Man vs Claw

• Outcome: Review
• Latest audit: May 28, 2026, 12:56 PM
• Version:
  1.0.0

Overview

This is a coherent chess-voting skill, but it asks agents to periodically trust and overwrite local skill instructions from remote web content without verification.

Install only if you want an agent to participate in this public chess game. Do not enable the heartbeat self-update flow unless you are comfortable with remote web content changing local skill instructions; review fetched files before replacing local ones, protect the API key, and use more specific invocation language to avoid accidental network calls.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (6)

Missing User Warnings

Medium

• Confidence: 94% confidence
• Finding: The skill instructs authenticated requests using an API key but provides no warning about protecting that credential, avoiding shell history leakage, or preventing accidental disclosure in logs and transcripts. In an agent-skill context, this increases the chance that secrets are stored, echoed, or forwarded unsafely during routine execution.

Missing User Warnings

Medium

• Confidence: 97% confidence
• Finding: The skill tells the agent to overwrite local skill files directly from remote URLs without integrity verification, confirmation, backup, or pinning to a trusted version. This creates a remote update channel where compromise of the server, DNS, or transport trust could replace local instructions with malicious content and persist that change on disk.

Vague Triggers

Medium

• Confidence: 95% confidence
• Finding: The trigger phrase "play chess" is very broad and overlaps with common user requests that may have nothing to do with this specific third-party skill. This can cause unintended activation and route ordinary chess-related conversations into an external skill, increasing the chance of unneeded network calls or user confusion.

Vague Triggers

Medium

• Confidence: 91% confidence
• Finding: The trigger "join game" is ambiguous and commonly used across many benign contexts, making reliable and safe skill invocation difficult. In a skill that connects to an external API, broad activation increases the risk of unintended engagement with the remote service and poor user consent boundaries.

Vague Triggers

Low

• Confidence: 87% confidence
• Finding: The phrase "check game state" is generic and may collide with ordinary status-check requests in other gaming or workflow contexts. While the impact is lower than broader gameplay triggers, it still creates a risk of accidental invocation of a network-connected skill when the user intended something else.

Missing User Warnings

Low

• Confidence: 96% confidence
• Finding: The skill explicitly recommends storing a long-lived API key in a predictable local file path without any guidance on file permissions, encryption, or secret-management practices. On multi-user systems, shared environments, backups, or developer workstations, this increases the chance of credential disclosure and unauthorized voting or account misuse.

VirusTotal

65/65 vendors flagged this skill as clean.