Security audit

Toutiao Publish.Skill

Security checks for vulnerabilities and agentic risk

Overview

This skill is for real Toutiao account login and public article publishing, but it lacks clear user confirmation and safety boundaries for credentials, posting, local scripts, and third-party image generation.

Review before installing. Only use this skill if you are comfortable letting an agent operate a logged-in Toutiao account, handle your phone/password flow, send cover prompts to an external image service, download local image files, and publish live public content. Require a manual preview and explicit confirmation before any final publish click, and avoid sharing raw passwords through logs or chat when possible.

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
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands

Findings (11)

Tp4

High

Category: MCP Tool Poisoning
Confidence: 95% confidence
Finding: The declared description promises an end-to-end automatic publishing assistant for Toutiao Hao, including article generation, login, form filling, cover upload, and publishing. The supplied code only connects to a locally running browser via Puppeteer, finds or creates a page, and navigates to the Toutiao publish URL. This is only a small supporting step toward publishing and does not implement the core advertised capabilities. Therefore the description materially overstates what this code chunk actually does.
Content: No source excerpt is available for this finding.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 97% confidence
Finding: 声明描述的是一个完整的头条号文章自动生成并发布工具，但提供的代码片段实际只实现了“上传封面”这一子步骤，且还依赖用户或其他流程预先打开头条发布页并提供已连接的浏览器环境。虽然“上传封面”属于声明中的一项核心能力，但该代码的实际主功能与声明的整体目的相比明显更窄，不能代表所宣称的自动写作和发布助手。因此描述与代码行为存在实质性不匹配。
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 96% confidence
Finding: The skill instructs automated entry of a phone number and password into a third-party site without any warning about credential sensitivity, storage, or safe handling boundaries. In an agent environment, this can normalize collection and use of highly sensitive credentials and increase the risk of accidental logging, reuse, or exposure through tooling and transcripts.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 97% confidence
Finding: The skill describes publishing content to a live external platform, including the final confirmation step, without an explicit warning that this is a real external action with potentially irreversible consequences. In context, this is more dangerous because the skill is specifically designed to automate public posting, so an agent could publish unintended, unreviewed, or policy-violating content on the user's behalf.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 79% confidence
Finding: The description is written only in Chinese and targets the Chinese platform '今日头条', which may imply a fixed language/locale experience without documenting user choice or locale justification. This can conflict with language or locale flexibility expectations when no opt-in or scope limitation is stated.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 96% confidence
Finding: The skill is explicitly designed to generate content and publish it automatically, and later instructs cleanup/deletion of files, yet it does not require a user-facing confirmation before irreversible actions. In a publishing context this is especially dangerous because accidental or manipulated output can be posted to a real account and temporary artifacts can be removed before the user reviews what happened.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 93% confidence
Finding: The instructions tell the agent to handle account credentials and to perform remote downloads without any privacy or transmission warning. This creates risk of credential exposure, unauthorized retention, and silent transfer of user-related data to external services during login or cover-image generation.
Content: No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The rules instruct executing local Node.js and PowerShell commands to navigate and manipulate the environment, which is broader than simply publishing an article. Local command execution increases the attack surface because a content-focused skill can now run arbitrary scripts, interact with the filesystem, and potentially be repurposed for unintended actions.
Content: No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium

Category: Not specified by scanner
Confidence: 91% confidence
Finding: The skill directs the agent to call an external image-generation service and download files from it, which expands data flow beyond the declared publishing capability. This introduces third-party dependency, privacy, and supply-chain risk because prompts or derived content may be transmitted off-platform without explicit user consent or clear necessity.
Content: No source excerpt is available for this finding.

Missing User Warnings

Low

Category: Not specified by scanner
Confidence: 88% confidence
Finding: The skill directs the agent to download a generated image to local disk but does not disclose that it will create a local file or constrain where that file may be written. While lower impact than credential handling, silent local file creation can surprise users, consume storage, and create privacy or provenance issues if downloaded assets persist beyond the task.
Content: No source excerpt is available for this finding.

Vague Triggers

Low

Category: Not specified by scanner
Confidence: 90% confidence
Finding: The manifest description only says '自动发布今日头条' ('automatically publish to Toutiao') without any activation constraints, trigger phrases, or exclusions. In a manifest file, this lack of specificity can make the skill's intended invocation scope ambiguous and increase the chance of unintended activation.
Content: No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.