Back to skill

Security audit

多平台违禁词检测

Security checks for vulnerabilities and agentic risk

Overview

This word checker needs review because its setup exposes API keys and its automatic recharge instructions can turn untrusted payment details into local commands.

Before installing, require approval for helper installation and each recharge, verified payment details, and payment arguments passed safely as data. Keep the API key scoped to this integration and never print it; rotate it if exposed. Submit confidential material only if you are authorized to share it with RedFox.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
references/alipay-recharge.md
Finding

Remote payment fields are interpolated into shell commands without safe argument handling

Content
View full analysis

Impact

An attacker-controlled payment response can cause arbitrary local command execution when the documented shell workflow is followed. This occurs during payment-link preparation, before the user completes payment.

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
Findings (11)

Tainted flow: 'api_url' from os.environ.get (line 127, 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/check_sensitive_words.py (reported line 157)May include surrounding context.

python
try:
        # 发起 HTTPS POST 请求
        response = requests.post(api_url, headers=headers, json=params, timeout=30)

        if response.status_code >= 400:
            raise Exception(f"HTTP请求失败: {response.status_code}, {response.text[:500]}")

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The text-checking behavior broadly matches the declared purpose, and the HTTPS API call is a supporting implementation detail rather than evidence of an unrelated purpose. However, several substantive advertised capabilities are absent from this code chunk, particularly multimodal input, replacement suggestions, and the automatic recharge-and-resume workflow. Printed instructions naming alipay-bot commands do not execute those commands. The code also provides no independent evidence that the backend uses an official database or implements the stated platform standards. These findings are limited to the supplied script; other skill components could potentially implement the missing workflow.

Content

No source excerpt is available for this finding.

A routine content check escalates into an automatic payment workflow when the service reports insufficient credits.

High
Category
Prompt Injection
Confidence
91% confidence
Finding

An insufficient-credit response escalates a content-checking request into payment submission using service-supplied payment details, without an explicit requirement here for transaction-specific user approval. This crosses a financial authorization boundary and could enable unwanted spending or payment redirection if those details are compromised; the referenced payment document is unavailable, so any downstream safeguards remain unverified.

Content

No source excerpt is available for this finding.

The API-backed content-checking workflow lacks a disclosure about third-party processing of user content.

Medium
Category
Output Handling
Confidence
86% confidence
Finding

The README identifies an external API provider but does not explain whether submitted text, uploaded files, extracted image text, or scraped webpage content is transmitted to that provider. Users checking unpublished marketing materials or confidential documents could unintentionally disclose sensitive information; the API-key security guidance does not address this separate content-privacy risk. The documentation supports a disclosure-gap finding, but does not establish actual transmission behavior or malicious exfiltration.

Content

No source excerpt is available for this finding.

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

Credential setup persists and prints the full API key without warning about exposure.

Medium
Category
Output Handling
Confidence
98% confidence
Finding

The instructions explicitly persist the API key in shell configuration or a user-wide environment variable and then print its full value for verification. This unnecessarily exposes a billable credential to terminal or agent logs and expands its availability to unrelated processes, potentially enabling unauthorized API usage and credit depletion.

Content

No source excerpt is available for this finding.

Automatically installs an external CLI without warning the user about local system changes.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

When the CLI is absent, the skill directs the agent to install external software without disclosing the local changes or obtaining explicit installation approval; npx -y also suppresses its package-fetch confirmation. Installing a payment helper may support the stated workflow, but a word-checking request does not itself authorize executing an installer or modifying the user's local environment.

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
98% confidence
Finding

The installation command uses the mutable @latest tag, allowing newly published package code to execute with the agent process's local permissions without version review. Although the scanner incorrectly describes this CLI as an MCP server, the supply-chain risk is real: a compromised or malicious release could access local files and credentials, and automatic installation during a payment workflow increases exposure.

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
95% confidence
Finding

This occurrence repeats the unpinned @latest installation command in failure-handling guidance, reinforcing the same supply-chain exposure as the earlier installation step. It is not a separate execution path by itself, but it references an unsafe installation procedure that can execute an unreviewed release with local user permissions.

Content

No source excerpt is available for this finding.

A word-checking failure is reframed as an instruction to initiate a payment workflow using third-party order details.

Medium
Category
Prompt Injection
Confidence
92% confidence
Finding

A third-party API response is promoted into imperative instructions for an agent to initiate a payment workflow using the returned URL, amount, and merchant information without validation or a confirmation gate in this file. Although the script does not itself transfer money, a malicious or compromised backend could exploit the advertised automatic-recharge workflow to induce unauthorized payment initiation or direct the user to an attacker-controlled payment destination.

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/check_sensitive_words.py (reported line 157)May include surrounding context.

python
try:
        # 发起 HTTPS POST 请求
        response = requests.post(api_url, headers=headers, json=params, timeout=30)

        if response.status_code >= 400:
            raise Exception(f"HTTP请求失败: {response.status_code}, {response.text[:500]}")