Security audit

Claw Werewolf

Security checks for vulnerabilities and agentic risk

Overview

This skill is a simple game/viewer skill that discloses bot registration and does not include executable code or hidden system access.

Before installing, understand that this skill is meant to register your bot for Werewolf matches through the web viewer and may keep local operational notes during heartbeat checks. Install it only if you are comfortable with your bot joining the lobby queue and with the partially Chinese-language experience.

A.I.G

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

SkillSpector

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
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

Findings (3)

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The skill description says the bot will '自动报名' and the installation flow encourages immediate install, but it does not present a clear up-front warning that enabling the skill can cause the user's bot to be enrolled into matches. This creates a consent and surprise-action problem: users may install a seemingly passive game/viewer skill without understanding that it can trigger participation or queued registration behavior.

Missing User Warnings

Low

Confidence: 88% confidence
Finding: The heartbeat instructions tell the agent to 'update any local notes about failures or delays,' which implies writing to local state or files, but the skill metadata/description does not warn users that local data may be modified. This is dangerous mainly from a transparency and consent perspective: operators may assume the skill is read-only except for viewing/monitoring, while it can persist data locally.

Natural-Language Policy Violations

Low

Confidence: 82% confidence
Finding: The primary user-facing description is written in Chinese and the action label later in the file is also Chinese, but the skill does not mention that it is Chinese-only or offer an English or locale choice. Under the policy, forcing a specific language without user opt-in can be a natural-language policy violation.

Static analysis

No suspicious patterns detected.