Back to skill

Security audit

实时资讯观点博客写作助手

Security checks for vulnerabilities and agentic risk

Overview

This writing skill is coherent, but it tells agents to run unpinned npx tools and browse arbitrary URLs for screenshots without clear safety limits.

Install only if you are comfortable with the agent doing live web research, visiting source URLs in a browser, and writing screenshots locally. Prefer pinned, preinstalled tools, avoid logged-in/private pages, review screenshots before publishing, and restrict screenshot URLs to trusted public sources.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (3)

T08 · Insecure Dependencies

Error
Location
SKILL.md:48
Finding

Unpinned Package Execution Through npx

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 48
Vulnerability Type: Untrusted third-party package resolution and execution
Risk Level: High

Vulnerable Code

bash
npx skills search "<主题>"  # 如果有相关 skill

Technical Analysis

The workflow directs the agent to execute the unversioned skills package through npx. If the package is not already installed locally, npx can resolve it from the configured package registry, download it, and execute its CLI or package lifecycle code.

The skill does not specify an exact version, lockfile, integrity hash, trusted publisher, approved registry, or local-only execution requirement. The generic package name also increases exposure to package replacement, dependency confusion, typosquatting, and registry compromise.

Attack Path

  1. A user invokes the skill with a news topic.
  2. The agent follows the research workflow and executes npx skills search.
  3. npx finds no trusted local installation and resolves the package from the configured registry.
  4. A malicious, compromised, or unexpectedly resolved package is downloaded.
  5. Package lifecycle code or the package CLI executes with the permissions of the agent process.
  6. The package can access files, environment variables, network resources, and other assets available to that process.

Impact Assessment

Successful exploitation can result in arbitrary code execution under the operating-system account running the agent. The accessible scope may include project files, writable user files, process environment variables, configured credentials, and network resources available to the agent.

The instruction does not itself establish elevated privileges or persistence. Impact is bounded by the permissions and isolation controls applied to the agent process.

Remediation
View remediation

Remediation Suggestions

  • Remove runtime installation and execution of unspecified packages.
  • Add the dependency to a reviewed project manifest and pin an exact version.
  • Commit and enforce a lockfile with verified integrity metadata.
  • Obtain packages only from an approved registry and verify the publisher and package provenance.
  • Review the package and its transitive dependencies before approval.
  • Disable package lifecycle scripts where operationally practical.
  • Invoke only a preinstalled local binary, such as with npx --no-install.
  • Run the tool in a restricted environment with minimal filesystem, environment-variable, and network access.

T08 · Insecure Dependencies

Warning
Location
SKILL.md:188
Finding

Unpinned Playwright Execution Through npx

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 188
Vulnerability Type: Untrusted third-party package resolution and execution
Risk Level: Medium

Vulnerable Code

bash
npx playwright screenshot <URL> --full-page <output-path>

Technical Analysis

The screenshot workflow executes playwright through npx without an exact version or a local-only constraint. When Playwright is absent locally, npx may retrieve and execute package code from the configured registry.

Although Playwright is a known package name, the absence of version pinning, a lockfile, integrity verification, and an approved installation process leaves the command exposed to supply-chain compromise or unexpected package resolution.

Attack Path

  1. The skill reaches the screenshot stage.
  2. The agent executes the documented npx playwright command.
  3. npx resolves the package from a registry if a trusted local copy is unavailable.
  4. A compromised registry response, package release, dependency, or configuration supplies malicious code.
  5. The downloaded code executes with the agent process permissions.

Impact Assessment

Exploitation may provide arbitrary code execution within the agent account. A malicious package could read or alter project content, access available environment variables, make outbound network requests, or modify other files writable by the process.

No evidence shows that the skill deliberately seeks administrative privileges or establishes persistence. The practical scope depends on the sandbox and privileges assigned to the agent.

Remediation
View remediation

Remediation Suggestions

  • Install a reviewed Playwright release as a pinned project dependency.
  • Use a committed lockfile and verify package integrity and provenance.
  • Replace runtime package retrieval with invocation of a trusted local binary.
  • If npx remains necessary, use npx --no-install after controlled installation.
  • Use an approved package registry and prevent fallback to untrusted registries.
  • Run browser automation in a container or sandbox with minimal filesystem and network permissions.
  • Keep Playwright and its browser runtime updated through a controlled dependency-review process.

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:188
Finding

Unrestricted URL Navigation in Screenshot Workflow

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 188
Vulnerability Type: Server-side request forgery and unsafe browser navigation
Risk Level: Medium

Vulnerable Code

bash
npx playwright screenshot <URL> --full-page <output-path>

Technical Analysis

The skill accepts news links and instructs the agent to pass a URL directly to an automated browser. No validation requirements restrict URL schemes, destination hosts, resolved IP addresses, redirects, embedded credentials, localhost, private network ranges, or link-local cloud metadata services.

An attacker can therefore supply a crafted URL that causes the browser to request a resource reachable from the agent environment but unavailable to the attacker directly. Because the command captures the rendered response, sensitive internal content could be written to the generated screenshot directory.

Attack Path

  1. An attacker supplies a crafted news link or persuades the agent to use a malicious source URL.
  2. The skill reaches the screenshot stage and passes that URL to Playwright.
  3. The URL directly targets, or redirects to, a localhost service, private network host, link-local endpoint, or other restricted resource.
  4. The browser accesses the destination using the agent environment's network position.
  5. The returned content is rendered and saved as a screenshot under generated-images/.
  6. The screenshot may subsequently be included in the generated article output or otherwise exposed to the requesting user.

Impact Assessment

Exploitation may disclose information from internal web applications, local development services, administrative interfaces, or cloud metadata endpoints reachable by the browser. It may also allow interaction with internal HTTP endpoints that perform state-changing actions on request.

This finding does not demonstrate direct operating-system privilege escalation. Its scope is limited to r ...[truncated 113 chars]

Remediation
View remediation

Remediation Suggestions

  • Permit only explicitly supported schemes, preferably https.
  • Reject file:, data:, javascript:, and other non-HTTP navigation schemes.
  • Reject URLs containing embedded credentials.
  • Resolve the destination before navigation and block loopback, private, reserved, multicast, and link-local IP ranges for both IPv4 and IPv6.
  • Revalidate every redirect destination and protect against DNS rebinding.
  • Maintain an allowlist of approved public source domains where feasible.
  • Run browser automation in an isolated network namespace with no access to internal services or cloud metadata endpoints.
  • Use a clean browser context without ambient cookies, client certificates, or authenticated sessions.
  • Constrain screenshot output to a dedicated directory using sanitized, generated filenames.
  • Require explicit user approval before visiting an unknown or newly resolved host.
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 (5)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The description states the skill outputs 公众号/博客文章 and the entire instruction set is written as a Chinese-only workflow, implying a fixed language/locale behavior. There is no opt-in or alternative language choice for users, which violates the language/locale policy criterion.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The skill instructs use of npx skills search without pinning a specific package/server version, which makes execution depend on whatever package version is currently resolved at runtime. That creates a supply-chain risk: a compromised or breaking upstream release could execute unexpected code or alter behavior during research workflows.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill requires automatic browser screenshots and local file writes but provides no warning about capturing authenticated sessions, personal data, or sensitive on-screen information. In this context, screenshots of social media, dashboards, or personalized pages could inadvertently store tokens, usernames, recommendations, or other private content in generated artifacts.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The skill invokes npx playwright screenshot without version pinning, so the tool fetched and executed at runtime may change over time or be replaced by a malicious upstream artifact. Because this command drives a browser and accesses external URLs, an unexpected package version could increase exposure beyond ordinary content rendering.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The description states the skill is a '实时资讯观点博客写作助手' that outputs '公众号/博客文章', all in Chinese, which implies a fixed language/locale behavior. Under the policy rules, forcing a specific language without an explicit user choice or opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.