Back to skill

Security audit

公众号文案创作

Security checks for vulnerabilities and agentic risk

Overview

The article-writing workflow is coherent, but automatic payment-software installation and insufficiently controlled recharge initiation warrant review.

Review the recharge workflow before installing. Require approval before installing payment software and before each recharge, verify the merchant and amount, and use a reviewed, fixed software version. Search keywords go to RedFox, so avoid confidential topics. The files do not establish credential theft or automatic charging.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (11)

Tainted flow: 'headers' from os.environ.get (line 170, credential/environment) → requests.post (network output)

Critical
Category
Data Flow
Confidence
90% confidence
Finding

Credentials or environment variables flow to a network sink. This is a high-confidence indicator of credential exfiltration.

Content

Scanner excerpt · scripts/fetch_gzh_trends.py (reported line 85)May include surrounding context.

python
print(f"Params: {json.dumps(params, ensure_ascii=False)}", file=sys.stderr)

    try:
        response = requests.post(base_url, headers=headers, json=params, timeout=60)

        if debug:
            print(f"状态码: {response.status_code}", file=sys.stderr)

A routine writing workflow escalates into automatic payment when the data provider reports insufficient credits.

High
Category
Prompt Injection
Confidence
96% confidence
Finding

A provider-controlled insufficient-credit response escalates a writing request into payment initiation and submission without an explicit transaction-specific approval requirement in this file. The referenced payment implementation is unavailable, so actual charging safeguards cannot be verified, but the visible workflow risks unauthorized payment attempts or charges if payment tools have standing authorization.

Content

No source excerpt is available for this finding.

A data-query response instructs the agent to initiate a payment workflow using API-supplied payment details, without requiring user approval.

High
Category
Prompt Injection
Confidence
99% confidence
Finding

On an insufficient-credit response, the script instructs the agent to invoke payment tools using API-supplied cashier URLs, amounts, and merchant descriptions without requiring explicit user approval. This crosses from a read-only article query into a financial workflow, allowing a compromised or malicious API response to steer payment initiation toward an unverified destination; the script does not itself prove that funds can be transferred without further authorization.

Content

No source excerpt is available for this finding.

Keyword searches use an external service without an explicit disclosure that user-supplied search terms are transmitted to that service.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The skill directs keyword-based searches backed by Redfox, whose external API and credentials are identified at lines 40–47. Although credential handling is documented, the README does not warn that user-provided keywords, which may contain confidential product names or topics, are sent to a third-party service.

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, network.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

The search workflow sends user-derived keywords to a third-party service without a privacy warning.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The workflow derives search keywords from the user's request and sends them to an external data provider without an explicit privacy notice or sensitive-data check. Although external search is central to the advertised functionality, keywords could disclose confidential product names, unpublished plans, or personal topics that users did not intend to share with that provider.

Content

No source excerpt is available for this finding.

The workflow automatically installs a downloaded CLI without warning the user about local system changes or package-code execution.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

The workflow directs the agent to download and execute an installer automatically when the CLI is absent, without requiring informed authorization for code execution or persistent local changes. The -y option also suppresses npx's package-install confirmation. A request to write articles or replenish credits does not inherently authorize installing software, and compromised installer code could access resources available to the agent.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @alipay/agent-payment'.

Medium
Category
MCP Rug Pull
Confidence
99% confidence
Finding

The command uses @latest rather than an immutable package version, allowing newly published package code to execute with the agent's local privileges. A compromised package release or dependency could expose local credentials or modify files; its role in a payment workflow increases the sensitivity. Although the scanner calls this an MCP server, the actual issue is execution of an unpinned npm CLI installer.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @alipay/agent-payment'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
80% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · scripts/fetch_gzh_trends.py (reported line 85)May include surrounding context.

python
print(f"Params: {json.dumps(params, ensure_ascii=False)}", file=sys.stderr)

    try:
        response = requests.post(base_url, headers=headers, json=params, timeout=60)

        if debug:
            print(f"状态码: {response.status_code}", file=sys.stderr)

Every successful query appends an unsolicited advertisement for purchasing a separate database.

Low
Category
Excessive Agency
Confidence
94% confidence
Finding

L352 unconditionally promotes a full-database procurement offering and directs users to a sales email address. The manifest describes article research, writing, and payment handling for insufficient query credits, not advertising separate data products; this promotion is unrelated to completing the requested query.

Content

No source excerpt is available for this finding.