Tool use lets Claude search the web or run scripts to help answer a prompt or produce requested work. In Claude Code, a single turn can combine repository searches, file edits, builds, and tests; the model runs in the cloud while tools run on your machine. section 1.1 section 5.1
Treat access as consequential. Claude Code can access files and tools you permit, but it has no physical sandbox to prevent unwanted behavior. Review permission requests and check the resulting work. section 5.1
Investigate an error
Give Claude Code an error message and ask it to investigate. It can search the repository for where the message is produced, trace the call stack and relevant data, then edit files and run the build or tests. section 5.1
Figure 5.1 Claude Code runs inside your repository with no sandbox. The permissions model asks your consent to access files and run tools, but there is no physical backstop to prevent unwanted behavior.
Use the figure to see how repository access and consent-based permissions coexist with the absence of a physical sandbox. section 5.1
Figure 1.2 How context builds over the turns of a conversation. The input grows cumulatively and there can be huge jumps from tool use when Claude is researching file content or web pages.
Use the figure to understand how research through tools can add substantially to conversation context. section 1.4.1

















