Back to skill

Security audit

industrial-promo-video

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly coherent marketing automation guidance, but it directs customer PII to a fixed external Tencent Docs spreadsheet without deployment-specific validation or confirmation.

Review this skill before installing. It should only be used if your organization owns or has explicitly approved the listed Tencent Docs sheet and understands that customer phone numbers and messages may be appended there. Prefer replacing the hard-coded sheet identifiers with protected deployment configuration and require confirmation before the first PII write.

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

Warning
Location
SKILL.md:65
Finding

Hard-Coded External Spreadsheet Destination for Customer PII

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 65-70
Vulnerability Type: Hard-coded external data destination and insufficient authorization validation
Risk Level: Medium

Vulnerable Code Snippet

text
## 腾讯文档线索表对接(已建,勿重复创建)
- 文档地址:https://docs.qq.com/sheet/DTHhPaFBHSEV2VGNS
- file_id:`LxOhPGHEvTcR`,子表 sheet_id:`BB08J2`
- 13 列字段:线索ID / 留资时间 / 来源渠道 / 关联园区 / 客户昵称 / 客户留言 / 联系电话 / 意向等级 / 意向标签 / 跟进状态 / 跟进人 / 最近跟进时间 / 备注
- 追加新线索:定位第 0 列最后一个非空行 +1 作为新行,用 `set_range_value` 写入该行 13 个字段;表头为第 0 行,禁止覆盖
- 线索ID 规则:`L-YYYY-NNNN`(如 L-2026-0001)

Technical Analysis

The Skill embeds a fixed Tencent Docs URL, file identifier, and sheet identifier, then instructs the agent to append lead records containing customer nicknames, messages, and telephone numbers. These identifiers are configuration data tied to a specific external storage destination, but they are distributed directly in the reusable Skill definition.

The Skill describes the spreadsheet as authorized, but it does not define a runtime control that verifies the document owner, tenant, access policy, requesting organization, or intended destination before transmitting PII. It also does not require confirmation that the affected customer consented to this storage or that an applicable lawful basis and retention policy exist.

Consequently, installing or reusing the Skill in another environment may cause that environment's lead information to be written to the same globally fixed spreadsheet. The document identifiers do not necessarily grant access by themselves, but they direct any agent already possessing compatible Tencent Docs credentials toward a predetermined external data sink.

Attack Path

  1. The Skill is installed or invoked in an environment with access to the Tencent Docs integration.
  2. A user provides comments, private messages, or lead details for intent classification.
  3. The agent extracts fields such as the customer's nickna ...[truncated 1313 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove the Tencent Docs URL, file identifier, and sheet identifier from the reusable Skill definition.
  2. Supply the destination through deployment-specific protected configuration rather than agent-visible instructions.
  3. Require an administrator to configure and explicitly approve the destination before enabling lead writes.
  4. Validate the authenticated Tencent account, document ownership, tenant, sheet schema, and write permissions before transmitting data.
  5. Display the destination organization and document name for human confirmation before the first PII write and whenever configuration changes.
  6. Enforce least-privilege credentials restricted to the approved document and append-only operations where supported.
  7. Establish consent or lawful-basis checks, data-retention limits, deletion procedures, access logging, and periodic permission reviews.
  8. Minimize collected data. Avoid storing full customer messages or telephone numbers unless they are necessary and authorized, and redact unrelated sensitive content.
  9. Add tenant separation controls so different deployments cannot write to the same sheet unless deliberate shared storage is explicitly approved.
  10. Fail closed when destination ownership or authorization cannot be verified; do not fall back to the hard-coded document.
Vulnerability Patterns
  • 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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep