文旅素材搜索

• Outcome: Review
• Latest audit: May 29, 2026 at 9:35 PM UTC
• Version:
  1.0.22

Overview

This skill openly searches and downloads tourism videos, but purchase-like chat commands can immediately run a shell download and write files without a separate confirmation.

Install only if you are comfortable with purchase/down-order chat commands immediately downloading remote media into local storage. Prefer adding an explicit confirmation step, keeping downloads constrained to a dedicated folder, and reviewing TRADE_API_BASE/WENLV_API_ORIGIN settings before use, especially before enabling any real transaction API.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (5)

Vague Triggers

Medium

• Confidence: 84% confidence
• Finding: The skill defines broad natural-language triggers for purchase and selection such as '购买', '下单', and related variants, and maps them directly to side-effectful actions. In conversational systems, ambiguous trigger phrases increase the risk of unintended purchases and automatic local downloads from ordinary or quoted user text.

Missing User Warnings

High

• Confidence: 97% confidence
• Finding: The skill description says that after purchase it will automatically download video to local storage and output the file path, but this side effect is not presented as a prominent warning or consent boundary. Silent local file writes are risky because users may think they are only browsing/searching while the agent performs persistent system actions.

Missing User Warnings

High

• Confidence: 98% confidence
• Finding: The purchase flow mandates shell execution to download remote content (`bash scripts/download_video.sh`) without a prominent safety warning or opt-in checkpoint. Executing shell-based downloads from conversation-driven triggers expands the attack surface to command execution, unsafe file writes, and abuse of network access.

Missing User Warnings

High

• Confidence: 98% confidence
• Finding: The skill explicitly forbids a confirmation step and requires immediate purchase/download execution upon matching a purchase instruction. Removing confirmation for an action that causes remote fetches and local file writes materially increases the chance of accidental or socially engineered execution.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The documentation explicitly instructs the agent to automatically download purchased video files to local storage after a mock transaction, without requiring explicit runtime user consent or clarifying the destination and overwrite behavior. Unprompted file writes create a real safety boundary crossing: a remote API response can cause local persistence of attacker-controlled content, increasing risk of disk abuse, unsafe file placement, or downstream handling of untrusted media.

VirusTotal

67/67 vendors flagged this skill as clean.