Back to skill

Security audit

Expense Report Wizard

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent expense-report assistant, but it can automatically process sensitive reimbursement data through connected external services without clear per-action consent.

Install only if you are comfortable using it with reimbursement data and connected Tencent/WeCom services. Prefer local-only processing unless you intentionally approve OCR uploads, online document export, meeting lookup, and notifications; use a dedicated folder containing only the invoices for this reimbursement; review any archive contents before processing; and avoid exposing Tencent Cloud secrets in chat, logs, screenshots, or shared shell history.

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
Findings (1)

other

Warning
Location
SKILL.md:52
Finding

Automatic External Processing and Notification of Sensitive Financial Data Without Per-Action Consent

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:52, SKILL.md:382, and SKILL.md:637
Vulnerability Type: Automatic external transmission of sensitive financial data
Risk Level: Medium

Relevant Skill Instructions

The following excerpts are faithful English translations of the relevant Skill instructions:

markdown
Line 52:
Optional enhancements are automatically enabled when connected, including batch
OCR, online document export, meeting-calendar verification, and compliance
notifications. These may connect to Tencent Docs, Tencent Cloud, Tencent Meeting,
and WeCom.

Line 382:
When a batch OCR service such as Tencent Docs is connected, automatically enable
it and invoke mcp__tencent-docs__ocr.extract for individual recognition or
mcp__tencent-docs__ocr.toexcel for batch recognition.

Line 637:
When a severe compliance issue exists, send a WeCom notification through
wecomcli-msg.

Technical Analysis

The Skill handles information that may include invoice images, company names, tax identifiers, employee names, travel routes, accommodation records, vendors, transaction amounts, and internal financial policies.

The instructions state that connected external integrations are enabled automatically. In particular, invoice data may be submitted to Tencent Docs OCR without requiring explicit approval immediately before transmission. Severe compliance findings may also be sent through WeCom without an instruction to preview the message, verify its recipients, minimize its contents, or obtain per-message confirmation.

Existing authorization for an integration does not necessarily constitute informed consent to transmit every subsequently supplied invoice or compliance result. The absence of a per-action confirmation boundary increases the chance of unintended disclosure, especially where an integration is preconfigured at an organization or agent level.

The finding is limited to the documen ...[truncated 2039 chars]

Remediation
View remediation

Remediation Suggestions

  1. Disable automatic external processing by default. Use local or built-in processing unless the user explicitly selects an external provider.
  2. Before each external OCR operation, display:
    • The service receiving the data
    • The files and fields that will be transmitted
    • The reason for transmission
    • Whether the provider retains or logs the submitted data
  3. Require explicit confirmation immediately before uploading invoice content. A previously configured integration should not be treated as blanket consent.
  4. Apply data minimization and redaction before transmission. Remove tax identifiers, personal names, invoice numbers, and unrelated image regions unless required for the requested task.
  5. Require confirmation before every WeCom notification. Present a preview of the exact message and identify the destination organization, channel, group, or recipient.
  6. Avoid including full invoice records in notifications. Send a minimal alert and keep detailed findings in the local report.
  7. Validate that the user is authorized to disclose the supplied corporate and employee data to the configured external services.
  8. Document provider retention, deletion, encryption, region, and access-control behavior.
  9. Record an auditable consent event containing the selected service, approved files, approved fields, destination, and timestamp.
  10. Provide a persistent local-only mode that prevents all OCR uploads, document exports, calendar access, and messaging calls.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (7)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The metadata description and trigger language indicate the skill may activate on very broad phrases like '帮我整理报销' without narrow boundaries. In a skill that can read local folders, parse receipts, and export data externally, over-broad activation raises the risk of unintended execution on ordinary conversations and accidental access to sensitive financial files.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill says it will '自动走完全流程' for common natural-language requests, but it does not define hard stop points before sensitive actions. Because later steps include local folder scanning, recursive file discovery, ZIP extraction, OCR, and external exports, auto-running the workflow could cause unintended processing of personal or corporate financial data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill promotes automatic enhancement via online documents, OCR, meeting schedules, and messaging integrations, but it does not clearly warn that reimbursement materials may contain invoices, names, travel records, company names, tax data, and other sensitive business information sent to third-party services. Users may unknowingly authorize disclosure of financial and personal data outside the local environment.

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 users to place Tencent Cloud secrets in environment variables for OCR integration but does not include any security guidance for handling those credentials. Exposing API secrets in shared shells, logs, screenshots, or misconfigured environments can lead to unauthorized use of cloud services and access to associated billing or data-processing capabilities.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The export workflow offers one-click export to Tencent Docs and related platforms without an explicit warning that full reimbursement content may be uploaded externally. Because expense reports can contain invoice images, names of colleagues, travel patterns, company identifiers, tax numbers, and compliance findings, silent or poorly explained export behavior creates a meaningful confidentiality risk.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The document states that the skill currently supports only China mainland invoice formats. This is a locale restriction stated as a hard limitation rather than a selectable or opt-in region mode, which can be a natural-language policy concern under language/locale constraints.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

The file-processing section describes recursive scanning of local folders and automatic ZIP extraction, but it does not prominently warn users that this can process many files beyond the immediate visible set. That increases the chance of accidental ingestion of unrelated or sensitive files, especially in mixed-use download folders or archives.

Content

No source excerpt is available for this finding.