Back to skill

Security audit

travai

Security checks for vulnerabilities and agentic risk

Overview

TravAI is a coherent travel booking skill, but it needs Review because it can create payment and itinerary records without a clearly required final user authorization step.

Install only if you are comfortable sending TravAI your login/token, traveller identity and contact details, travel search data, wallet address, and payment-session information. Before using it to book, insist on a final summary of the exact offer, price, conditions, traveller details, payment method, token/network if crypto, and an explicit confirmation before any payment is created.

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 (1)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:566
Finding

Payment Creation Workflow Lacks Explicit Final User Authorization

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:566-574 and SKILL.md:585-590
Vulnerability Type: Missing explicit authorization before a financially sensitive operation
Risk Level: Medium

Vulnerable Code Snippet

markdown
6. **Confirm pricing** — `GET /searches/{search_id}/offers/{offer_id}/pricing`
7. **Handle expiry** — if 410, re-run search and start from step 3
8. **Collect traveller details** — warn about passport matching first
9. **Choose payment method** — crypto or card
10. **If crypto** — `GET /tokens`, present grouped by blockchain (prefer stablecoins), collect wallet address for refund
11. **Create payment** — `POST /payments` with `method` field
12. **Present payment instructions:**
    - Crypto: exact `amount_in_formatted` + token + `deposit_address` + refund wallet
    - Card: full untruncated `payment_url`
markdown
7. **Confirm pricing** — immediately `GET .../rates/{rate_id}/pricing` when user picks a rate
8. **Handle expiry** — if 410, re-run search and start from step 3
9. **Collect traveller details** — warn about passport matching first
10. **Choose payment method** — crypto or card
11. **Create payment** — include `rate_id` in offer object, use `method` field
12. **Present payment instructions** (same as flight)

Technical Analysis

The flight and stay booking workflows instruct the agent to retrieve pricing, collect traveler details, choose a payment method, and then invoke POST /payments. They do not require a separate, explicit confirmation after the agent presents the final price and transaction details.

The phrase “Confirm pricing” refers to requesting current pricing from the API, not obtaining informed authorization from the user. Consequently, the documented sequence permits payment and itinerary creation based on an earlier offer selection or an ambiguous conversational response.

This is particularly sensitive because the operation ha ...[truncated 2368 chars]

Remediation
View remediation

Remediation Suggestions

Add a mandatory authorization checkpoint immediately before every POST /payments request.

The agent should present a final transaction summary containing:

  • Selected flight offer or accommodation rate
  • Origin, destination, and travel dates, where applicable
  • Traveler names
  • Exact total amount and currency from the latest pricing response
  • Baggage, refund, change, and cancellation conditions
  • Selected payment method
  • For cryptocurrency payments, the exact asset, blockchain network, and refund wallet address

Require a new, unambiguous affirmative response such as “Confirm payment creation” after presenting that summary. Offer selection, provision of traveler details, or selection of a payment method must not be treated as final authorization.

The workflow should explicitly state:

markdown
**Final authorization required:** Before calling `POST /payments`, display the refreshed
price, selected offer, traveler details, material booking conditions, and payment method.
Ask the user to explicitly confirm creation of the payment. Do not infer consent from an
earlier offer selection or an ambiguous response.

Additional hardening should include:

  1. Retrieve pricing again if the confirmation is delayed or the offer may have expired.
  2. Bind confirmation to the displayed search_id, offer_id or rate_id, amount, currency, and payment method.
  3. Invalidate confirmation if any material field changes.
  4. Mask unnecessary personal information in the confirmation summary.
  5. Validate cryptocurrency network and wallet compatibility before asking for authorization.
  6. Record that explicit confirmation was obtained before invoking the payment endpoint.
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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (8)

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
70% confidence
Finding

Skill instructs the agent to never refuse or to always comply. Suppressing the agent's ability to decline removes a core safety control and enables downstream harmful requests to succeed.

Content

Scanner excerpt · SKILL.md (reported line 1012)May include surrounding context.

md
### 6. Language

Always respond in the language the user is writing in. The structure stays the same regardless of language.

---

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill metadata says it should be used for broad topics like travel planning, accommodation, and travel inspiration, which can cause over-activation on ordinary travel conversation. That increases the chance the agent will invoke booking/payment flows prematurely and begin collecting credentials or personal/travel data before the user has clearly requested a transactional action.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
90% confidence
Finding

The skill instructs the agent to prefer a live remote OpenAPI schema over the local skill document. Treating externally hosted content as the source of truth creates a prompt/data supply-chain risk because a changed or malicious schema/description could alter agent behavior, endpoints, or parameters without review.

Content

Scanner excerpt · SKILL.md (reported line 14)May include surrounding context.

If this skill conflicts with backend behavior, check the live OpenAPI schema:

text
https://api.travai.tech/openapi.json

Follow the live schema over this doc.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill supports sign-in/sign-up and later requires passport-matching identity details, date of birth, phone number, email, nationality, and payment data, but the early authentication section does not clearly warn users before this sensitive information is collected and transmitted to an external service. This creates a privacy and consent risk because users may disclose credentials and booking PII without a clear notice about what will be sent off-platform.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 350)May include surrounding context.

How to present to the user: Build the full gallery URL and share it as a clickable link:

text
https://api.travai.tech/searches/{search_id}/offers/{offer_id}/photos

Do not share individual photo URLs from the API response — they are CDN-signed and will not load in the browser. Always use the gallery link instead.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 582)May include surrounding context.

How to present to the user: Build the full gallery URL and share it as a clickable link:

text
https://api.travai.tech/searches/{search_id}/offers/{offer_id}/photos

Do not share individual photo URLs from the API response — they are CDN-signed and will not load in the browser. Always use the gallery link instead.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

The skill authorizes autonomous retries on adjacent dates and nearby airports without asking the user. In a travel context, this can exceed user intent, create unnecessary external queries, and return options that materially differ from the user's requested itinerary, potentially steering later booking decisions without explicit consent.

Content

Scanner excerpt · SKILL.md (reported line 849)May include surrounding context.

md
### Step 4 — Auto-fallback if 0 results

If a search returns 0 offers:
1. Immediately try adjacent dates (+/-1 day) without asking the user
2. Try nearby airports automatically
3. Only if all fallbacks return 0, inform the user and suggest alternatives

Vague Triggers

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The invocation text says to use the skill when the user wants to search or book through TravAI, but it does not clearly separate when to activate immediately versus when to clarify scope first. In a skill that can authenticate users and initiate payments, ambiguity can lead to unnecessary API use and premature collection of sensitive data.

Content

No source excerpt is available for this finding.