Back to skill

Security audit

Phantra AI: Camera Moves

Security checks for vulnerabilities and agentic risk

Overview

The skill provides camera guidance and saves scene drafts, with explicit user approval required before paid rendering.

Expect the workflow to save a complete scene; specify advice-only if you only want camera suggestions. Paid rendering requires your approval after pricing. The full package was inspected, but the separately supplied tools were outside this review.

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
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (1)

The skill directs full-scene composition and persistence despite explicitly excluding whole-scene work.

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

The manifest at L03 limits the skill to camera-move and framing guidance and explicitly excludes handling the whole scene. L74-L75 nevertheless instruct the agent to assemble the full scene, including style, sound, and avoidance instructions, and save it using write_scene, extending behavior beyond that stated scope.

Content

No source excerpt is available for this finding.