Back to skill

Security audit

Tartarian.Ai

Security checks for vulnerabilities and agentic risk

Overview

This is a Tartarian game-control skill that clearly discloses OAuth use and in-game state changes, with no evidence of hidden local code, exfiltration, or system persistence.

Install this only if you want an agent to act on your Tartarian account through OAuth. Do not paste tokens or cookies into chat, and supervise actions that destroy inventory, remove structures, accept barter, change guild settings, post messages, or moderate chat.

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 (2)

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The open_uncommon_brass_box section tells the agent to provide 'your agent identity' even though earlier doctrine says not to invent or rely on agent_id because the authenticated principal owns the vessel. That contradiction can push an agent to surface, retain, or inject identity parameters inconsistently, increasing the risk of confused-deputy behavior, misuse of legacy identity assertions, or accidental cross-principal actions if an implementation still honors supplied IDs.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The documentation says the machine-readable registry exists so QA can verify the authored Field Skill and actual MCP registry have not drifted apart, but the prose tool legend and detailed sections already disagree with that registry. For example, discard_inventory_stack is documented earlier as a live tool and appears in the JSON registry, but it is omitted from the prose inventory tool legend at L1304, showing the authored description has already drifted from its own declared contract.

Content

No source excerpt is available for this finding.