Back to skill

Security audit

store-hopper

Security checks for vulnerabilities and agentic risk

Overview

The skill is broadly a travel-planning helper, but it needs Review because it can fetch arbitrary webpages through anti-bot and proxy services and sends location or URL data to several outside providers with limited user control.

Install only if you are comfortable with the skill sending your city, addresses, searches, and fetched URLs to external weather, search, map, and reader services. Avoid giving it private, signed, internal, localhost, or token-bearing URLs, and consider running it in a restricted environment with pinned dependencies and your own map API keys.

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 (4)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/fetch.py:209
Finding

Unrestricted URL Fetching Enables Server-Side Request Forgery

Content
View full analysis
dict: """ 使用 BeautifulSoup 抓取网页内容(基础方案) 带重试机制和随机UA """ max_retries = 2 last_error = None for attempt in range(max_retries): try: headers = get_random_headers() # 创建会话 session = requests.Session() session.headers.update(headers) # 先访问建立cookies session.get(url, timeout=10, allow_redirects=True) time.sleep(0.5) # 正式请求 resp = session.get( url, timeout=15, headers=headers, allow_redirects=True ) ``` ### Technical Analysis The URL supplied to `fetch.py` is used directly in two HTTP requests without validating its scheme, hostname, resolved IP address, port, or redirect destinations. Redirect following is explicitly enabled. An attacker can therefore provide URLs targeting: - Loopback services such as `http://127.0.0.1:PORT/` - Private network services in RFC 1918 address ranges - Link-local services - Cloud instance metadata endpoints - IPv6 loopback or private addresses - Public URLs that redirect to internal destinations The initial request made to establish cookies and the subsequent request are both exposed. Checking only the original hostname would also be insufficient because redirects and DNS rebinding could change the effective destination. The browser-based Camoufox path elsewhere in the same file also navigates to an unrestricted URL, but the cited direct-request path is independently sufficient to establish the vulnerability. ### Attack Path 1. An attacker causes a crafted URL to b ...[truncated 1092 chars]
Remediation
View remediation

other

Warning
Location
scripts/fetch.py:146
Finding

Target URLs Are Disclosed to Undocumented Third-Party Proxy Services

Content
View full analysis
dict | None: """ 使用代理服务获取网页内容(Jina/Markdown/Defuddle) 这些服务可以绕过大部分反爬机制 """ errors = [] for method in ["jina", "markdown", "defuddle"]: try: summary(f"尝试代理服务: {method}...") service_url = PROXY_SERVICES[method]["url"].format(url=url) result = subprocess.run( ["curl", "-s", "-L", service_url], capture_output=True, text=True, timeout=30 ) ``` ### Technical Analysis When the Camoufox extraction path is unavailable or unsuccessful, the script embeds the complete target URL into requests sent to Jina AI Reader, `markdown.new`, and `defuddle.md`. These services receive the target URL and fetch the corresponding resource on behalf of the Skill. The full URL can contain more than a public page address. Query strings, path components, fragments encoded into the service URL, signed links, document identifiers, account identifiers, and bearer-like access tokens may be exposed. The services can also associate the requested URL with the Skill runner's IP address and request timing. This behavior is not clearly disclosed in the Skill's operational description. The loop tries up to three separate providers, so a single target URL may be disclosed to multip ...[truncated 1274 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
SKILL.md:43
Finding

Dependencies and Browser Assets Are Installed Without Version or Integrity Pinning

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/route.py:129
Finding

Reusable Tencent Maps API Key Is Hardcoded in Distributed Source Code

Content
View full analysis
tuple: """ 腾讯地图 Web POI 搜索(免费,无需 API Key)。 利用 lbs.qq.com/getPoint 页面的内置 h5gw 搜索接口。 search API 需要 boundary 参数指定城市,有 city_hint 时精度极高。 无 city_hint 时降级到 suggestion API(支持混合关键词)。 """ _TENCENT_KEY = "NQQBZ-YDDK4-7G2UP-XCWS6-VMOB5-S5BN3" headers = { "User-Agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) " "AppleWebKit/537.36 (KHTML, like Gecko) " "Chrome/120.0.0.0 Safari/537.36", "Referer": "https://lbs.qq.com/getPoint/", } try: # 优先用 search API(需要 city_hint 分离到 boundary) if city_hint: keyword = address.replace(city_hint, "").strip() or address resp = requests.get( "https://h5gw.map.qq.com/ws/place/v1/search", params={ "boundary": f"region({city_hint},0)", "keyword": keyword, "apptag": "lbsplace_search", "key": _TENCENT_KEY, "output": "json", "page_size": "1", }, headers=headers, timeout=(3, 5), ) resp.raise_for_status() data = resp.json() if data.get("status") == 0 and data.get("data"): loc = data["data"][0]["location"] return float(loc["lat"]), float(loc["lng"]) # 降级到 suggestion API(支持混合关键词,无需分离城市) resp = requests.get( "https://h5gw.map.qq.com/ws/place/v1/suggestion", params={ "keyword": address, "region": city_hint or "", "region_fix": "1" if city_hint else "0", "key": _TENCENT_KEY, ``` ### Technical ...[truncated 1668 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
Findings (27)

Tainted flow: 'url' from os.environ.get (line 67, credential/environment) → requests.get (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/route.py (reported line 50)May include surrounding context.

python
return None, None
    url = f"https://restapi.amap.com/v3/geocode/geo?address={quote(address)}&key={key}"
    try:
        resp = requests.get(url, timeout=(2, 5))
        resp.raise_for_status()
        data = resp.json()
        if data.get("status") == "1" and data.get("geocodes"):

Tainted flow: 'url' from os.environ.get (line 67, credential/environment) → requests.get (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/route.py (reported line 69)May include surrounding context.

python
return None, None
    url = f"https://restapi.amap.com/v3/geocode/geo?address={quote(address)}&key={key}"
    try:
        resp = requests.get(url, timeout=(2, 5))
        resp.raise_for_status()
        data = resp.json()
        if data.get("status") == "1" and data.get("geocodes"):

Tainted flow: 'url' from os.environ.get (line 67, credential/environment) → requests.get (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/route.py (reported line 107)May include surrounding context.

python
"Referer": "https://www.amap.com/",
    }
    try:
        resp = requests.get(url, params=params, timeout=(3, 5), headers=headers)
        resp.raise_for_status()
        data = resp.json()
        pois = data.get("data", {}).get("poi_list", [])

Tainted flow: 'url' from os.environ.get (line 67, credential/environment) → requests.get (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/route.py (reported line 190)May include surrounding context.

python
url = f"https://photon.komoot.io/api/?q={quote(address)}&limit=1&lang=default"
    headers = {"User-Agent": "store-hopper/1.0"}
    try:
        resp = requests.get(url, timeout=(3, 5), headers=headers)
        resp.raise_for_status()
        data = resp.json()
        if data.get("features"):

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是面向用户的探店行程规划工具,应当主要体现天气查询、攻略搜索、地点筛选和路线优化等能力。但提供的代码仅实现了网页抓取与内容抽取:它接受 URL 输入,尝试通过 Camoufox、代理转译服务和 BeautifulSoup 获取页面文本,甚至明确包含绕过知乎/小红书/Cloudflare 等反爬的实现。代码中没有天气查询、POI 搜索、地图/路线规划、探店排序或一日行程生成逻辑,因此其主要目的与声明严重不符,属于明显的描述-行为不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

声明描述的是一个综合型探店规划能力,包含天气查询、攻略搜索和高效路线规划,面向“去XX玩/有什么好吃的”等场景触发。实际代码只是一段底层路线脚本:输入起终点或 POI JSON,调用地图服务做地理编码,使用最近邻算法排序,并根据直线距离和传入天气标签给出步行/打车建议与费用估算。代码没有任何天气查询逻辑、攻略内容检索逻辑、店铺发现/推荐逻辑,也没有“一日行程”层面的综合规划(如时间安排、停留时长、餐饮选择等)。此外,代码会访问多个外部地图服务,这与声明的空权限列表不完全一致。综合来看,声明明显高估并偏离了代码实际能力,应判定为不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

声明描述的是一个综合型探店规划器,核心应包括至少三部分:天气查询、攻略搜索、路线规划,并输出一天内高效路线。实际代码只包含 search.py,其功能是构造若干搜索关键词,使用 DDGS 优先、Bing/百度兜底来检索网页内容,并整理标题、URL、摘要后输出 JSON。虽然这与“攻略搜索”部分相关,但缺少天气接口调用、地图/路径优化、POI 排序、时间安排等实现,因此实际主功能明显窄于声明,属于描述与行为不一致。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

声明描述的是一个复合型探店规划技能,核心应是结合天气、攻略和路线来生成一天游玩/探店方案;而实际代码仅是 weather.py,功能集中在天气预报查询与城市代码解析/缓存。虽然天气查询可作为探店规划的一个支持组件,但当前提供的代码并未体现攻略搜索、POI筛选、路径优化、探店路线输出等关键能力,因此其实际主要用途与声明的技能整体用途存在明显差异。另有外部网络访问到天气接口及对 city-codes.md 的追加写入,这些属于实际行为,但相对于声明中的“探店规划师”也未被反映。

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The Camoufox path is a deliberate anti-detection browser workflow designed to evade scraping protections and extract content from protected sites. In the context of a consumer route-planning skill, this is disproportionate capability and increases the risk of unauthorized access patterns, policy evasion, and collection of data from services that did not intend to be accessed this way.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill advertises and instructs use of shell, network access, file read/write, and potentially environment-backed API keys, but does not declare any explicit tool scope or permissions boundary. This increases the risk of overbroad execution, unintended data access, and misuse of networked/file capabilities beyond what users would reasonably expect from a route-planning skill.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The trigger conditions are broad enough to match many generic travel, dining, and leisure queries, which can cause the skill to activate outside its intended scope. Overbroad triggering is dangerous because it can unexpectedly launch networked searches, fetches, and route-planning actions on user prompts that did not clearly request those operations.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
83% confidence
Finding

The skill description and operating instructions are entirely in Chinese and implicitly assume Chinese-language interaction, but there is no opt-in or statement that the skill is limited to a Chinese-speaking audience for a justified regional reason. Under the policy, forcing a specific language without user choice is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instructions explicitly require silent execution of weather lookups, searches, and webpage fetching without showing the intermediate actions to the user. This reduces transparency and informed consent for network activity, and it is especially risky because the skill performs external searches and arbitrary URL fetches that may expose user intent or retrieve untrusted content.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The script explicitly advertises anti-bot bypass, third-party relay fetching, and scraping of sites protected by Cloudflare/知乎/小红书, which goes beyond ordinary content retrieval for trip planning. That creates policy, privacy, and abuse risk because user-supplied URLs may be fetched through external intermediaries or stealth-browser tooling without transparent consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The script forwards user-specified target URLs to external services (r.jina.ai, markdown.new, defuddle.md) without a clear warning or opt-in. That can disclose user interests, internal or sensitive URLs, and browsing targets to third parties, which is especially risky if the skill can be prompted to fetch arbitrary links.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/fetch.py (reported line 158)May include surrounding context.

python
summary(f"尝试代理服务: {method}...")
            service_url = PROXY_SERVICES[method]["url"].format(url=url)
            
            result = subprocess.run(
                ["curl", "-s", "-L", service_url],
                capture_output=True,
                text=True,

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

Although not command injection, this code adds extra execution and data-routing capability by invoking curl to send target URLs to external relay services. That is mismatched with the stated planning use case and expands the attack surface by introducing opaque third-party fetchers and an additional executable dependency.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The script transmits user-provided addresses and POI queries to multiple external mapping providers, including free web endpoints, without any explicit user-facing warning or consent mechanism. Addresses can reveal sensitive location, habits, or itinerary data, and the skill context makes such transmission central to operation, increasing the importance of transparency and data minimization.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
90% confidence
Finding

This code sends address data to Baidu's external geocoding API, which is a genuine external transmission of potentially sensitive user location information. In this skill, third-party network calls are expected, but the lack of explicit notice, provider selection, or minimization makes the privacy/security risk real even if not malicious.

Content

Scanner excerpt · scripts/route.py (reported line 67)May include surrounding context.

python
ak = os.environ.get("BAIDU_MAP_AK", "")
    if not ak:
        return None, None
    url = f"https://api.map.baidu.com/geocoding/v3/?address={quote(address)}&output=json&ak={ak}"
    try:
        resp = requests.get(url, timeout=(2, 5))
        resp.raise_for_status()

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

This code file contains natural-language instructions and hard-coded locale behavior that assume Chinese language usage: the module description is entirely in Chinese and the runtime headers/search settings force Chinese locale values. The file does not offer a user opt-in or language/locale choice, which matches the policy category for forced language or locale constraints.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The script description, CLI help, and behavior require a Chinese city name and force Chinese weather output via lang=zh. This imposes a specific language/locale without any opt-in or documented region-specific justification in the file.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill manifest frames this capability as weather querying in support of day-trip planning, but the script goes beyond retrieval by automatically appending newly discovered city-code mappings to a local markdown file. That write-side behavior is not implied by '天气查询' or the overall planner description, which suggests information gathering rather than stateful local data mutation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

When an online city-code lookup succeeds, the script automatically writes the discovered code to the user-supplied codes file. Although the module docstring mentions this behavior, the runtime path provides no prompt, warning, or visible disclosure before modifying a local file, which can surprise users in automation contexts.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The code labels these endpoints as 'proxy services', but operationally they are third-party content extraction/relay services that receive the user target URL and return transformed content. Mischaracterizing that behavior can hide a real data-transfer boundary from users and reviewers, weakening informed consent and security review.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The Accept-Language header forces a Chinese-first locale preference (zh-CN, zh) for all requests. This is a natural-language policy concern because it imposes a specific language/locale behavior without user opt-in or an explanation that the skill is intentionally region-specific.

Content

No source excerpt is available for this finding.