Back to skill

Security audit

Toutiao Agent Skill

Security checks for vulnerabilities and agentic risk

Overview

This skill is for automated Toutiao posting, but it also imports package-authored memory and encourages automatic account actions without clear per-post approval controls.

Review carefully before installing. Only use this with a dedicated low-risk Toutiao account, inspect restore.sh first, avoid importing bundled memory into active agent memory unless you explicitly want it, pin installer versions where possible, and require manual review before any post is published or location is added.

Vulnerability Patterns
  • 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
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (3)

T03 · Remote Payload Retrieval and Execution

Error
Location
README.md:47
Finding

Unverified Remote Script Is Piped Directly into Bash

Content
View full analysis
Remediation
View remediation

T02 · Agent Memory Poisoning

Error
Location
restore.sh:22
Finding

Bundled Operational Directives Are Installed into Persistent Agent Memory

Content
View full analysis
/dev/null || mkdir -p "$WORKSPACE_DIR/memory" && cp -r "$SCRIPT_DIR/memory/"* "$WORKSPACE_DIR/memory/" ``` The copied `memory/2026-03-25.md` directs the Agent to read `knowledge/toutiao-growth.md` on every startup and records a decision to load and learn the knowledge base automatically whenever the Agent starts. ### Technical Analysis The restoration script copies package-controlled example records directly into `~/openclaw/workspace/memory`, which is shared persistent Agent state. These records are not merely historical data: they contain startup-oriented operational directives that tell later sessions to load additional package-controlled instructions. The package describes these files as examples, but it does not isolate them under an examples directory, ask for per-file consent, detect naming collisions, or distinguish package data from the user's own history. Consequently, package-authored instructions may be treated as trusted user memory in later sessions. This write is not required to provide static publishing templates or reference documentation. Installing package-controlled instructions into active long-term memory exceeds the minimum filesystem access needed for those functions. ### Attack Path 1. The user runs `bash restore.sh` as instructed. 2. The script copies every bundled file under `memory/` into the active OpenClaw memory directory. 3. A later Agent session scans or loads persistent memory. 4. The Agent encounters the package-authored startup directive and loads `knowledge/toutiao-growth.md`. 5. The additional bundled publishing, engagement, and automation strategies continue to influence future se ...[truncated 776 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
package.json:45
Finding

Installation Instructions Execute Unpinned Third-Party Packages

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (41)

Chaining Abuse

High
Category
Tool Misuse
Confidence
96% confidence
Finding

The use of '| bash' creates an unsafe command chain that executes untrusted remote content immediately. In this skill context, the README is a setup guide for a new environment, so users are especially likely to run commands verbatim, increasing the chance of full host compromise if the fetched script is malicious or tampered with.

Content

Scanner excerpt · README.md (reported line 49)May include surrounding context.

bash
# 安装 Node.js (v24+)
curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.39.0/install.sh | bash
nvm install 24

# 安装 OpenClaw

Rp1

Medium
Category
MCP Rug Pull
Confidence
91% confidence
Finding

The README instructs users to run npx clawhub@latest install toutiao-agent, which pulls and executes the latest version of a remote package without pinning to a known-good release. That creates a supply-chain risk: if the package is compromised or changed unexpectedly, users may execute attacker-controlled code during installation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The README promotes automatic posting, scheduled account activity, data collection, and browser cookie saving, but does not clearly warn users about account takeover, privacy leakage, platform policy violations, or the sensitivity of stored session data. In this skill context, those omissions are more dangerous because the tool is explicitly designed to automate a third-party content account and persist operational/account data over time.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

整个技能说明文档均以中文编写,且没有提供其他语言选项、用户语言偏好说明,或说明该语言限制为何是必要的。根据规则,未获用户选择即默认强制特定语言/locale,属于自然语言层面的组织政策风险。

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
89% confidence
Finding

The installation command uses npx clawhub@latest, which pulls and executes the latest published package version rather than a pinned, reviewed release. That creates a supply-chain risk: if the upstream package is compromised or changes behavior, users may execute unreviewed code during installation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill clearly describes scheduled auto-posting elsewhere, but the usage section does not present a prominent warning that it will publish content on the user's behalf at fixed times. This can lead to unintended actions from a logged-in browser session, causing unauthorized or surprising posts, account misuse, or platform-policy violations.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The entire skill is written as a fixed Chinese-language workflow and specifies Chinese content conventions such as required regional terms and tags, but it does not indicate that language/locale is user-selectable or that the skill is intentionally limited to a Chinese-only regional use case. Under SQP-3, forcing a specific language without user opt-in is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly instructs the agent to read recent memory files and then output a synthesized summary into chat. That creates a direct natural-language exfiltration path: sensitive user-provided notes, operational issues, account status, or token information stored in memory can be surfaced to any chat participant or downstream consumer without need-to-know filtering.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The workflow automatically scans knowledge, memory, and temp directories and generates a replacement summary file, but it does not disclose or gate that behavior with explicit user consent or warning. This is dangerous because memory files may contain sensitive operational notes, and silent overwriting of the prior summary can destroy audit context or propagate stale/incorrect synthesized guidance into downstream publishing.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
83% confidence
Finding

This markdown file imposes a single language throughout the skill content, which can constitute a language/locale policy violation when no user opt-in or documented regional justification is provided. There is no indication that users may choose another language or that the restriction is required for a region-specific compliance purpose.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The entire skill document is written in Chinese and provides prescriptive content guidance in Chinese only, with no indication that users may choose another language or locale. Under the stated policy, forcing a specific language without user opt-in can be a natural-language policy violation unless the locale restriction is clearly documented and justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
81% confidence
Finding

This file prescribes specific Chinese forms of address and phrasing throughout, but it does not indicate that the skill is limited to Chinese-language audiences or that users can opt into another language/locale. Under the policy, forcing a specific language or locale without opt-in is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill includes explicit automation steps to copy a local file into an upload directory and then use browser automation to publish content and upload media to an external platform, ending with a final publish click. There is no requirement for explicit user confirmation, preview, authorization boundary, or safety interlock before modifying third-party account content, which creates a real risk of unauthorized or unintended posting if the workflow is triggered automatically or with the wrong assets/account context.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The document explicitly encourages users to add location tags to posts ('发布时添加菜场定位') without any privacy or safety guidance. Repeatedly sharing precise, real-world locations tied to routines can enable stalking, profiling, or inference of the creator’s home/work area, especially in a content plan built around frequent posting from local markets.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

The file title and throughout-document instructions prescribe a China-specific, Chinese-language content strategy, including required use of labels like '上海' and '嘉定'. Under the policy, forcing a specific language or locale without opt-in or explicit justification is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The document explicitly plans automated publication to Toutiao and emphasizes growth objectives, but it does not require consent, review gates, data minimization, or safeguards against posting unintended user/content data. In a skill/workflow context, guidance that normalizes unattended publishing can directly lead to privacy leaks, accidental disclosure, or policy-violating posts if upstream content or account state is not carefully validated.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The entire skill file is written in Chinese and presents its operational content only in that language, without offering any language or locale choice. Under the stated policy, forcing a specific language without user opt-in can be a natural-language policy violation unless the locale restriction is clearly documented and justified.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The report stores identifiable usernames, follower counts, relationship status, and planned next steps for specific individuals without any privacy boundary, minimization, or consent context. Even if sourced from a public platform, aggregating and persisting this profile-style information in memory increases privacy risk, enables targeting, and may violate internal data-handling expectations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The comment interaction table reproduces user identifiers together with quoted comment content and engagement details, creating a reusable record of individual users' statements. This kind of persistence makes the data more sensitive than ephemeral display on a platform because it supports profiling, cross-referencing, and potential misuse beyond the original context.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The operational notes and user-directed instructions are entirely in Chinese, including the explicit instruction to manually publish on a specific Chinese platform. There is no indication that the skill offers a language/locale choice or that the Chinese-only constraint is explicitly justified as a region-specific requirement, which can violate language/locale policy expectations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The file explicitly documents scheduled and automatic publishing tasks, including a statement that the user has authorized automatic execution and guidance to assume the user is already logged in and publish directly. This creates a real risk of account actions being taken without fresh confirmation, visibility, or safeguards, which can lead to unintended posting, misuse of an authenticated session, or abuse if the workflow is triggered with modified content.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The install command uses npx clawhub@latest install toutiao-agent, which pulls and executes the latest remote package without pinning a specific version. This creates a supply-chain execution risk: if the upstream package is compromised or a malicious update is published, users may execute attacker-controlled code during installation.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The installation steps again instruct users to run npx clawhub@latest, which repeats the same unpinned remote-code execution risk. Duplicating the unsafe pattern in both command and steps increases exposure because users are more likely to execute it directly without scrutiny.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The manifest instructs users to configure account login information and describes automated publishing, but it does not provide clear warnings about credential handling, account risk, or the consequences of automated actions. In this context, the skill operates against a real third-party publishing account, so missing warnings materially increase the chance of unsafe credential exposure or unexpected account actions.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The usage section states that the skill will automatically publish content on a fixed daily schedule, but it does not clearly define activation boundaries, approval gates, or how automation is enabled or disabled. In a content-publishing skill tied to a real account, ambiguous autonomous behavior can lead to unintended posting, account misuse, policy violations, or reputational damage.

Content

No source excerpt is available for this finding.