Back to skill

Security audit

Airport Transfer Guide

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed Chinese-language airport transfer guide that fetches public travel context and saves local helper output, with no evidence of hidden persistence, credential access, or destructive behavior.

Install only if you are comfortable with a Chinese-language travel skill that may contact public airport, ifly.com, flightqueue.com, and flight-board pages, and may save transit_context.json plus screenshots locally. Treat visa, immigration, and baggage conclusions as guidance to verify with the airline or official airport/border sources.

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
  • 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
Findings (13)

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The documented behavior presents the skill as a traveler guidance assistant, but the actual capabilities include scraping remote sites and using headless Chrome screenshots, while not actually performing some promised tasks such as baggage-tag analysis. This mismatch is dangerous because reviewers, orchestrators, or users may grant trust and activation based on the benign description while the skill exercises stronger collection/execution capabilities than expected.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The instruction states that all output must use Simplified Chinese, including titles, tables, steps, and image captions. This forces a specific language/locale without user opt-in, which is a natural-language policy violation under the stated auditing rules.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The README states that the skill provides Chinese-language instructions, and the Chinese description reinforces that behavior. Because this is a natural-language locale constraint presented as the default behavior with no opt-in or alternative language option, it conflicts with the policy against forcing a specific language without user choice.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
90% confidence
Finding

The skill declares broad capabilities to invoke Python, perform outbound network access, and potentially read/write local files, but it does not define an explicit tool/permission scope. In an agent environment, this creates an over-privileged execution surface where prompt-triggered behavior can fetch arbitrary remote content or write artifacts without clear restriction, increasing the blast radius of misuse or prompt injection through traveler-supplied inputs.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The manifest says to trigger when someone asks about airport transfers, connecting flights, transit procedures, layovers, terminal changes, immigration needs, baggage handling, or what to do after a delay. This is a wide, open-ended list without explicit boundaries or negative examples, so ordinary travel-chat requests could invoke the skill unintentionally.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This JSON registry uses Chinese-only natural-language description text in metadata, and the airport entries continue that pattern throughout the file. Because SQP-3 applies to all file types and covers language/locale policy violations, this is a policy concern when the skill does not offer a language choice or document a justified locale constraint.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill's declared purpose is transit guidance, but this code additionally captures rendered screenshots from registry-defined URLs and writes PNG files to disk. That creates a broader-than-advertised execution and persistence capability, increasing risk if the registry is modified to point at sensitive internal endpoints, local resources, or attacker-controlled pages.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

Executing local browser subprocesses is a significant capability escalation for an airport-transfer assistant and is not clearly necessary for providing text guidance, maps, or queue estimates. In skill context this is more dangerous because the browser is directed to registry-provided URLs, enabling interaction with remote content through a full browser engine rather than simple HTTP requests.

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_transit_context.py (reported line 607)May include surrounding context.

python
for path in paths:
            try:
                result = subprocess.run(
                    [path, '--version'],
                    capture_output=True,
                    timeout=5

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
89% confidence
Finding

The code launches a local Chrome process on arbitrary URLs sourced from the airport registry and renders remote content in headless mode. Even without shell injection, this expands the skill's capabilities from simple HTTP fetching to active local browser execution, which can be abused for SSRF-like access to internal network resources, unintended local file access attempts via browser-supported URL schemes, or exposure to browser-side vulnerabilities if registry entries are tampered with.

Content

Scanner excerpt · scripts/fetch_transit_context.py (reported line 637)May include surrounding context.

python
url
            ]
            
            result = subprocess.run(cmd, timeout=timeout, capture_output=True)
            
            if result.returncode == 0 and Path(output_path).exists():
                return True

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README explicitly describes fetching data from third-party sources and taking page screenshots, but it does not warn users that running the script may make outbound network requests or transmit query parameters such as airport, terminal, and transit timing to external services. In a travel-assistant context this can expose itinerary metadata and surprise users who expect a local-only skill, making it a real transparency and privacy issue even if it is not overtly malicious.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

The markdown explains that the script retrieves real-time information from third-party aggregators, official airport sites, and CDP-rendered pages, which implies outbound network requests and transmission of user-supplied itinerary context such as airport, terminals, and transit time. There is no explicit warning that the tool contacts external services or that request metadata may be exposed to those services.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

This markdown file documents creation of an output file via --output transit_context.json, and later describes cdp_screenshot_path plus CDP-based local screenshots. The document explains functionality but does not warn users that running the script will write files to disk, potentially including local screenshot artifacts.

Content

No source excerpt is available for this finding.