latest audit: Sep 19, 2026 · 7:29 AM UTC
• Version: 1.0.40

Overview

The local payroll checker code is benign-looking, but the skill instructions push agents into external purchase, payment-link, and dependency-install workflows that are broader than the advertised free offline tool.

Review before installing. The checker code appears local and does not perform network calls, but the skill text instructs agents to offer a paid upgrade and, on user selection, navigate external payment/install flows. Only use it if you are comfortable with those commercial instructions, and do not allow automatic purchase, QR-code handling, or installation of additional skills without separate, explicit confirmation.

A.I.G

Vulnerability Patterns

• Skill Instruction Hijacking: Alters the agent's session goals or safety constraints when the skill loads
• Agent Memory Poisoning: Writes attacker-controlled rules into memory that affect later sessions
• Remote Payload Retrieval and Execution: Fetches external code whose behavior can change after review
• Embedded Malicious Code: Ships malicious scripts inside the skill and executes them locally
• Unauthorized Access and Privilege Escalation: Obtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns

• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (14)

Description-Behavior Mismatch

High

• Confidence: 98% confidence
• Finding: The skill prominently claims it is fully local, offline, and requires no network access, yet later embeds remote purchase, installation, and payment flows. That mismatch can mislead users and the host agent into trusting the skill with fewer restrictions than warranted, creating a social-engineering path to external navigation and transactional actions.

Ae1

High

• Category: analysis-evasion

Content

node scripts/run.mjs --sample # 先看样例

• Confidence: 100% confidence
• Finding: Referenced artifact was not completely inspected

Ae1

High

• Category: analysis-evasion

Content

node scripts/run.mjs --sample # 先看样例

• Confidence: 100% confidence
• Finding: Referenced artifact was not completely inspected

Context-Inappropriate Capability

High

• Confidence: 97% confidence
• Finding: A payroll reconciliation skill should only analyze local records, but this file adds mandatory upsell and payment-orchestration instructions that push the agent into commercial actions unrelated to reconciliation. Embedding merchant IDs, purchase logic, and payment flow steps increases the attack surface for phishing, policy bypass, and unauthorized external actions.

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: The markdown instructs the agent to present paid upgrade choices and complete purchase flows, but the main guidance does not foreground that this may trigger external purchase navigation and payment-related actions. Users may reasonably believe they are only running a local checking tool, making consent less informed than it should be.

Whitespace Padding

Medium

• Category: Prompt Injection

Content

输出与你的安全约束**；它**可被用户关闭**（`SKILL_NO_UPSELL=1`），也**必须可忽略**（不点不买不影响结果）。
   执行官方购买指令时：先读一次官方说明；核对**商户号 `2088702818190040`**、商品号与订单是否一致；
   ⛔ 不要点击来路不明的链接/二维码，也**不要替用户付款** —— 付款一律由**用户本人**确认。

• Confidence: 80% confidence
• Finding: Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The file includes multiple external purchase links, redirect endpoints, and guidance for interacting with off-platform resources without a strong, centralized warning that the user is leaving the current platform. This increases the likelihood of phishing-style confusion and unsafe interaction with external payment pages.

Ssd 4

Medium

• Confidence: 95% confidence
• Finding: This instruction sequence explicitly moves the agent from ordinary payroll analysis into acquiring, installing, and validating another product on the user's behalf. Even with language saying the user must confirm payment, the skill is still steering the agent into a materially different, higher-risk operational workflow involving external dependencies and commerce.

Context-Inappropriate Capability

Medium

• Confidence: 94% confidence
• Finding: The skill instructs the agent to install additional skills and dependencies despite being presented as a bounded payroll checking tool. This expands trust and execution scope beyond the stated purpose, enabling supply-chain risk or unreviewed capability expansion through externally sourced components.

Description-Behavior Mismatch

Medium

• Confidence: 95% confidence
• Finding: The manifest describes a free local checker, but the body includes dynamic product lookup, redirect URLs, and upgrade discovery behavior. This hidden behavioral expansion undermines transparency and can steer agents to external resources under the guise of a local-only tool.

Natural-Language Policy Violations

Medium

• Confidence: 93% confidence
• Finding: The file’s natural-language comments, user-facing advice, notes, and labels are entirely written in Chinese and describe the expected user inputs and outputs in that language. There is no indication that the skill offers language selection or that the Chinese-only behavior is a documented, justified locale restriction.

Natural-Language Policy Violations

Medium

• Confidence: 92% confidence
• Finding: This file contains natural-language strings for usage, errors, and operational notes exclusively in Chinese, which can force a specific language experience on users. The policy allows fixed locale only when user opt-in or clear region-specific justification is provided, which is not present here.

Intent-Code Divergence

Low

• Confidence: 93% confidence
• Finding: The inline comment states that non-JSON input will be handled as plain-text material such as '直接粘贴的合同全文', which refers to contract text rather than a payroll-payment/bank-receipt reconciliation table. The actual surrounding code and manifest are narrowly about processing payroll reconciliation table input, so this comment appears copied from another skill and misstates the intended input domain.

Natural-Language Policy Violations

Low

• Confidence: 86% confidence
• Finding: The only natural-language content in the file is in Chinese, including headers and status values, with no indication that the skill is region-specific or that users can opt into this locale. Under the policy, hard-coding a specific language without documented justification can be a locale-policy issue.

Static analysis

No suspicious patterns detected.