概念

security agent safety stub

创建2026-05-10

更新2026-05-10

阅读量级1 分钟

概念导读

Prompt injection is an attack where untrusted content attempts to override, redirect, or manipulate an LLM or agent's intended instructions.

Prompt Injection

Prompt injection is an attack where untrusted content attempts to override, redirect, or manipulate an LLM or agent's intended instructions.

For tool-using agents, prompt injection can lead to unsafe tool calls, data exfiltration, or persistent memory contamination.

agent-safety — Broader safety context.
clawguard — Runtime defense against indirect prompt injection.
ssl-skill-representation — Skill representation can interact with tool-use security.

Related