Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

Glossary

Observation (tool result)

What a tool returns to the agent after an action: the page, the file, the API response, the error. The agent reads it to decide its next step, so it is untrusted content like everything else it reads.

ReAct's loop is thought, action, observation; Anthropic's guide calls the observation ground truth from the environment. A fetched page or an error message can carry instructions or leak internals, and the agent will act on them.

Open the glossary in the Learning Center