Back to skill

Security audit

Douyin Browser Publish

Security checks for vulnerabilities and agentic risk

Overview

This skill is for publishing videos to a logged-in Douyin account, but it can click the final publish button without an explicit user confirmation.

Review this skill carefully before installing. Use it only if you are comfortable letting an agent control a logged-in Douyin creator session, upload videos, fill metadata, and potentially publish publicly. Require a manual review and explicit confirmation before the final publish click, and check for the built-in '琪琪' branding, child-story descriptions, and fixed hashtags in generated metadata.

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
Findings (1)

T01 · Skill Instruction Hijacking

Warning
Location
scripts/prepare_publish.py:126
Finding

Fixed Branding and Promotional Metadata Injection

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:64-65, scripts/extract_publish_meta.py:51-53, and scripts/prepare_publish.py:126-141
Vulnerability Type: Fixed promotional content injected into authenticated publication workflows
Risk Level: Medium

Evidence

SKILL.md:64-65:

text
标题:`{故事名} 琪琪睡前故事`
描述:`儿童成长故事,适合3-8岁小朋友 #儿童故事 #睡前故事 #亲子时光`

scripts/extract_publish_meta.py:51-53:

python
title = f"{story_name} 琪琪睡前故事"
description = "儿童成长故事,适合3-8岁小朋友"
topics = ["#儿童故事", "#睡前故事", "#亲子时光"]

scripts/prepare_publish.py:126-141:

python
title = meta.get("title", f"适合3-8岁的睡前故事 | {story_name} | 琪琪的魔法故事屋")
description = meta.get("description", "儿童成长故事,适合3-8岁小朋友")
keywords = meta.get("keywords", "")
elif os.path.exists(story_file):
    with open(story_file, 'r', encoding='utf-8') as f:
        story = json.load(f)
        story_name = story.get("name", story_name)
        title = f"适合3-8岁的睡前故事 | {story_name} | 琪琪的魔法故事屋"
        description = story.get("text", "")[:100] or "儿童成长故事,适合3-8岁小朋友"
        keywords = extract_keywords(description)
else:
    title = f"适合3-8岁的睡前故事 | {story_name} | 琪琪的魔法故事屋"
    description = "儿童成长故事,适合3-8岁小朋友"
    keywords = extract_keywords(description)

Technical Analysis

The Skill is described as accepting a video path, title, and description, but its instructions and metadata-generation scripts contain fixed branding and promotional defaults. When project metadata is missing or incomplete, these values are automatically inserted into titles, descriptions, and tags.

Because the browser workflow ultimately clicks the publication button in an authenticated Douyin session, generated defaults can become public content without preserving the user's intended metadata. This constitutes output manipulation within a privileged publishing workflow. The concern is strongest where the fixed brand name is inserted into the title, rather th ...[truncated 1534 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove fixed brand names and promotional phrases from all default title and description templates.
  2. Require the title and description to come from explicit user input or trusted project configuration.
  3. Use neutral, clearly documented defaults only when the user expressly permits fallback metadata.
  4. Display the exact final title, description, tags, and video path before publication.
  5. Require explicit user confirmation immediately before clicking the publication button.
  6. Record whether each metadata field came from user input, a project file, or a fallback value.
  7. Reject incomplete metadata in unattended publication workflows instead of silently inserting branded content.
  8. Add tests verifying that absent metadata cannot introduce undeclared names, brands, hashtags, or promotional phrases.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (7)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

声明的核心用途是“自动完成上传发布”到抖音,但提供的代码块只是一个元数据提取脚本。它读取本地目录中的 JSON/Markdown/视频文件,生成 publish_meta.json,并打印提取结果。代码没有任何与抖音平台交互、浏览器控制、网络请求、页面自动化、上传文件或点击发布相关的逻辑。因此其主要行为与声明的主要用途存在实质性不一致。虽然该脚本可能是发布流程的辅助组件,但就这段代码本身而言,描述明显夸大了实际能力。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

该脚本的实际职责是“发布前准备/入队”,而不是“自动发布到抖音”。它会扫描项目目录中的 mp4、publish_metadata.json、story.json,提取或生成标题、描述、关键词和标签,并把结果写到 ~/.openclaw/workspace/publish-queue 下的 pending_*.json。整个代码中没有任何浏览器控制、网络请求、抖音平台交互、登录、上传文件或点击发布等行为。因此其主要目的与声明存在实质差异。虽然这些准备步骤可能是发布流程的支持环节,但声明描述的是完整自动发布能力,而代码仅完成本地预处理与排队,属于明显不匹配。

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The workflow instructs the agent to click the final '发布' button immediately, without a mandatory user confirmation or final review. In this context, that is dangerous because it performs an irreversible action on the user's authenticated Douyin account, potentially publishing incorrect, private, or policy-violating content due to automation error, prompt ambiguity, or UI misidentification.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The trigger phrases '发抖音' and '发布到抖音' are broad and can cause the skill to activate in situations where the user has not explicitly authorized a full publish workflow. Because the skill culminates in posting content to a live account, overbroad activation increases the chance of unintended publication from ambiguous natural-language requests.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This code file contains natural-language instructions and output entirely in Chinese, including the top-level docstring and CLI usage/help text. Under the policy, forcing a specific language without user opt-in or a documented justification is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The module docstring and CLI usage/output text are entirely in Chinese, and the script's generated titles/descriptions are also hard-coded in Chinese. This imposes a specific language on users without any opt-in, selection mechanism, or documented region-specific justification, which matches the locale policy violation criteria.

Content

No source excerpt is available for this finding.