NEWPractise supported commands in browser simulations, alongside the lesson.Try a browser practice exercise.Try a lab♥ Codelivly is 100% free, no paywall. If it helped you, chip in to keep it running.100% free. Help keep it so.Support us
Preview · 30 min
Master AI agent sequencing. Learn how attackers exploit text-based planning frameworks like ReAct to hijack logic, and defend with deterministic state machines.
Knowing the parts of an agent is only step one. The power, and the danger, is in how the agent sequences them to reach a multi-step goal. It does not simply pick a tool and fire it. It reasons about the state, chooses an action, observes the result and adjusts its plan.
The dominant method is ReAct (Reasoning + Acting), often extended with Chain of Thought (CoT) or Tree of Thoughts (ToT). The planning is not a fixed algorithm. The LLM writes structured text (Thought: ..., Action: ..., Observation: ...) inside its own context window.
For a defender this is a key weakness. The agent reasoning is just text in short-term memory, so it can be manipulated. If someone injects fake Thoughts or Observations, the agent may follow a chain of steps while believing that the plan is its own.
Create a free account to unlock the rest of this task, complete tasks, and earn Rep.
Sign in