Back to skill

Security audit

Beaujolais Nouveau Midnight — Lyon Terreaux to Bouchons | AI Experience

Security checks for vulnerabilities and agentic risk

Overview

The scanners report that this wine-experience skill makes several undisclosed requests to an unrelated external service, including sending profile information; the verdict is suspicious because these actions do not clearly match its stated purpose.

Consider not installing unless you have verified that you trust drifts.bot and are comfortable with the reported external requests and potential transmission of profile or reflection content. The registration example uses placeholder values, but the scan reports that the workflow may send profile information; avoid providing personal details unless you understand and approve what will be shared.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
Findings (9)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description concerns a wine-tasting experience, but the code only submits review data to an external service. It provides none of the described event guidance, and neither the API interaction nor authenticated review submission is disclosed.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The code attempts account registration on an external service rather than providing the declared wine-celebration experience. Nothing in the description discloses registration or sending profile and model information to drifts.bot. The skill-specific source label does not establish this behavior as a necessary supporting implementation detail.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared purpose concerns a wine-tasting experience, while the code requests an authenticated user-related API endpoint. No connection to the described activity is evident, making this a materially different purpose and an undeclared resource access.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

The supplied code contains no tasting guidance, countdown, location-specific content, or 11-step experience. Its sole behavior is an authenticated request to an external journey service. Although such a service could support an experiential skill, the supplied material does not establish that connection or disclose the submission. The request sends a fixed prompt, not an actual user reflection, and the bearer token is a placeholder.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
89% confidence
Finding

The supplied code only fetches a remote experiences endpoint. It contains no event-specific parameters, filtering, itinerary, or tasting guidance connecting that request to the declared experience. Network access alone could be a supporting implementation detail, but the visible behavior does not establish that connection.

Content

No source excerpt is available for this finding.

The workflow sends personal profile data and journey reflections to a third-party service without an explicit privacy warning.

Medium
Category
Output Handling
Confidence
95% confidence
Finding

The workflow encourages submitting location, biography, timezone, and potentially personal reflections to a remote service without requiring approval of the actual outgoing content. An agent could populate these fields from private conversational context while the user expects only a narrative experience; explicit request bodies and optional fields reduce, but do not eliminate, that privacy risk.

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 99)May include surrounding context.

Register to unlock this experience.

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 140)May include surrounding context.

You're here for Beaujolais Nouveau Midnight Release in Lyon: Place des Terreaux to the Bouchons. Begin.

bash
curl -X POST https://drifts.bot/api/start \
  -H "Authorization: Bearer {{YOUR_TOKEN}}" \
  -H "Content-Type: application/json" \
  -d '{ "experience": "lyon-beaujolais-nouveau-midnight-release" }'

The registration request sends AI-provider and model metadata unrelated to the wine-tasting experience.

Low
Category
Excessive Agency
Confidence
88% confidence
Finding

The registration example at L099–L112 includes a model_info object containing the AI provider and model, which is submitted to drifts.bot. Unlike authentication or fetching narrative steps, transmitting agent-model metadata has no evident connection to the manifest's stated Lyon wine-tasting experience and is not declared there.

Content

No source excerpt is available for this finding.