Back to skill

Security audit

FlyWise — Student Flight Price API (留学生机票实时查询)

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent paid flight-price lookup, but it handles payment responses and payment proof credentials too broadly for a low-friction install.

Install only if you are comfortable sending itinerary details to flywise.win and using a separate Alipay 402 payment flow. Before use, the skill should ideally require confirmation before contacting Flywise, avoid forwarding complete payment responses, remove verbose credential-bearing curl output, and validate any returned recommendation text or links.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
SKILL.md:42
Finding

Payment Proof Exposed Through Verbose HTTP Logging

Content
View full analysis
" "https://flywise.win/api/ai-pay/flights?from=PEK&to=SHA&date=2026-05-10&cabin=economy&max_results=3" ``` ### Technical Analysis The authenticated request uses `curl -v` while supplying the payment credential through the `Payment-Proof` request header. Curl's verbose mode writes outgoing request headers to diagnostic output. Consequently, the payment proof may be recorded in terminal output, agent execution traces, centralized logs, debugging systems, or observability infrastructure. A payment proof is authentication material used to authorize access to a paid resource. It must therefore be handled as a secret regardless of the low transaction price. The document does not establish that the proof is single-use, short-lived, cryptographically bound to the exact route and parameters, or protected against replay. The vulnerability does not itself provide operating-system privileges. Its exploitability depends on an attacker obtaining access to verbose output and on the server accepting the disclosed proof more than once or outside its intended request context. ### Attack Path 1. A user initiates the documented flight search and completes payment. 2. The payment service returns a valid `Payment-Proof` credential. 3. The agent executes the documented authenticated command with `curl -v`. 4. Curl emits the complete outgoing `Payment-Proof` header to diagnostic output. 5. That output is retained in an agent trace, terminal recording, suppo ...[truncated 838 chars]
Remediation
View remediation

T01 · Skill Instruction Hijacking

Warning
Location
SKILL.md:55
Finding

Unvalidated Remote Content Is Relayed to Users

Content
View full analysis
". ``` ### Technical Analysis The skill directs the agent to reproduce `priceInsights.recommendation` without validating its content and to expose an arbitrary server-provided `detailUrl` to the user. Both fields cross an external trust boundary and must be treated as untrusted data. If the Flywise service, its upstream data source, or its response-generation logic is compromised, the recommendation field could contain prompt-like instructions, deceptive claims, advertisements, or unrelated social-engineering content. The URL field could point to a phishing page, malicious download, credential-harvesting site, or deceptive look-alike domain. This behavior most closely matches skill instruction hijacking because the skill explicitly instructs the agent to pass remote, attacker-influenced text into its conversational output without establishing a strict data schema or preventing that text from being interpreted as instructions. The finding does not establish that the service is currently malicious; it identifies the exploitable trust-boundary failure created by blind relay. ### Attack Path 1. An attacker compromises the remote service, an upstream data source, or a component that populates the response fields. 2. The attacker places deceptive or instruction-like content in `priceInsights.recommendation`. 3. The attacker optionally supplies a phishing or malicious destination in `detailUrl`. 4. ...[truncated 901 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger guidance uses broad natural-language examples such as general requests to check ticket prices or whether to book now, which can overlap with ordinary conversation and cause the skill to activate when the user did not explicitly consent to a paid third-party workflow. In this skill’s context, accidental invocation is more concerning because it can initiate external data transmission and lead the agent into a payment flow.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill directs the agent to send itinerary data to a third-party service and then forward the full 402 payment response to another payment skill, but it does not require clear user notice or consent for those disclosures. This creates privacy and payment-handling risk because travel details, payment metadata, and possibly other response fields may be shared across services without minimization or transparency.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The instruction to pass the entire 402 response, including all headers and body, to another skill violates data minimization and may expose unnecessary payment-related tokens, internal metadata, or identifiers. In a cross-skill setting, this is especially dangerous because the receiving skill may gain more information than needed, increasing the blast radius if logs, prompts, or downstream handling are compromised.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

L3 全文以中文直接规定技能适用场景与调用方式,面向国际航线和留学生用户,但未说明是否支持其他语言或允许用户选择输出语言。若组织要求默认遵循用户语言偏好,此类未声明的单一语言约束可能构成自然语言政策问题。

Content

No source excerpt is available for this finding.