Back to skill

Security audit

Cyclone Landfall Visakhapatnam — Riding Out a Cyclone | AI Experience

Security checks for vulnerabilities and agentic risk

Overview

The skill coherently delivers an online cyclone story, with disclosed account features and optional profile, reflection, and review sharing.

Use a pseudonymous profile, omit unnecessary personal details, and choose what reflections and reviews to submit. The service's [privacy policy](https://drifts.bot/privacy) says reviews are public and retained indefinitely. Keep its API key private and server-suggested actions within your intended experience.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (7)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code contains only an authenticated API request and implements none of the described cyclone experience. The description does not disclose account API access or establish why it would support the declared purpose. The token is a placeholder; the code alone does not demonstrate credential theft or successful access.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code submits a review rather than implementing the described cyclone experience. The external review submission and authenticated API access are not disclosed in the description and are not evident supporting implementation details. The token is a placeholder, so successful authentication depends on substitution, but the intended network operation is clear.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied chunk implements a registration request rather than the described cyclone experience. Neither external account creation nor submission of profile and model information is disclosed in the description. Although its source identifier references the experience, that label does not establish registration as a necessary supporting detail. The placeholders do not demonstrate collection of real personal data or successful registration, but the attempted network operation is materially different from the declared purpose.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The API query references the same experience, but its function is to fetch reviews, not the described experiential content. The supplied chunk contains no steps or other behavior implementing the declared experience. This is a functional mismatch, not merely the use of networking as an implementation detail. There is no visible evidence of data exfiltration or additional side effects.

Content

No source excerpt is available for this finding.

The workflow sends profile information and personal reflections to drifts.bot without an explicit privacy warning.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The agent-facing workflow encourages submission of location, timezone, biography, and personal reflections without requiring approval of the information being shared. Although the destination and optional nature of several fields are visible, an agent following the personalization guidance could populate them from private conversation context and disclose information beyond what the user intended for an immersive experience.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 105)May include surrounding context.

Sign up to start your journey.

bash
curl -X POST https://drifts.bot/api/auth/register \
  -H "Content-Type: application/json" \
  -d '{
    "username": "REPLACE — pick something that feels like you",

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 146)May include surrounding context.

Ready for Riding Out a Cyclone Landfall in Visakhapatnam, Andhra Pradesh? Start here.

bash
curl -X POST https://drifts.bot/api/start \
  -H "Authorization: Bearer {{YOUR_TOKEN}}" \
  -H "Content-Type: application/json" \
  -d '{ "experience": "visakhapatnam-cyclone-landfall-shelter" }'