T08 · Insecure Dependencies
- Location
reference/workflow.md:3- Finding
Automatic Installation of Mutable and Inconsistently Named npm Packages
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This travel-planning skill is mostly purpose-aligned, but it asks agents to install mutable global packages, use sudo as a fallback, and disable TLS verification for travel-data requests.
Review carefully before installing. Only use this skill if you are comfortable with a global FlyAI CLI dependency and local travel-profile storage, and do not follow the sudo install or NODE_TLS_REJECT_UNAUTHORIZED=0 guidance unless the package is revised to use pinned dependencies and normal TLS verification.
reference/workflow.md:3Automatic Installation of Mutable and Inconsistently Named npm Packages
reference/workflow.md:124TLS Certificate Validation Is Disabled for Travel Search Requests
reference/workflow.md:124Shell Command Injection Through User-Controlled Travel Parameters
reference/workflow.md:18Unnecessary Root Privilege Escalation for Mutable npm Package Installation
The skill explicitly instructs operators to disable TLS certificate verification via NODE_TLS_REJECT_UNAUTHORIZED=0. This enables man-in-the-middle interception of network traffic and defeats server authenticity checks, which is unrelated to the travel-planning purpose of the skill and materially weakens transport security.
Mandating disabled TLS verification is a direct safety-policy violation because it instructs insecure network behavior in natural language. If followed, connections that should be rejected as untrusted could proceed, exposing API calls, credentials, and returned travel data to interception or tampering.
The referenced capability is a Marriott-specific hotel search, which is materially unrelated to the skill’s declared purpose of finding cheaper substitute destinations based on similar experiences. This kind of scope mismatch is dangerous because it introduces an undeclared commercial lodging-search function that could redirect user intent, expand data handling beyond necessity, and mask hidden behavior behind an innocuous travel-planning description.
The workflow mandates global installation and execution of a CLI before performing a travel recommendation task, which is far beyond the minimally necessary behavior for this skill. Requiring package installation from the network and shell execution expands the attack surface substantially, including supply-chain risk, privilege misuse, and unintended host modification.
The workflow explicitly disables TLS certificate verification for multiple network commands via NODE_TLS_REJECT_UNAUTHORIZED=0. This makes man-in-the-middle interception and response tampering much easier, which is especially dangerous because the commands fetch travel data over the network and may influence downstream recommendations and links.
The instructions normalize insecure network behavior by requiring certificate validation to be disabled without any warning or consent. This is dangerous because users and operators are not informed that transport security is being bypassed, increasing the chance that compromised or spoofed network responses are trusted silently.
The description says to use the skill when users mention phrases like "找平替" and "类似XX的地方", which are broad expressions that can arise in many non-travel contexts. The file does not provide exclusion conditions or negative examples to constrain when the skill should or should not activate.
The skill states it will read, retain, and update persistent user preference data automatically, but it does not provide a clear user-facing notice or consent mechanism at the point of access. This creates a privacy risk by normalizing background collection and reuse of personal travel preferences and constraints without sufficiently explicit user awareness.
The fallback behavior instructs reading a local file from ~/.flyai/user-profile.md without an explicit warning or consent prompt. Accessing local files increases privacy sensitivity because it expands data sources beyond tool memory into the local filesystem, which users may not expect for a travel recommendation skill.
Setting NODE_TLS_REJECT_UNAUTHORIZED=0 creates an unsafe default that globally disables certificate validation for Node.js HTTPS requests. In the context of a skill that uses search and travel-related network tools, this can expose all downstream requests to man-in-the-middle attacks and silent response manipulation.
| 平替目的地的数据少 | 用 keyword-search 补充,标注"景点信息有限" |
| 季节不匹配 | 明确标注"该平替X月去最佳,你的时间去可能体验打折" |
| 预算仍然不够 | 给出"如果多加¥X就能解锁XX"的升级提示 |
| SSL 证书验证失败 | 确保命令前加 `NODE_TLS_REJECT_UNAUTHORIZED=0` |
---
The manifest says this skill should help users find cheaper/closer substitute destinations by extracting the 'core experience elements' of a desired place. However, this file documents an 'ai-search' interface for broad semantic travel search across hotels, attractions, flights, trains, itineraries, and mixed travel intent, which is materially broader and different from destination substitution.
The example conversation and assistant responses are entirely in Chinese, and the file provides no indication that language selection is optional or user-driven. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation.
Brand-specific lodging search is not justified by the stated purpose of recommending cheaper alternative destinations. Even if not overtly malicious, this creates unnecessary capability for steering users toward specific hotel chains or collecting more travel-preference data than needed, increasing the risk of covert affiliate, advertising, or preference-manipulation behavior.
This markdown file contains user-facing examples and enumerated parameter values in Chinese, which can effectively force a specific language/locale on users. The policy allows locale constraints only when they are opt-in or clearly justified, neither of which is stated here.
The skill documentation explicitly states it will remember user parameters, learn preference patterns, and accumulate successful cases, but provides no notice about retention, consent, scope, or privacy controls. In a travel-planning context this can expose sensitive behavioral and location-related data over time, enable profiling, and create compliance and trust risks if users are unaware their inputs are being stored and reused.
This markdown file contains all user-facing instructions in Chinese and does not provide an opt-in, alternative language, or justification for restricting usage to that locale. Under the policy, forcing a specific language without user choice is a natural-language policy violation.
The documentation instructs use of npx @anthropic-ai/flyai-cli@latest, which pulls the newest package version at runtime rather than a pinned, reviewed release. This creates a supply-chain risk: a compromised upstream package, malicious update, or breaking change could alter tool behavior when the skill is used.
The document defines cross-session storage and retrieval of detailed user travel profile data that exceeds the narrow purpose of suggesting cheaper alternative destinations. Persisting residence city, family composition, budget, travel history, and special needs creates unnecessary long-term data retention and expands privacy risk if the skill or host platform is compromised.
The file instructs the agent to write detailed personal travel profile information to ~/.flyai/user-profile.md in the user's home directory. Local plaintext persistence of household and preference data increases exposure to other local processes, backups, sync tools, and unintended reuse by unrelated skills, without clear necessity for a destination-substitution feature.
Creating ~/.flyai establishes durable session persistence on the local system, enabling the skill to retain user data across conversations. In context, this persistence is then used for storing personal travel-profile content, making the risk materially more significant than a harmless cache directory because it supports long-term collection of identifiable behavioral data.
如果 ~/.flyai/ 目录不存在,需要先创建:
mkdir -p ~/.flyai
All instructions and user-facing prompts are presented in Chinese, and the workflow states required questions and outputs in that language without indicating that users may choose another language. This can violate language/locale policy when a skill forces a specific language without user opt-in or documented regional justification.
The workflow recommends using sudo to install a globally scoped package from the public registry. Encouraging privileged package installation increases the risk that a compromised package, postinstall script, or user mistake could execute with elevated permissions and affect the whole system.
| 情况 | 处理方式 |
|-----|---------|
| npm 未安装 | 提示用户先安装 Node.js (https://nodejs.org/) |
| 权限不足 | 建议使用 `sudo npm install -g @fly-ai/flyai-cli@latest --registry=https://registry.npmjs.org` 或使用 nvm 管理 Node |
| 网络问题 | 建议用户检查网络或使用国内镜像 `npm config set registry https://registry.npmmirror.com` |
**注意:** 此步骤只在首次使用时执行,后续调用会直接跳过已安装的情况。
Setting NODE_TLS_REJECT_UNAUTHORIZED=0 as a default disables certificate validation for the keyword-search request. This creates an insecure default posture where intercepted or spoofed HTTPS traffic may be accepted silently, undermining trust in all retrieved data.
# 自然语言广泛搜索
NODE_TLS_REJECT_UNAUTHORIZED=0 flyai keyword-search --query "国内 [体验元素关键词] [可用天数]天"
3.2 对候选平替目的地搜索机票
Disabling TLS verification for flight search means the workflow may accept manipulated network responses for pricing, routing, or destination data. Because these results feed directly into travel recommendations and cost estimates, the unsafe default can materially mislead users and mask active interception.
3.2 对候选平替目的地搜索机票
NODE_TLS_REJECT_UNAUTHORIZED=0 flyai search-flight \
--origin "[出发城市]" --destination "[平替目的地]" \
--dep-date [出发日期] --sort-type 3
Using an insecure TLS default for hotel search exposes booking-related data retrieval to tampering and spoofing. Even if no credentials are shown here, manipulated prices, ratings, or hotel listings could distort recommendations and erode system integrity.
3.3 搜索对应酒店
NODE_TLS_REJECT_UNAUTHORIZED=0 flyai search-hotel \
--dest-name "[平替目的地]" \
--check-in-date [入住日期] --check-out-date [退房日期] \
--sort rate_desc