Agentic control
Tool calls
Use when the agent edits files, runs commands or reads documents, and people want to follow each action as it happens.
Edits, commands, and reads as a compact activity feed.
Part of Agentic control. The stage where the agent stops answering and starts doing. It covers the whole handover: agreeing what the agent may do, watching it work, stepping in, and checking what changed afterwards. Reach for these as soon as an action touches a customer, money or a record.
Do
- Give each kind of action its own compact block: a reasoning note, a file write, a command, a read.
- Show the evidence in the block, such as the changed lines or the command output.
- Finish with a summary of the files touched and how many lines changed.
Don't
- Don't dump raw tool JSON. Show the file name, the command and the result.
- Don't leave the activity indicator running after the last event.
- Don't mix the feed into the answer text. Keep actions and prose apart.
States
- Streaming: Events arrive one at a time under a header, with a thinking indicator on the right.
- Thinking note: Two lines of reasoning in plain text, with no container.
- File write: A block with the file name, the line count and the added lines in green.
- Command run: The command and its passing output.
- File read: The file name, its details and one line on what the agent learned from it.
- Complete: The indicator goes and a row of file chips shows lines added and removed.
Usage
import { ToolStreamPattern } from "./halaska-kit";
<ToolStreamPattern />
Built from
Agent Glyph, Stack, Thinking Indicator, Typography
Related patterns
Halaska UI · Components · Patterns · Screens · Install