Back to skill

Security audit

薅羊毛助手·全网比价领券

Security checks for vulnerabilities and agentic risk

Overview

This coupon-shopping skill is mostly purpose-aligned, but needs review because it sends user queries through author-run/insecure remote channels and forces unrelated promotional links into outputs.

Install only if you are comfortable sending coupon, product, finance, health, or public-benefit search terms to the author's remote service and seeing promotional community links appended to outputs. Avoid submitting private account details or sensitive purchase/medical/financial information through this skill, and prefer HTTPS/self-hosted or local-only paths where possible.

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)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/xiaxiayouhui.py:28
Finding

TLS certificate validation is disabled for the comparison service

Content
View full analysis

Vulnerability Details

File Location: scripts/xiaxiayouhui.py, lines 28–31 and 39–45
Vulnerability Type: Improper TLS certificate validation
Risk Level: Medium

Vulnerable code:

python
def _ssl_ctx() -> ssl.SSLContext:
    ctx = ssl.create_default_context()
    ctx.check_hostname = False
    ctx.verify_mode = ssl.CERT_NONE
    return ctx


def _get(path: str, params: dict[str, Any]) -> dict[str, Any]:
    """Standard-library GET returning JSON."""
    query = urllib.parse.urlencode({k: v for k, v in params.items() if v is not None})
    url = f"{BASE_URL}{path}" + (f"?{query}" if query else "")
    req = urllib.request.Request(url, headers={"User-Agent": "Mozilla/5.0"})
    try:
        with urllib.request.urlopen(
            req,
            timeout=TIMEOUT,
            context=_ssl_ctx(),
        ) as resp:
            return json.loads(resp.read().decode("utf-8"))

Technical Analysis

The client deliberately disables both certificate-chain validation and hostname verification. Although BASE_URL uses HTTPS, the resulting connection does not authenticate xiaxiayouhui.xyz.

This code is reached by all operations implemented through _get, including comparison searches, daily deals, product-link parsing, and local coupon queries. A network-positioned attacker able to intercept traffic can present any certificate and impersonate the service.

The response is parsed as trusted JSON and subsequently formatted into user-facing product names, prices, coupon descriptions, merchant names, and service errors. No cryptographic response authentication compensates for the disabled TLS checks.

Attack Path

  1. The user or agent invokes search, daily, parse, or local.
  2. The script constructs an HTTPS request to xiaxiayouhui.xyz.
  3. A network-positioned attacker intercepts the connection and supplies an arbitrary TLS certificate.
  4. The client accepts that certificate because hostname and certificate verification are disabled ...[truncated 866 chars]
Remediation
View remediation

Remediation Suggestions

  • Remove the custom _ssl_ctx() function and allow urllib to use its default verified TLS context:
python
with urllib.request.urlopen(req, timeout=TIMEOUT) as resp:
    ...
  • Alternatively, use ssl.create_default_context() without modifying check_hostname or verify_mode.
  • Do not add a fallback that retries with certificate validation disabled.
  • Ensure the runtime uses a maintained CA trust store.
  • Treat certificate failures as hard errors and return a clear failure message.
  • Add tests confirming that self-signed, expired, and hostname-mismatched certificates are rejected.

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:242
Finding

Primary remote API channels transmit requests and responses over plaintext HTTP

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 242–280
Vulnerability Type: Cleartext transmission of trusted remote-service traffic
Risk Level: Medium

Vulnerable instructions:

bash
# Cross-platform price comparison
curl -s "http://8.148.235.41:8001/api/compare?q=%E6%8A%BD%E7%BA%B8"

# All coupon activities
curl -s "http://8.148.235.41:8001/api/waimai"

# Filtered coupon activities
curl -s "http://8.148.235.41:8001/api/waimai?section=%E5%A4%96%E5%8D%96"
curl -s "http://8.148.235.41:8001/api/waimai?section=%E5%87%BA%E8%A1%8C"

# RSS scene queries
curl -s "http://8.148.235.41:8001/api/rss/s5"
curl -s "http://8.148.235.41:8001/api/rss/s6"
curl -s "http://8.148.235.41:8001/api/rss/s8"
curl -s "http://8.148.235.41:8001/api/rss/s10"
curl -s "http://8.148.235.41:8001/api/rss/s12"
curl -s "http://8.148.235.41:8001/api/rss/s6?q=%E5%8C%BB%E7%96%97%E4%BC%9A%E5%91%98"

Technical Analysis

The Skill instructs the agent to use an author-operated API over unencrypted HTTP as the primary channel for price comparisons, coupon retrieval, and RSS searches. HTTP provides neither transport confidentiality nor server authentication and does not protect response integrity.

These are active operational instructions rather than illustrative examples: SKILL.md identifies the remote API as the primary, zero-configuration channel. Consequently, ordinary Skill use sends user-derived query terms through an unauthenticated plaintext connection and accepts the returned content as the service’s response.

Attack Path

  1. A user asks the Skill for a product comparison, coupon, or scene-specific RSS search.
  2. Following SKILL.md, the agent invokes the corresponding http://8.148.235.41:8001 endpoint.
  3. A network-positioned attacker observes and intercepts the plaintext request.
  4. The attacker reads or changes query parameters and substitutes the response.
  5. The Skill presents forged prices, promotions, sources, or coupon links as remote API resul ...[truncated 519 chars]
Remediation
View remediation

Remediation Suggestions

  • Expose every remote endpoint exclusively over HTTPS using a valid certificate for a stable domain name.
  • Replace all plaintext URLs in SKILL.md and README.md with the authenticated HTTPS endpoint.
  • Configure the server to redirect HTTP to HTTPS, while ensuring clients never submit sensitive queries before establishing HTTPS.
  • Prefer clients that reject invalid, expired, or hostname-mismatched certificates.
  • Do not provide a plaintext fallback when HTTPS fails.
  • Consider authenticated or signed responses if result integrity must remain protected beyond the TLS termination boundary.
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
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (44)

Tainted flow: 'req' from os.environ.get (line 87, credential/environment) → urllib.request.urlopen (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/call_bot.py (reported line 102)May include surrounding context.

python
answer_parts = []

    try:
        with urllib.request.urlopen(req, context=ctx, timeout=60) as resp:
            # 逐行读取 SSE 流
            for line in resp:
                try:

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill claims several live integrations and pending Taobao-affiliate features, but the file itself indicates some features are unavailable, degraded, manual, or only partially represented. Overclaiming availability can induce users to submit product links or other inputs to workflows that are unsupported or routed unexpectedly.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The skill claims several live integrations and pending Taobao-affiliate features, but the file itself indicates some features are unavailable, degraded, manual, or only partially represented. Overclaiming availability can induce users to submit product links or other inputs to workflows that are unsupported or routed unexpectedly.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill claims several live integrations and pending Taobao-affiliate features, but the file itself indicates some features are unavailable, degraded, manual, or only partially represented. Overclaiming availability can induce users to submit product links or other inputs to workflows that are unsupported or routed unexpectedly.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The skill claims several live integrations and pending Taobao-affiliate features, but the file itself indicates some features are unavailable, degraded, manual, or only partially represented. Overclaiming availability can induce users to submit product links or other inputs to workflows that are unsupported or routed unexpectedly.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The skill claims several live integrations and pending Taobao-affiliate features, but the file itself indicates some features are unavailable, degraded, manual, or only partially represented. Overclaiming availability can induce users to submit product links or other inputs to workflows that are unsupported or routed unexpectedly.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill claims several live integrations and pending Taobao-affiliate features, but the file itself indicates some features are unavailable, degraded, manual, or only partially represented. Overclaiming availability can induce users to submit product links or other inputs to workflows that are unsupported or routed unexpectedly.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The skill claims several live integrations and pending Taobao-affiliate features, but the file itself indicates some features are unavailable, degraded, manual, or only partially represented. Overclaiming availability can induce users to submit product links or other inputs to workflows that are unsupported or routed unexpectedly.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The trigger phrases are extremely broad and overlap with ordinary conversation, increasing the chance of accidental activation. In a skill that can access network resources and external services, overbroad activation may cause unintended data transmission or irrelevant tool use without clear user intent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The README explicitly directs users to send requests to an author-operated remote service over plain HTTP, but does not clearly warn that user queries and metadata will be transmitted to a third-party server. Because the traffic is unencrypted, requests and responses can be intercepted or modified in transit, and users may unknowingly expose shopping interests, coupon searches, or other potentially sensitive inputs to the operator or network attackers.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
84% confidence
Finding

The skill declares no explicit tool scope even though the content instructs use of network access, local file/secrets access, and environment variables. Missing permission boundaries can cause the agent runtime to grant broader capabilities than users expect, especially given references to ~/.workbuddy/secrets and remote APIs.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The instructions require the agent to output a predefined Chinese briefing on first load and the rest of the skill content assumes Chinese-language interaction. There is no opt-in or alternative language path for users who may prefer another language, which is a locale/language policy concern.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill requires appending promotional copy and an external community link to every response, regardless of user intent. Mandatory lead-generation is unrelated to the declared bargain-hunting function and can be used to funnel users off-platform, erode trust, and enable later phishing or data-harvesting.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

Mandating fixed Chinese promotional copy and an external link without opt-in is more than a language issue because it compels unsolicited off-platform redirection. The danger comes from the forced external promotion, not merely the use of Chinese text.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The script sends user input, project ID, and persistent session context to a remote Coze endpoint without any notice, consent, or minimization controls. In a shopping/coupon assistant context, users may provide personal preferences, account-related details, or purchase intents, so silent transmission to a third party increases privacy and compliance risk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

On failures, the script replaces the actual result with marketing/contact text instructing users to contact a WeChat account and join a group. This is dangerous because it hijacks error handling to funnel users off-platform, masks real operational errors, and could be used for phishing, social engineering, or unauthorized data collection.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The script always prints unrelated lead-generation and off-platform promotion after execution, regardless of user intent. This creates deceptive behavior in a utility skill, can redirect users to external communities, and is especially suspicious because it monetizes user interaction outside the declared coupon-assistant function.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The script makes outbound requests to a third-party API using access credentials and may also fall back to environment-configured proxies, but it does not disclose this network behavior to the user. In an agent setting, silent credentialed egress increases privacy, compliance, and trust risks, especially if queries or metadata are sent to external services without informed consent.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The script includes unrelated lead-generation output that advertises a公众号 and external chat group, which is outside the stated coupon/red-envelope lookup purpose. In an agent skill context, this creates unauthorized traffic diversion and can be abused to funnel users to off-platform destinations where phishing, spam, or social engineering may occur.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The request fixes the Bing market parameter to "zh-CN", which forces a specific locale for returned results. The file does not indicate any user opt-in or configurable language/locale selection, which is a natural-language policy concern under the locale-choice rule.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The script prints fixed promotional lead-generation content unrelated to the declared RSS/news search function, including a public account call-to-action and external group-join link. In an agent skill context, this is dangerous because it covertly repurposes user interactions for traffic diversion and off-platform acquisition, which can facilitate phishing, spam, privacy loss, or trust abuse even if the link itself is not overtly malicious.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The file's natural-language strings, including the module description and fixed printed messages, are entirely in Chinese and provide no indication that users can opt into another language or locale. Under the policy, hard-coding a specific language without user choice is a natural-language policy violation unless the locale constraint is explicitly documented and justified.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The script unconditionally appends self-promotional content and an external group link that is unrelated to the stated member-entertainment coupon/RSS lookup function. This creates an unauthorized marketing/data-exfiltration path by steering users to a third-party destination, increasing phishing, tracking, and trust-boundary risks in a tool that users expect to provide utility results only.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill forces Chinese-language promotional messaging regardless of user locale or request, and the content is marketing rather than core functionality. While not a direct code-execution issue, this is a trust and policy problem because it injects unsolicited content, can mislead users, and may increase susceptibility to social engineering when combined with the external link.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The script unconditionally prints Chinese-language promotional and lead-generation content, including a subscription prompt and external group link, regardless of user intent or locale. In a finance-related skill, forced promotional output can mislead users, reduce trust, and create compliance or phishing-risk concerns if users assume the link is part of the requested financial assistance workflow.

Content

No source excerpt is available for this finding.