Single agent

ReAct loop

Alternate reasoning and acting until the goal is met.

1 agentCore

How it works

1Write a Thought: what to do next and why.
2Take one Action: a single tool call.
3Observe the result and update the plan.
4Repeat until done — always with a max-step limit and a clear success condition.

Use it when

Open-ended tasks where the next move depends on what the last action revealed — exploration, debugging, multi-hop research.

Reach for something else when

The path is known up front (use a chain), or the loop has no stopping rule and can run away.

Where you stay in the loop

You set the goal, the available tools, and the stopping rule; the agent drives the loop. Stay close on open-ended runs — read the Thoughts to catch a plan going sideways before it burns steps.

In the wild

Fixing a failing test: think → run test → read error → edit code → re-run → stop when green.

Hand this to your agent

Run as a ReAct agent.

Loop in exactly this format:
  Thought: <your reasoning>
  Action: <one tool call>
…then stop and wait for me to paste the Observation. Never take more than one
action per turn. Give a Final Answer when the goal is met or after 8 steps.

Tools: <list>
Goal: <...>

Replace the <…> placeholders, paste it into your agent, and it'll scaffold the workflow with you.