Back to skill

Security audit

三角洲随机地图装备生成

Security checks for vulnerabilities and agentic risk

Overview

The skill is a small game randomizer, but it does not clearly disclose that each run contacts a custom third-party web service.

Review this skill before installing if you do not want a game utility to contact an undisclosed third-party service. The main risks are exposing connection metadata to that service, receiving manipulated randomizer content if the service changes or is compromised, and possible hangs because network calls have no timeout.

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 (2)

other

Note
Location
my_task.py:4
Finding

Undisclosed Third-Party Network Access

Content
View full analysis

Vulnerability Details

File Location: my_task.py:4-5, 10-11; network behavior is not disclosed in SKILL.md:13-43
Vulnerability Type: Undisclosed External Network Access
Risk Level: Low

Vulnerable Code

python
MAP_API = "https://sanjiaozhou.fushisanlang.cn/api/random_map_status"
ARM_API = "https://sanjiaozhou.fushisanlang.cn/api/random_arm_equipment"

# ...

map_data = requests.get(MAP_API).json()[0]
arm_data = requests.get(ARM_API).json()[0]

Technical Analysis

Each execution contacts two endpoints hosted on the custom third-party domain sanjiaozhou.fushisanlang.cn. The documentation describes the skill as a random map and equipment generator but does not disclose that its data is retrieved from an external service.

Although HTTPS protects data in transit, the remote operator receives ordinary connection metadata such as the user's public IP address and request timing. The service also controls the data incorporated into the skill's output. The code does not validate HTTP status codes, content types, response size, or response structure beyond accessing expected fields.

The retrieved content is only formatted as text; it is not evaluated or executed. Therefore, this does not constitute remote payload execution.

Attack Path

  1. A user installs and runs the skill based on its documented random-generation functionality.
  2. The skill silently sends two HTTPS requests to the custom third-party service.
  3. The service observes the user's connection metadata.
  4. If the service or its infrastructure is compromised, an attacker can return manipulated JSON values.
  5. Those values are incorporated into the generated result and presented to the user.

Impact Assessment

The issue exposes network metadata to an inadequately disclosed third party and creates an integrity dependency on that service. A compromised service can manipulate displayed map, equipment, or weapon-code data wit ...[truncated 188 chars]

Remediation
View remediation

Remediation Suggestions

  • Clearly document both external endpoints, the reason for contacting them, and the associated privacy implications.
  • Prefer a local, versioned dataset if remote access is unnecessary for the random-generation feature.
  • If remote access is retained, use an approved and documented service with an appropriate privacy policy.
  • Call response.raise_for_status() before parsing each response.
  • Validate the response content type, maximum size, JSON schema, and expected value types before incorporating data into output.
  • Consider restricting accepted response fields and lengths to reduce the effect of malicious or malformed server responses.

T09 · Insecure Skill Coding Practices

Warning
Location
my_task.py:10
Finding

Outbound HTTP Requests Lack Timeouts

Content
View full analysis

Vulnerability Details

File Location: my_task.py:10-11
Vulnerability Type: Unbounded Network Operations
Risk Level: Medium

Vulnerable Code

python
map_data = requests.get(MAP_API).json()[0]
arm_data = requests.get(ARM_API).json()[0]

Technical Analysis

Both requests.get() calls omit the timeout argument. The Requests library does not impose a default request timeout, so execution can remain blocked while waiting for a connection or response.

The broad exception handler does not mitigate this condition because it only runs after an exception is raised. A remote endpoint that accepts a connection and then sends data extremely slowly or never completes the response may prevent the operation from reaching that handler for an unbounded period.

Attack Path

  1. The user or an automated agent invokes my_task.py.
  2. The program connects to one of the configured remote endpoints.
  3. The endpoint becomes unavailable, accepts the connection without completing its response, or is compromised and deliberately stalls.
  4. Because no connection or read timeout is configured, the request remains blocked.
  5. The skill fails to return its JSON result, potentially tying up the invoking worker or workflow until an external process limit terminates it.

Impact Assessment

An unavailable or malicious remote service can cause denial of service against the skill invocation. The impact is limited to availability of the running process and any workflow waiting for its result. The flaw does not grant additional system privileges, access to local data, or arbitrary code execution.

Remediation
View remediation

Remediation Suggestions

  • Set explicit connection and read timeouts for every request, for example:

    python
    map_response = requests.get(MAP_API, timeout=(3, 10))
    map_response.raise_for_status()
    map_data = map_response.json()[0]
    
    arm_response = requests.get(ARM_API, timeout=(3, 10))
    arm_response.raise_for_status()
    arm_data = arm_response.json()[0]
    
  • Catch requests.Timeout, requests.ConnectionError, and requests.HTTPError separately to provide controlled error handling.

  • Apply narrowly bounded retries with exponential backoff only for transient failures.

  • Enforce an overall execution deadline at the caller or worker level as defense in depth.

  • Validate the response before indexing arrays or accessing nested fields.

Vulnerability Patterns
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • 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
Findings (2)

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
88% confidence
Finding

The skill declares a pip dependency on requests, which indicates network capability, but it does not declare any tool scope such as permissions or allowed-tools. This creates an unnecessary trust gap: a reviewer or runtime may not understand that the skill can make outbound network requests, which weakens sandboxing and informed consent even if the current description suggests only local random generation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

The natural-language content of the skill, including its description, headings, and usage instructions, is entirely in Chinese. Under the policy, forcing a specific language without user opt-in can be a locale-policy violation when no alternative language option or justification is provided.

Content

No source excerpt is available for this finding.