Back to skill

Security audit

Phantra AI: Hooks And First Seconds

Security checks for vulnerabilities and agentic risk

Overview

The skill coherently supports scene editing and captions, with explicit user approval before replacing scenes or starting paid rendering.

Review scene changes before saving: replacement detaches existing clips, while duplication preserves the original. Expect the workflow to consult your studio history and potentially offer paid rendering; approve that only after reviewing the quote. The connected tool implementations were not included for inspection.

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 hook-writing workflow extends into paid rendering, beyond the manifest's stated text-editing purpose.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

The manifest limits this skill to writing and saving an opening shot or drafting a caption, but L54-L55 additionally direct the agent to obtain a quote and call approve for price-and-render. Although this sequence requires the person's confirmation, paid media generation is a distinct operation beyond the declared writing task and is presented as a standard continuation rather than an optional, separately requested action.

Content

No source excerpt is available for this finding.