MC Ecosystem Adaptation Engineer

• Outcome: Review
• Latest audit: Aug 6, 2026 · 9:52 AM UTC
• Version: 1.0.1

Overview

The skill mostly matches Minecraft mod management, but it also performs automatic IP-based location detection, stores local licensing/device state, and can make high-impact file changes that need careful review.

Install only if you are comfortable with automatic third-party IP geolocation, local licensing and usage-state files, network queries to mod services, downloaded executable JARs, and game-directory mutations. Prefer explicit language selection, offline mode where available, disabling automatic downloads unless needed, and manual review/backups before auto-fix or save restore.

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
• MCP Least Privilege: Underdeclared Capability, Wildcard Permission, Missing Permission Declaration
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection

Findings (43)

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 86% confidence
• Finding: The skill advertises capabilities that imply file, network, shell, and environment access, but the manifest does not declare any permissions or safety boundaries. This prevents informed consent and weakens agent-side policy enforcement, especially for features like downloads, backups, and command execution wrappers.

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 95% confidence
• Finding: The documented purpose focuses on Minecraft mod management, but the analyzed behavior includes license enforcement, machine fingerprinting, usage tracking, local auth state, online verification, and payment upsell flows that are not clearly disclosed. Hidden identity, tracking, and monetization logic materially expands the trust boundary and may expose users to privacy loss or unexpected system/network actions.

Description-Behavior Mismatch

High

• Confidence: 93% confidence
• Finding: This file implements a local payment/licensing system, including free-trial tracking, tiering, usage limits, and activation logic, which is unrelated to the stated Minecraft mod-management functionality. In an agent skill context, hidden monetization and authorization enforcement increases attack surface, can manipulate user behavior, and is especially suspicious because it is not necessary for core mod ecosystem operations.

Context-Inappropriate Capability

Medium

• Confidence: 95% confidence
• Finding: The code derives a persistent machine identifier from hardware/host attributes and also enables privileged 'author mode' via an environment variable. Device fingerprinting without clear necessity creates privacy risk and persistence, while environment-variable-based special mode can be abused by local users or wrappers to alter feature visibility and potentially bypass intended controls.

Context-Inappropriate Capability

Medium

• Confidence: 90% confidence
• Finding: The skill includes online license verification that contacts an external server and transmits local entitlement data, which is outside the declared mod-management feature set. Unnecessary network-capable code in a skill expands the trust boundary and creates risk of unexpected telemetry, tracking, or future abuse if the endpoint is enabled.

Description-Behavior Mismatch

Medium

• Confidence: 94% confidence
• Finding: The analyzer performs outbound Modrinth queries by default when local data is unavailable, which causes user-derived mod identifiers and environment metadata to leave the local system without an explicit opt-in. In the context of a crash/log analyzer, that is a privacy and trust-boundary issue because logs may reveal a user's installed modpack composition and runtime environment, and the network behavior is easy to miss.

Intent-Code Divergence

Medium

• Confidence: 88% confidence
• Finding: The module documentation presents the feature as local crash-report analysis and repair guidance, but the implementation also performs online mod-version discovery and generates download links. This mismatch undermines informed consent and can cause operators to run networked behavior they did not expect from the documented feature scope.

Context-Inappropriate Capability

Medium

• Confidence: 94% confidence
• Finding: The module makes outbound requests to third-party IP geolocation services and then persists country-derived location metadata to disk automatically. This creates a privacy and data-handling risk because users' network-derived information is disclosed to external services and retained locally without a clear trust boundary, consent flow, or minimization controls.

Description-Behavior Mismatch

Medium

• Confidence: 95% confidence
• Finding: This module implements payment-gating and donation/payment collection UI that is not necessary for a Minecraft mod ecosystem management skill. Even without remote code execution, undisclosed monetization logic expands the skill's behavior beyond its declared purpose and can coerce users into payment under misleading expectations.

Context-Inappropriate Capability

Medium

• Confidence: 97% confidence
• Finding: The code writes a payment-themed HTML file and automatically opens it in the user's browser via webbrowser.open(). Automatically launching browser content for monetization is an intrusive side effect unrelated to the core skill purpose and can be abused for phishing-style pressure, deceptive UX, or unexpected system interaction.

Context-Inappropriate Capability

Medium

• Confidence: 72% confidence
• Finding: The CLI exposes license/tier administration actions such as set_tier, reset_usage, and reset_free_period alongside normal end-user functionality. If this tool is distributed to untrusted users, these administrative paths can enable local privilege or quota bypass unless strong authorization is enforced in the underlying auth_manager implementation.

Missing User Warnings

Medium

• Confidence: 85% confidence
• Finding: The README advertises automated download, backup, replacement, and optional confirmation bypass for mod files, but it does not prominently warn users that this can change local game state and potentially break installations or introduce untrusted binaries. In a tool that modifies the user's mods directory, under-disclosing this risk increases the chance of unsafe use and accidental damage.

Missing User Warnings

Medium

• Confidence: 88% confidence
• Finding: The save synchronization feature sends Minecraft save data to a network storage provider, yet the README gives no privacy or security warning about transmitting potentially sensitive files to a third-party service. Save files can contain personal content, server details, or other user data, so omission of disclosure meaningfully increases privacy and data-handling risk.

Missing User Warnings

Medium

• Confidence: 80% confidence
• Finding: One-click upgrade and save restore features can modify or overwrite user data, yet the documentation does not warn about those risks or indicate confirmation/backup requirements. In a game-management skill, silent state-changing actions are security-relevant because they can cause integrity loss, unwanted downloads, or destructive restoration.

Natural-Language Policy Violations

High

• Confidence: 89% confidence
• Finding: Claiming IP-geolocation-based language auto-detection without clear user opt-in introduces a privacy-risking behavior that is unnecessary for the core mod-management purpose. Using network-derived location for locale selection can expose user metadata and normalize undisclosed tracking in a context where users would not reasonably expect it.

Missing User Warnings

Medium

• Confidence: 97% confidence
• Finding: The verification request sends both the license key and a machine-derived identifier to a remote service without any visible disclosure, consent, or just-in-time confirmation. This exposes sensitive identifiers for correlation and tracking, and users may have no reasonable expectation that local use of a Minecraft management skill triggers outbound identity-linked telemetry.

Missing User Warnings

Medium

• Confidence: 89% confidence
• Finding: The code forwards the auto_confirm flag directly into a workflow that downloads replacement JARs, writes them into the mods directory, and deletes the prior version, all without an execution-time user prompt. Even with backups, this is a destructive autonomous action path and increases the risk of unintended or unsafe package replacement if recommendations or download sources are wrong.

Missing User Warnings

Medium

• Confidence: 96% confidence
• Finding: The code sends mod information extracted from user logs to an external service during online lookup, but there is no visible per-run disclosure, confirmation, or prominent privacy notice at the point of use. Even if only mod IDs and version context are transmitted, that still exposes user environment details to a third party and may violate user expectations in a local diagnostic tool.

Missing User Warnings

Medium

• Confidence: 83% confidence
• Finding: The skill writes a runnable .bat file to disk automatically, without explicit user confirmation at the point of creation. In an agent context, silently materializing executable helper scripts increases the chance of social engineering, accidental execution, and unsafe trust in generated automation, even if this specific script mostly opens URLs and creates directories.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The code sends the user's apparent public IP-derived context to multiple external geolocation providers without any visible consent or warning. Even if the code does not explicitly transmit the IP value in parameters, the providers receive the user's request metadata and can infer location, which is a privacy-sensitive operation for an i18n helper.

Missing User Warnings

Low

• Confidence: 90% confidence
• Finding: The module writes location-derived data such as country code, country name, and timestamp to a persistent disk cache without user awareness. Persistent storage of inferred location increases privacy exposure on shared systems and can outlive the user's expectation of a transient language-detection feature.

Natural-Language Policy Violations

Medium

• Confidence: 92% confidence
• Finding: The module is designed to auto-detect language from IP geolocation first and falls back to system locale, with module import triggering initialization by default. In this skill context, that makes the behavior more concerning because a utility component performs potentially privacy-sensitive detection automatically rather than waiting for explicit user choice.

Missing User Warnings

Medium

• Confidence: 89% confidence
• Finding: The tool performs outbound network lookups based on dependency identifiers extracted from an untrusted JAR, without explicit user notice or opt-in. In this skill context, that can leak analysis activity, environment metadata such as IP/user agent/timing, and sensitive mod/dependency interests to third parties, which is especially concerning because the broader skill advertises geolocation/language behavior and operates on user-supplied artifacts.

Missing User Warnings

Medium

• Confidence: 88% confidence
• Finding: The module automatically downloads executable JAR files from remote sources, including transitive dependencies, with no explicit execution-time confirmation or trust prompt. In the context of a mod-management skill, this increases supply-chain risk because users may unknowingly persist unreviewed third-party code to disk and later load it into Minecraft or related tooling.

Vague Triggers

Low

• Confidence: 78% confidence
• Finding: The strings describe cloud save synchronization and automatic repair in broad terms without constraining when they should run or what safeguards apply. In the context of a mod-management skill that can alter files and synchronize saves, vague prompts can normalize high-impact actions and increase the chance of unintended destructive operations or unsafe agent behavior.

VirusTotal

VirusTotal findings are pending for this skill version.

Static analysis

No suspicious patterns detected.