Back to skill

Security audit

Xiaohongshu Comic Creator

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed Xiaohongshu posting workflow, but it uses sensitive account cookies, live publishing, and mutable external installs without enough user-control safeguards.

Review this skill carefully before installing. Use pinned and trusted versions of the external skills, protect the Xiaohongshu cookies like a password, and require the agent to show the exact title, body, images, and target account for approval before any post is published.

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:54
Finding

Unpinned Executable and Mutable Skill Dependencies

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:54, 70-71, 91
Vulnerability Type: Unpinned third-party executable and Skill dependencies
Risk Level: Medium

Vulnerable Code

bash
npx clawhub@latest --workdir {your_agent_workspace_dir} install nano-banana-pro
bash
npx clawhub@latest --workdir {your_agent_workspace_dir} install tavily-search
npx clawhub@latest --workdir {your_agent_workspace_dir} install playwright-mcp
bash
npx clawhub@latest --workdir {your_agent_workspace_dir} install xiaohongshu-mcp

Technical Analysis

The installation instructions execute clawhub through the mutable @latest tag and install four third-party Skills by name without immutable versions, cryptographic digests, signatures, a lockfile, or another integrity-verification mechanism.

Because npx may retrieve and execute the current registry release, the code that runs can change after this Skill has been reviewed. The behavior of the subsequently installed Skills can likewise change independently of this repository. A compromised publisher account, package registry, release pipeline, or transitive dependency could therefore introduce arbitrary code into the Agent environment.

This is particularly sensitive because the installed components are expected to perform network searches, generate images using API credentials, control a browser, and publish through an authenticated Xiaohongshu integration.

Attack Path

  1. An attacker compromises the clawhub package, one of the named Skill distributions, its publisher account, release pipeline, registry entry, or a transitive dependency.
  2. The attacker publishes a modified release under the version selected by @latest or updates an unpinned named Skill.
  3. A user follows the documented installation command.
  4. npx downloads and executes the altered installer, or the installer retrieves the altered Skill.
  5. The malicious component executes with the permissions available to the Agen ...[truncated 1056 chars]
Remediation
View remediation

Remediation Suggestions

  1. Replace clawhub@latest with a reviewed, exact version.
  2. Pin every installed Skill to an immutable version or content digest.
  3. Verify package signatures or cryptographic checksums before execution.
  4. Commit and enforce a lockfile or equivalent dependency manifest.
  5. Maintain an allowlist of reviewed publishers, packages, versions, and hashes.
  6. Disable lifecycle scripts where feasible and inspect package contents before installation.
  7. Perform installation and execution in a sandbox with restricted filesystem and network access.
  8. Isolate credentials by component so search, image-generation, browser, and publishing tools receive only the secrets they require.
  9. Mount Xiaohongshu cookies read-only and expose them only to the reviewed publishing component.
  10. Monitor dependency provenance and require security review before version updates.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (16)

Rp1

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The skill instructs users to run npx clawhub@latest, which fetches and executes the latest published package version at runtime rather than a reviewed, pinned version. This creates a supply-chain risk: if the package or one of its transitive dependencies is compromised, the agent environment could execute attacker-controlled code during skill installation.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

This invocation again uses npx clawhub@latest, meaning users are told to execute an unpinned remote package with whatever code is current at install time. In a skill that also handles API keys and browser automation, compromise of the installer path could lead to credential theft or arbitrary code execution on the host.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

This is another unpinned npx clawhub@latest execution path, so the same supply-chain execution risk applies. Because the skill encourages installing multiple external MCP-related components, each unpinned install expands the attack surface and makes the workflow more dangerous in practice.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill describes automated posting to Xiaohongshu but does not warn users that it can act on a real external account, potentially publishing content, triggering moderation, violating platform rules, or causing reputational damage. In agentic workflows, lack of an explicit warning and confirmation step increases the likelihood of unintended account actions.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The install instruction for the Xiaohongshu MCP skill also relies on npx clawhub@latest, allowing unaudited code changes to be executed in environments that may contain login cookies and publishing capability. If abused, this could directly affect an external social media account as well as the local machine.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill asks users to export and store Xiaohongshu login cookies without emphasizing that cookies are highly sensitive authentication material equivalent to account access. If mishandled, leaked, or consumed by untrusted tooling, these cookies could enable account takeover, unauthorized posting, and exposure of private account data.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

Line L003 states 'All conversation use Chinese,' which imposes a specific language requirement. The document does not provide user opt-in, alternatives, or a clear region-specific justification for this constraint, so it conflicts with the language/locale policy guidance.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

Lines L033-L034 require 'Chinese language' and 'speech bubbles with actual Chinese text visible.' This is a locale-specific output mandate presented as unconditional policy, with no indication that users may choose another language or that the restriction is justified by a region-specific use case.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The template includes 'speech bubbles containing Chinese text' and 'Chinese language' as fixed prompt components. Because this instruction is embedded in the default template, it enforces a single language without opt-in or documented justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The section says each panel must include specific Chinese dialogue and provides Chinese-only dialogue placeholders. This is another unconditional locale restriction that lacks user choice or an explicit, justified regional limitation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

Lines L073-L075 mark Chinese text in every speech bubble as 'MANDATORY.' This is a direct language policy constraint that does not offer alternatives or explain a justified locale-specific requirement.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The 'Common Mistakes to Avoid' section marks English text in bubbles as incorrect whenever Chinese is required. In the context of this document's repeated unconditional Chinese-only directives, this reinforces a language policy violation rather than a user-configurable preference.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instructions require specific Chinese text for every panel and present this as mandatory output, which imposes a language choice without user opt-in. This is a natural-language policy concern because the workflow does not indicate that other languages are supported or that Chinese is optional based on user preference.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The workflow includes an automated publishing step via an external MCP tool without requiring an explicit user-facing warning or confirmation immediately before upload. In a skill that generates content and then publishes it, this can cause unintended disclosure or posting of AI-generated material, especially if the user did not realize execution would trigger a live external action.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The file instructs the agent to publish posts through a feed creation API but does not require any explicit user notification or confirmation before performing an external side effect. In a skill that automates content generation and posting, this creates a real risk of unintended publication to a third-party platform, which can cause reputational harm, accidental data disclosure, or unauthorized account activity.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The document requires titles to be under 20 Chinese characters and all examples and formatting guidance assume Chinese-language content. Because the file does not state that this is a region-specific or Chinese-only skill, nor offer a language/locale choice, it appears to force a specific locale without user opt-in.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.