Back to skill

Security audit

Content Creator Bundle

Security checks for vulnerabilities and agentic risk

Overview

This bundle is not overtly malicious, but it needs Review because it can act on live social-media accounts and relies on unverified installation steps and weakly documented credential handling.

Review this skill before installing. Only use it with least-privilege platform credentials, store secrets outside committed config files, verify any installer and dependency versions before running them, and require explicit confirmation or dry-run review before scheduled posts, bulk publishing, likes, replies, or private-message actions.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (1)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:69
Finding

Unpinned Third-Party Components and Unverified Remote Installer Execution

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:69-80; README.md:7-13
Vulnerability Type: Unverified and unpinned third-party dependencies
Risk Level: Medium

The installation instructions retrieve multiple mutable third-party skills without specifying trusted versions, hashes, or signatures. The documented setup process also clones a remote repository and executes install.ps1, although that script is absent from the audited artifact and therefore could not be inspected.

Affected code in SKILL.md:69-80:

powershell
# 内容改写
py -m clawhub install humanizer
py -m clawhub install md2wechat
py -m clawhub install wechat-article-extractor

# 定时发布
py -m clawhub install cron

# 数据追踪
py -m clawhub install tavily-search

# 评论互动
py -m clawhub install wechat-management

Affected code in README.md:7-13:

powershell
# 克隆套装仓库
git clone https://github.com/clawhub/content-creator-bundle.git
cd content-creator-bundle

# 安装所有依赖技能
./install.ps1

Technical Analysis

Dependency installation commands resolve package names without immutable version constraints. Consequently, the code installed when a user follows these instructions can differ from the code originally reviewed. No package checksums, cryptographic signatures, trusted publisher identifiers, or lock file are supplied.

The README workflow has the same trust issue at the repository level: git clone retrieves the repository's current default branch, after which a PowerShell installer is executed. Because neither a reviewed commit hash nor integrity verification is specified, an upstream compromise or later malicious modification could change the effective installer payload. Moreover, install.ps1 is not included in the audited artifact, preventing validation of its commands and privileges.

Exploitation depends on compromise, replacement, or malicious publication of one of the named upstream components. The audited do ...[truncated 1749 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin every ClawHub dependency to an explicitly reviewed, immutable version. Avoid resolving packages solely by mutable names.
  2. Maintain a lock file containing exact versions and integrity hashes for all direct and transitive dependencies.
  3. Pin the Git repository to a reviewed commit hash or signed release tag rather than cloning and executing the current default branch.
  4. Publish cryptographic checksums or signatures for install.ps1 and verify them before execution.
  5. Include the installer in the distributed artifact so its behavior can be audited together with the documentation.
  6. Require signature verification and validate that packages originate from expected publisher identities.
  7. Display the installer contents and obtain explicit user confirmation before executing it.
  8. Run installation with the least-privileged account required and avoid administrator or root execution.
  9. Isolate dependency installation in a sandbox, container, or disposable environment before granting access to production platform credentials.
  10. Add automated dependency provenance, integrity, and vulnerability checks to the release process.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (7)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README instructs users to configure platform accounts and use publishing features but does not warn about secure credential storage, token handling, or the consequences of automated posting. In a multi-platform content-distribution skill, poor credential hygiene or unguarded automation could lead to account compromise, unauthorized posting, reputational damage, or policy violations across several services at once.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The invocation examples are presented as generic conversational phrases without clearly stating when the skill is active or what scope the commands apply to. Because this bundle supports rewriting, scheduling, publishing, and comment interaction across platforms, unclear activation boundaries can lead to unintended execution of sensitive actions.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The example trigger phrase '帮我回复评论' is broad, natural-language input that could plausibly appear in ordinary conversation, increasing the chance of accidental activation. In a skill that can perform user-impacting platform actions, ambiguous triggers raise the risk of unintended comment-management behavior without clear user intent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill advertises scheduled publishing and comment-management capabilities that can take actions on users' external platform accounts, but it does not clearly warn that these features may post, reply, or otherwise act automatically on the user's behalf. This can lead to unintended account activity, reputational harm, or platform-policy violations if users enable automation without understanding the scope of actions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The configuration examples instruct users to supply platform account identifiers and secrets such as WECHAT_APP_ID and WECHAT_APP_SECRET, but there is no accompanying guidance on secure secret handling. Without warnings, users may hardcode credentials, store them insecurely, commit them to version control, or share them in logs, increasing the risk of account compromise.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

All user-facing instructions and trigger examples are presented only in Chinese, which effectively imposes a language constraint without opt-in or explanation. Under the policy, a forced language or locale should either offer user choice or be explicitly justified as region-specific.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

The skill documentation, examples, platform naming, and interaction flows are entirely presented in Chinese, and no opt-in or alternative language behavior is offered. Under the language/locale policy, forcing a specific language without user choice can be a natural-language policy issue unless the regional constraint is explicitly documented and justified.

Content

No source excerpt is available for this finding.