Back to skill

Security audit

claw_poster

Security checks for vulnerabilities and agentic risk

Overview

The skill appears to generate posters as advertised, but it sends user design content to a temporary, informally managed external MCP service that users cannot clearly verify.

Install only if you trust the publisher and the specific MCP endpoint operator. Avoid sending confidential campaign plans, customer data, credentials, or sensitive brand material through this skill, and verify generated images and download links before using them publicly.

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:10
Finding

Unverified Remote MCP Service Hosted on a Temporary Tunnel

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:10-11; duplicated in v2.5/SKILL.md:10-11 and v2.5/SKILL 2.md:10-11
Vulnerability Type: Untrusted external service dependency
Risk Level: Medium

Vulnerable Code

text
MCP service address: https://syncopated-retractively-anitra.ngrok-free.dev/mcp
If the tool is unavailable or the domain expires, tell the user to contact “Xiao Zou” for a new address.

The displayed snippet is an English translation of the corresponding source instructions.

Technical Analysis

The skill requires the agent to use an MCP service exposed through an ngrok-free.dev tunnel. Such tunnel addresses are temporary and externally controlled, and the project does not provide server authentication, endpoint ownership verification, certificate pinning, response integrity checks, or a trusted mechanism for distributing replacement endpoints.

The skill directs the agent to submit template-search queries and poster slot values to this service. Depending on the design request, those values may include user-provided brand names, account names, dates, campaign details, or other potentially confidential content. The remote service also controls template specifications, generated images, quality-assurance results, and download links returned to the agent.

The instruction to obtain a replacement address informally from a named individual does not provide a verifiable chain of trust. A compromised operator account, malicious replacement address, compromised tunnel, or reassigned endpoint could therefore place an attacker in control of the MCP responses.

No evidence establishes that the current endpoint is malicious. The vulnerability is the absence of controls needed to establish and preserve trust in a mutable external service dependency.

Attack Path

  1. An attacker compromises the remote service, obtains control of a replacement endpoint, or convinces users to configure an attacker ...[truncated 1350 chars]
Remediation
View remediation

Remediation Suggestions

  1. Replace the temporary tunnel with a stable endpoint hosted on an organization-owned and access-controlled domain.
  2. Authenticate the MCP server using scoped credentials and require authenticated, encrypted communication.
  3. Publish approved endpoint changes through signed, version-controlled configuration rather than informal person-to-person replacement instructions.
  4. Validate the destination host and reject redirects or connections to endpoints outside an explicit allowlist.
  5. Define and disclose what user data is transmitted, how it is retained, and who can access it.
  6. Obtain user consent before transmitting potentially confidential content to the external service.
  7. Minimize submitted data and avoid sending secrets, credentials, personal data, or unnecessary account information.
  8. Validate returned download URLs, MIME types, file sizes, and image contents before presenting them as trusted output.
  9. Add endpoint monitoring, certificate-expiration monitoring, audit logging, and a documented incident-response process.
  10. Apply the same changes to all duplicated skill files so an insecure copy cannot remain active.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (7)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill description includes broad, everyday trigger phrases like '帮我做个促销海报' and '生成一张小红书封面', which can cause the agent to invoke this skill in many ordinary conversations without strong disambiguation. Because the skill then directs the agent to contact an external MCP endpoint, unintended invocation can leak user prompts or cause unreviewed external actions more often than necessary.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill is written to operate exclusively in Chinese and does not offer a user-language choice, which can cause misunderstanding of user intent, incorrect slot filling, or confusing follow-up questions for non-Chinese users. In a workflow that generates externally rendered content, language mismatch increases the chance of producing incorrect or misleading outputs and can reduce the user's ability to notice mistakes before generation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The changelog section is written entirely in Chinese, including operational guidance, with no indication that the skill supports multiple languages or that Chinese-only output is intentional and justified for a region-specific use case. This can violate language/locale policy by effectively forcing a specific language on users without opt-in.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill description is broad enough to trigger on many generic image or poster requests without clear scoping, which can cause the agent to invoke an external MCP service unexpectedly. Because the skill sends user prompts to a third-party endpoint, over-triggering increases the chance of unintended data disclosure and workflow hijacking, especially when users did not explicitly ask to use this service.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
81% confidence
Finding

Mandating a fixed language/locale behavior without user opt-in can cause the agent to ignore user preferences and produce outputs in an unintended language or regional format. In a design-generation skill, this can lead to wrong-language content being embedded into generated images and transmitted to the external service, creating privacy, usability, and trust issues.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill is triggered by very broad, common image-generation phrases such as making a poster, notice image, or cover, which creates a high risk of over-invocation outside the author's intended scope. In an agent environment, this can route many ordinary user requests to an external MCP service automatically, increasing unintended data disclosure and reducing user control over tool use.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The file is written entirely in Chinese and the trigger examples and operational guidance assume Chinese-language usage, but it does not state that language selection is optional or limited to a justified region-specific context. This can violate language/locale policy because it implicitly forces a specific language without user opt-in.

Content

No source excerpt is available for this finding.