ClawHub
Back to skill

Security audit

Qwen Web Reviewer [GPT]

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed workflow for using Qwen Web as an external reviewer, with appropriate cautions around sensitive data, login, and file uploads.

Install only if you want an agent to send selected prompts or authorized excerpts to Qwen Web for review. Do not include confidential, regulated, proprietary, credential, or personal data unless you explicitly approve sending that material to Qwen, and be especially careful with file uploads.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Context Leakage

High
Category
Data Exfiltration
Confidence
85% confidence
Finding

Code or instructions that leak agent conversation context to external services, potentially exposing sensitive user interactions.

Content

Scanner excerpt · SKILL.md (reported line 82)May include surrounding context.

md
- After opening a new chat, verify that the composer is empty and the conversation header/URL changed. Re-select the newest suitable model and re-enable or verify Thinking because Qwen may reset these settings per chat.
- Never continue a previous conversation merely to preserve model settings. Context separation has priority over convenience.
- For a follow-up on the same subject, keep the same chat and reference the earlier answer explicitly. For a new subject, start a new chat and do not paste the old answer unless it is intentionally part of the new task.
- Record the chat label, model label, Thinking state, and round count in the final synthesis; do not expose session-bound URLs.

## Synthesis and reporting

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 25)May include surrounding context.

md
## Browser and session

1. Use the available browser-control skill before interacting with Qwen. Reuse the existing Qwen tab when it is live; do not create unnecessary tabs or reload an active conversation.
2. Verify that the visible page is Qwen Studio and that the account is signed in. A visible user profile, sidebar, model picker, and working composer are positive signs of authentication. Do not ask the user to log in or create a new tab when those signs are present; request manual login only when the current visible page actually shows `Войти`/`Sign in`, `Зарегистрироваться`/`Sign up`, a login form, CAPTCHA, bot check, or an equivalent blocking state.
3. Use the secure browser-auth capability for login or credential entry. Never request or enter a password, OTP, recovery code, or passkey through ordinary page automation. Passkeys are not supported in the Cloud Browser.
4. If Qwen requests CAPTCHA, 2FA, re-authentication, or a security decision, stop at the blocking step and offer the documented manual browser handoff. Do not solve or bypass a CAPTCHA without the required action-time confirmation, and do not claim that the session is persistent until a fresh Qwen page verifies it.