Back to skill

Security audit

Feishu Crm Automation

Security checks for vulnerabilities and agentic risk

Overview

This Feishu CRM skill mostly matches its stated purpose, but it embeds a specific CRM table destination that could route customer data to the wrong place.

Review this skill before installing. Only use it if the embedded Feishu table belongs to your organization or has been replaced with your own approved table configuration, and confirm reads or writes before letting it process customer contact details or sales records.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:111
Finding

Hardcoded Feishu CRM Destination May Cause Unauthorized Disclosure of Customer Data

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 111–113
Vulnerability Type: Hardcoded external data destination
Risk Level: High

Vulnerable Code

markdown
**Table URL**: https://my.feishu.cn/base/KgoqbzS5ZaTwLRskzKxcVucvn0e
**app_token**: KgoqbzS5ZaTwLRskzKxcVucvn0e
**table_id**: tblzrBS1Gb73OhIS

The original table label is presented in English here for reporting consistency; the URL, token, and table ID are reproduced exactly.

Technical Analysis

The skill embeds a fixed Feishu Bitable URL, application token, and table ID. Elsewhere, it instructs the agent to create CRM records containing customer names, contact information, business requirements, assigned sales personnel, and customer status.

Although a Feishu Bitable application token is generally a resource identifier rather than an API authentication secret, embedding a fixed destination is unsafe because the repository does not establish that the table belongs to the invoking user or that the user has approved it. The skill also does not require destination ownership verification before a read or write operation.

Consequently, when a user does not specify a table, an agent may interpret this hardcoded configuration as the default destination and transmit sensitive customer information to a table controlled by the skill author, another organization, or another tenant. This behavior also conflicts with the skill's earlier guidance that users without a configured table should be assisted in creating one.

Attack Path

  1. An attacker publishes or distributes the skill with a Feishu table under the attacker's control embedded in SKILL.md.
  2. A user installs the skill and asks the agent to add a customer without specifying a destination table.
  3. The agent obtains the fixed app_token and table_id from the skill configuration.
  4. Using the user's available Feishu integration authorization, the agent invokes the record-crea ...[truncated 1321 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove the hardcoded Feishu table URL, application token, table ID, and associated field identifiers from the distributed skill.
  2. Require each user or administrator to supply a destination table through tenant-scoped, access-controlled configuration.
  3. When no table is configured, stop the operation and guide the user through selecting or creating a table rather than silently using a default resource.
  4. Before the first read or write, display the resolved workspace and table destination and request explicit user confirmation.
  5. Resolve table metadata from the user-approved URL and verify that the table belongs to the expected tenant or workspace.
  6. Apply least-privilege Feishu permissions, restricting the integration to only the approved table and necessary record operations.
  7. Avoid logging customer fields, destination tokens, or record contents unnecessarily, and redact sensitive contact information from diagnostic output.
  8. Add tests ensuring that record operations fail closed when destination configuration is absent, ambiguous, or does not match the approved tenant.
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (8)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The trigger description is broad enough to overlap with common business conversation topics such as customer management, follow-up, reports, and spreadsheet autofill. Over-broad activation can cause the skill to run in contexts the user did not clearly intend, which is especially risky here because the skill performs CRM reads and writes against customer data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill supports writing customer records and querying contact-related data but does not state that the user will be informed before personal or business data is read, created, or modified. In a CRM context, this increases the chance of silent handling of sensitive customer information, unauthorized updates, and privacy or compliance issues.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill documentation hard-codes a specific Feishu Base URL, app_token, and table_id, which creates a direct access path to a real CRM dataset rather than describing a generic integration pattern. If this skill is distributed or exposed to unauthorized users, those identifiers can enable unintended targeting of a production data store and increase the risk of data disclosure, modification, or abuse.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The workflow retrieves all CRM records to generate a report, but the document does not indicate any user notice, consent, or minimization of accessed data. In a CRM context, this increases the risk of over-collection and inadvertent exposure of customer information, especially if the report is triggered in a broad or shared environment.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The document says the weekly report should filter customers created this week, but the implemented report counts customers with a recent follow-up date instead. This semantic mismatch can mislead users and downstream automation, causing reports to include a broader set of customer records than expected and potentially exposing or acting on incorrect business data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
81% confidence
Finding

This markdown file describes fetching all customer records and returning overdue customer names and statuses, which involves processing potentially sensitive business/customer data. The description does not include any warning or disclosure about viewing, handling, or sharing customer information, even though SQP-2 applies to markdown files when user data or privacy may be affected.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

L133 规定“所有时间使用Asia/Shanghai时区”,这是明确的语言/区域设置约束,但文档未提供用户选择,也未说明该技能仅面向中国区或该时区约束的合规/业务理由。根据规则,未获用户选择的固定 locale/tz 策略属于自然语言层面的组织政策风险。

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

该 markdown 文件给出的用户触发语只有一个自然语言示例“生成本周客户周报”,但没有说明必须在何种上下文、面向哪类数据源或有哪些不应触发的近似表达。虽然短语有一定领域特征,但作为技能触发描述仍缺少边界约束,容易与普通工作交流中的周报请求发生重叠。

Content

No source excerpt is available for this finding.