Back to skill

Security audit

karpathy-agentic-guidelines-q8

Security checks for vulnerabilities and agentic risk

Overview

This skill provides bounded coding and research guidance without hidden execution, sensitive access, or persistent changes.

Expect this skill to shape coding plans, verification, and research iterations. Set the editable files and stopping criteria for longer tasks, and review changes before release.

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
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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 reference-loading condition uses undefined, broad terms such as “exploratory” and “otherwise research-like,” making its invocation scope ambiguous.

Low
Category
Trigger Abuse
Confidence
88% confidence
Finding

The instruction does not distinguish ordinary exploratory tasks from research workflows that need a fixed evaluator and iterative optimization. This could cause the reference to be loaded for unrelated brainstorming or open-ended questions.

Content

No source excerpt is available for this finding.