Security audit

ClawBrain Benchmark

Security checks for vulnerabilities and agentic risk

Overview

This benchmark skill is not malicious, but it asks for broad command execution and describes file, terminal, web, and messaging tests without clear safety limits.

Review this skill before installing. It appears to be a benchmark description rather than malware, but only run it in a disposable or clearly bounded environment unless you are comfortable with the agent potentially using terminal commands, editing files, fetching web content, and sending messages as part of tests.

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

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (3)

Vague Triggers

Medium

Confidence: 95% confidence
Finding: The trigger phrases are very generic (e.g. 'run benchmark' / 'test model performance') and can easily overlap with normal user requests, causing the skill to activate unexpectedly. In a command-dispatch skill wired to an exec tool, accidental invocation increases the chance of unintended command execution or benchmark workflows affecting the host environment.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The skill advertises behaviors involving file operations, web fetching, messaging, and terminal actions, but does not warn users that running benchmarks may read/write data, send communications, or modify system state. Because the skill is user-invocable and configured for command dispatch through exec, the lack of safety boundaries or impact disclosure materially raises the risk of privacy exposure, unwanted external actions, and local system changes.

Natural-Language Policy Violations

Low

Confidence: 91% confidence
Finding: Natural-language instructions and invocation text are presented only in Chinese, which can amount to a language policy issue if the skill is intended for general use. The file does not offer multilingual support, user opt-in, or a documented reason that the skill is China- or Chinese-language-specific.

Static analysis

No suspicious patterns detected.