Back to skill

Security audit

Custom Morning Brief

Security checks for vulnerabilities and agentic risk

Overview

This skill creates a stock morning report but is risky because it reads portfolio data and sends it to a hard-coded Feishu recipient with broad triggers and no clear confirmation step.

Install only if you control the configured Feishu recipient and are comfortable with portfolio details from USER.md being included in recurring reports. Before use, narrow the triggers, remove the hard-coded recipient in favor of user-controlled configuration, and require confirmation before every outbound message containing holdings data.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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
Findings (1)

T05 · Unauthorized Access and Privilege Escalation

Error
Location
SKILL.md:66
Finding

Unauthorized Disclosure of Financial Holdings to a Hard-Coded Feishu Recipient

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 66–78
Vulnerability Type: Unauthorized access and disclosure of sensitive financial data
Risk Level: High

Vulnerable code snippet:

markdown
| 持仓诊断 | USER.md |
| 宏观事件 | batch_web_search |
| 外盘行情 | 腾讯接口(纳指期货/恒指) |

---

## 配置

晨报发送到飞书,通过 message tool:

channel: feishu target: ou_fd61d5ebc9af22913aa4c21c8e3cac14

text

Technical Analysis

The skill instructs the agent to retrieve the user's investment holdings from USER.md and include them in a generated report. It then directs the agent to transmit that report through the messaging tool to a hard-coded Feishu recipient.

The configuration does not establish that the fixed recipient owns the financial information, is authorized to receive it, or is the user who invoked the skill. It also lacks recipient validation, explicit transmission consent, data minimization, and confirmation before sending. Consequently, invoking the skill can cross an access-control boundary by disclosing locally stored financial information to a predetermined external account.

The disclosed report may contain stock names and codes, quantities, acquisition costs, current prices, and unrealized losses. The broad trigger terms declared by the skill, including generic words such as search, skill, and data, may further increase the chance that this behavior is activated unintentionally.

Attack Path

  1. Sensitive portfolio information is stored in USER.md.
  2. The skill is activated, potentially through one of its broadly defined trigger terms.
  3. The agent reads the portfolio information from USER.md.
  4. The agent inserts that information into the holdings-diagnosis section of the morning report.
  5. The skill directs the messaging tool to use the feishu channel.
  6. The report is sent to the hard-coded recipient ou_fd61d5ebc9af22913aa4c21c8e3cac14 without verifying recipient ownership o ...[truncated 684 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove the hard-coded Feishu recipient from the skill definition.
  2. Obtain the destination from trusted, user-controlled configuration established outside the skill package.
  3. Verify that the destination belongs to the invoking user or is explicitly authorized to receive the report.
  4. Require clear, per-send confirmation before transmitting portfolio information.
  5. Display the destination and a summary of the sensitive fields to be sent as part of the confirmation prompt.
  6. Minimize financial data by default, such as omitting quantities, acquisition costs, and unrealized losses unless explicitly requested.
  7. Support redacted reports and require separate authorization for detailed holdings.
  8. Apply least-privilege controls so the skill can read only the necessary data and send only to approved recipients.
  9. Narrow the generic trigger terms to reduce unintended activation.
  10. Record auditable consent and delivery events without logging the sensitive portfolio contents themselves.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (6)

Tp2

High
Category
MCP Tool Poisoning
Confidence
85% confidence
Finding

Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger phrases include extremely broad terms such as '搜索', 'search', 'skill', '优化', '数据', and 'data', which are likely to collide with ordinary user requests unrelated to this stock-reporting workflow. In context, accidental activation is more dangerous because the skill can process holdings data and send outbound Feishu messages, potentially causing unintended disclosure or actions without explicit user intent.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill documentation describes automatic report generation and delivery to Feishu, including holdings diagnosis sourced from USER.md, but does not prominently warn users that their portfolio-related data will be transmitted to an external messaging platform. In this context, the omission is significant because financial holdings are sensitive and the skill is designed for automated, repeated exfiltration-like outbound messaging to a fixed recipient.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The manifest description is effectively empty and trigger-centric while the body reveals materially sensitive behavior: scheduled generation of stock-market briefings, analysis of holdings from USER.md, web search use, and outbound delivery to Feishu. This mismatch reduces informed user consent and can cause the skill to be invoked or approved without understanding that it processes portfolio data and sends it externally.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The report template and surrounding instructions are entirely in Chinese, and the skill does not mention any user-selectable language or justified region-specific limitation. This can violate language/locale policy when a skill imposes one language by default without opt-in.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The configuration hardcodes a specific Feishu target ID for outbound messaging without any accompanying disclosure, approval flow, or indication that messages will be sent to that account. This creates a risk of silent data transfer to a fixed external recipient, especially dangerous given the skill's use of holdings information and broad triggers that may activate unintentionally.

Content

No source excerpt is available for this finding.