Back to skill

Security audit

Ego YouTube Video

Security checks for vulnerabilities and agentic risk

Overview

Review is needed because overlapping uploads can mix files and settings, and the script can report private visibility without confirming it.

Prefer a version that isolates each upload's configuration and requires confirmed visibility before saving. Until fixed, avoid overlapping uploads and verify the channel and actual visibility in YouTube Studio. Cookie checks are documented and remain internal in the inspected workflow. Review included source inspection and mocked checks; live YouTube behavior and Ego Lite itself were not tested.

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
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (3)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The supplied code shows no login check, browser interaction, local video upload, visibility settings, or return of a video ID and processing status. Running tests does not substantiate the declared publishing behavior. The underlying test script is unavailable, so its effects cannot be determined from this chunk.

Content

No source excerpt is available for this finding.

The probe returns all JavaScript-readable cookies, an unnecessary sensitive-data capability for uploading a video or checking login status.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The probe copies all JavaScript-readable cookies for the current document into its return value, unnecessarily exposing browser data to the automation caller and potentially downstream logs or agent context. The documented login check relies on page content and navigation rather than cookie values; HttpOnly cookies are not exposed, and this source does not establish external exfiltration.

Content

No source excerpt is available for this finding.

The default probe reads and returns browser cookies without disclosure.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

When no operation is specified, studioAct defaults to probe and returns document.cookie without any need established by the upload workflow. This silently expands an ordinary page inspection into sensitive-data collection; it is the same underlying exposure identified by SDI-2, not evidence of a separate exfiltration channel.

Content

No source excerpt is available for this finding.