Back to skill

Security audit

Who Is Undercover Skill

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly implements the advertised game, but it also ships an under-disclosed external InStreet integration with a hard-coded bearer token.

Review this package before installing. The normal skill entry appears to run a local game, but the published files also contain a runnable remote InStreet controller with an exposed shared API token and contradictory documentation about network access. Install only if you accept that risk, and the publisher should remove or rotate the token and clearly separate or disclose the online integration.

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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
instreet_game_controller.js:147
Finding

Hard-Coded Bearer API Credential in Distributed Source Code

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
instreet_adapter.js:2
Finding

Undisclosed External Transmission of Authenticated Gameplay Data

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
  • 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 (29)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

声明描述强调这是一个完整的“谁是卧底”AI游戏技能,包含智能AI对手和完整机制;但代码实际只是一个 InStreet 平台的网络适配层,通过 HTTP 请求操作远程房间和游戏动作。它确实与“谁是卧底”场景相关(如创建 spy 房间、提交描述和投票),因此不是完全无关代码;但其主要行为依赖外部服务,且执行了未声明的联网能力。更重要的是,代码本身没有展示智能AI对手、身份分配、回合控制、胜负判断等“完整游戏机制”的实现,因此声明与实际行为存在实质不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The description suggests a general AI implementation of 谁是卧底 with complete mechanics and configurable 4–10 player support. The code instead acts as a controller for an external InStreet/spy-room service: it creates a specific 6-player room, polls activity, and auto-submits random descriptions and votes. It also performs undeclared file output and includes a hardcoded API key, which are sensitive/undeclared capabilities not implied by the description. While the theme matches the game genre, the actual code is materially narrower and different in purpose and capabilities than the declared description.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code chunk is not game functionality for 'Who is Undercover'; it is a development/publishing utility script. Its primary behavior is validating the structure of a skill package by reading directories and files, checking metadata format, computing total size, and flagging binary files. That is materially different from the declared purpose of running a social deduction game with AI opponents and full mechanics. This is a clear description-behavior mismatch.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description says this skill is a playable AI social deduction game. However, the supplied code chunk does not implement any game functionality, player handling, AI opponents, or game mechanics. Instead, it is an operational deployment script for publishing the skill to ClawHub and syncing it locally. This is a materially different primary purpose from the declared description, so it should be flagged as a mismatch.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

描述总体上与“谁是卧底”游戏相关,代码也确实属于该游戏的辅助逻辑;但该代码块的实际功能并非完整游戏机制本身,而是一个实时状态监控/落盘脚本。它会读取 current_game.json、写入 game_status.txt,并修改游戏状态文件,这属于未声明的资源访问能力。更重要的是,状态报告中会为所有存活玩家显示 role,并用 🕵️/👤 区分卧底与平民,这会直接暴露隐藏身份,与正常“谁是卧底”游戏机制不一致,属于描述未体现的敏感行为。因此应判定为存在描述与行为不匹配。

Content

No source excerpt is available for this finding.

Memory Manipulation

High
Category
Memory Poisoning
Confidence
80% confidence
Finding

Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.

Content

Scanner excerpt · index.js (reported line 376)May include surrounding context.

js
}
  
  currentGames.delete(gameId);
  delete context.gameId;
  
  return {
    success: true,

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The entire skill-related improvement plan is written in Chinese, and there is no indication that users may choose another language or that the skill is intentionally limited to a Chinese-speaking or region-specific context. Under the policy, language constraints should be opt-in or clearly justified when a specific locale is enforced.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill description is presented in Chinese and the usage examples imply Chinese-language gameplay content, but the README does not indicate that language is optional or that the skill is intentionally limited to a Chinese-speaking audience. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
83% confidence
Finding

Lines describing Feishu group chat support and platform integration imply interaction beyond isolated local computation. This actively conflicts with the later claim that the implementation has 'no external dependencies' and 'no network latency', creating intent/documentation divergence within the file itself.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
79% confidence
Finding

The documentation states '无外部依赖' and '本地计算,无网络延迟', which implies a self-contained local game implementation. However, elsewhere it advertises Feishu group chat support, ClawHub integration, InStreet community integration, and even a stored 'room_link.txt', indicating networked/community-facing behavior beyond a purely local implementation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The game’s core content is hard-coded in Chinese, including the word pairs used for gameplay, and the rest of the file continues this pattern in generated descriptions and messages. Because the file provides no language selection, opt-in, or justification that the skill is intended only for a Chinese-language audience, this is a natural-language locale policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The file generates gameplay descriptions in Chinese and also returns Chinese status text elsewhere, which means users are implicitly forced into a single language. This violates the language/locale policy criterion unless the skill offers opt-in or clearly documents a justified regional scope.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

This code file contains natural-language metadata and command descriptions entirely in Chinese, and later user-facing responses are also hard-coded in Chinese. The skill does not offer a language selection or indicate that the locale restriction is intentional and justified, which conflicts with the policy against forcing a specific language without user opt-in.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest frames the skill as providing the game itself with AI opponents and complete mechanics, but this adapter delegates core gameplay operations to an external InStreet API. That is a semantic mismatch at the skill-description level because the implemented behavior is primarily remote service integration rather than self-contained game logic.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The module performs outbound requests to a third-party domain using a bearer token, but the manifest description only presents a local/social game experience and does not clearly disclose this remote dependency. Undisclosed external connectivity increases supply-chain and privacy risk because room metadata, moves, and potentially user-generated reasoning are transmitted off-platform to a service the user may not expect.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This file’s natural-language strings, comments, and user-facing console output are entirely fixed to Chinese, which can impose a specific language on users without opt-in. The policy explicitly flags forced language or locale behavior when no user choice or justification is provided.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill advertises a local AI social-deduction game, but the controller creates and operates remote game rooms through an external service and embeds a live API key. This expands the trust boundary, sends gameplay and metadata to a third party, and can expose users to undisclosed network interaction, tracking, or misuse of the hard-coded credential.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The file’s natural-language comments and user-facing strings are entirely in Chinese, starting with the script description, and the script emits Chinese status/error messages throughout. This imposes a specific language/locale without any opt-in or justification, which matches the language-policy violation criteria.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

Although the code comment suggests it only lists alive players, the implementation also discloses whether each player is undercover or not. In a social deduction game, this is sensitive game-state information; exposing it nullifies the gameplay and enables trivial cheating by anyone with access to the report.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The status report exposes each alive player's secret role by rendering a role-specific emoji, which defeats the core hidden-information mechanic of the game. Any user or component that can read the generated report or status file can immediately identify the undercover player, compromising game integrity and potentially leaking secrets across participants.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

The file presents all user-facing instructions in Chinese and does not offer an English alternative or indicate that the skill is intended only for Chinese-speaking users. Under the language/locale policy rule, forcing a single language without user opt-in can be a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

This markdown file is written in Chinese and labels the package as a publication checklist without indicating any user or maintainer language choice. Under the stated policy, forcing a specific language without opt-in can be a natural-language policy violation unless the locale is explicitly justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

This markdown file contains substantial Chinese-language instructions and descriptions alongside some English headings and closing text, but does not state that the skill is intended only for Chinese-speaking users or offer a language/locale choice. Per the policy, forcing or assuming a language without opt-in can be a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The manifest description is written entirely in Chinese, which can impose a language-specific experience on users without indicating that other languages are supported. The policy requires either offering language/locale choice or clearly documenting and justifying the locale constraint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The script includes a human description string in Chinese, which makes the demonstrated interaction language-specific. Because the file does not indicate that the skill is intentionally Chinese-only or provide any user language choice, this can conflict with language/locale policy requirements.

Content

No source excerpt is available for this finding.