Tool Result
What came back from the call, appended to context.
The output the harness returns after executing a tool call. It is appended to the conversation so the model can reason about it on the next turn.
Tool results are context, and they add up fast. A tool that dumps ten thousand tokens of logs will crowd out everything else — return summaries and pointers instead.
Tool results are the most common cause of context bloat, because they arrive in bulk and get appended by default. A command that prints ten thousand lines of build output is now permanently in the window, competing for attention with the task.
The build log that ate the window
A test command returns several thousand lines and the task is only about one failure.
Truncate at the tool boundary and keep the tail, or return a summary plus a pointer to the full log file. The interesting part of an agent’s job is usually in the last fifty lines, and it is the part most easily buried by the first five thousand.
- blogWriting effective tools for AI agents — using AI agents ↗
Treats tools as contracts between deterministic software and a non-deterministic caller: clear boundaries, token-efficient results, evaluation against realistic tasks.
- docsTool use with Claude ↗
The tool_use / tool_result round trip, client-executed versus server-executed tools, tool choice and parallel calls.
Tool Call
The model asking for a specific function with specific arguments.
Tool
A function the model is allowed to call to affect the world.
Context
Everything the model can currently see.
Progressive Disclosure
Reveal detail only when it becomes relevant.