Security audit

臺北市公車定點車機 Skill

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed Taipei bus lookup skill using public transit data, with only minor cautions about broad triggers and optional scheduled alerts.

Before installing, expect the skill to make public transit API calls and create local cache files for Taipei bus stops/routes. Use explicit Taipei bus questions for best results, and only enable the cron or push-notification examples when you intentionally want recurring background bus checks.

SkillSpector

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
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 (7)

Vague Triggers

Medium

Confidence: 90% confidence
Finding: The trigger condition activates on broadly phrased 'bus-related' requests without defining boundaries, exclusions, or confirmation steps. In an agent environment, this can cause the skill to engage unexpectedly on ambiguous travel, location, or logistics queries, increasing the chance of incorrect tool use, privacy overreach, or irrelevant actions.

Natural-Language Policy Violations

Medium

Confidence: 82% confidence
Finding: The output guidance and templates are exclusively in Chinese and present a de facto language requirement without checking user preference. This can lead to unsafe misunderstandings, especially for route, ETA, or anomaly information where clarity matters, and may cause the agent to ignore the user's requested locale.

Vague Triggers

Medium

Confidence: 91% confidence
Finding: The trigger list includes several broad, natural phrases such as asking where a bus is or whether it is arriving soon. In an agent environment, overly generic triggers can cause accidental invocation during ordinary conversation, leading to unintended data fetching, user confusion, or workflow interruption even if no direct code execution occurs.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The trigger list contains very broad everyday phrases such as '公車', '等公車', and related generic wording that can match a large amount of normal conversation. This increases the chance of accidental invocation, unintended routing into the skill, and possible context hijacking where the skill handles requests the user did not explicitly intend for it.

Vague Triggers

Medium

Confidence: 91% confidence
Finding: The trigger list includes broad natural-language phrases such as general bus-related queries that may appear in ordinary conversation. In an agent environment, overly permissive triggers can cause unintended skill activation, leading to unnecessary external API access, privacy leakage about user intent/location, or incorrect tool selection.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The mixed-language trigger set is underspecified and ambiguous, so short phrases like 'Taipei bus' or broad Chinese equivalents could match many benign conversations. This increases the risk of accidental activation and tool misuse, especially in multilingual chats where context disambiguation is harder.

Vague Triggers

Medium

Confidence: 95% confidence
Finding: The documented trigger phrases include very broad everyday language such as '等公交车', '公交车来了', and 'Taipei bus', which can plausibly appear in normal conversation without a clear intent to invoke the skill. In an agent environment, overly broad triggers can cause unintended activation, unnecessary external API calls, and incorrect routing of user requests to this skill.

Static analysis

No suspicious patterns detected.