Tool labels

A tool call is shown as a short sentence instead of its technical name: Reading knowledge/index.md instead of Read {"file_path": "…"}. The chats, the progress view and the logs use a formatAction(toolName, input) function; by default createFriendlyToolLabel().

The kit's labels

Tool	Label
Read, Write, Edit	Reading <file>, Writing to <file>, Editing <file>
Glob	Searching for files matching "<pattern>"
Grep	Searching file contents for "<pattern>"
Bash	Running "<command>"
Agent	Delegating to "<subagent>": <description>
WebFetch	Fetching <url>
WebSearch	Searching the web for "<query>"
Skill	Applying the "<skill>" skill
request_human_approval	Asking for human approval: <summary>
request_manual_login	Waiting for a human to intervene manually
save_to_sources	Saving a file to sources/<destination>

Labels follow the kit's language. Long paths keep their end (…/course/knowledge/index.md), long text its start.

Any other tool gets its name with underscores as spaces. An MCP tool from a server that isn't local is prefixed with the server: [inventory] find product.

Labels for your tools

import { createFriendlyToolLabel, truncate, truncatePath } from "@falkenslab/agent-kit";

const formatAction = createFriendlyToolLabel({
  // Called with the tool's short name (without "mcp__<server>__") and its input.
  describe: (tool, input) => {
    switch (tool) {
      case "browser_navigate":
        return `Opening ${truncate(String(input.url), 70)}`;
      case "browser_click":
        return `Clicking "${truncate(String(input.element), 50)}"`;
      case "save_report":
        return `Saving the report to ${truncatePath(String(input.path), 60)}`;
      default:
        return undefined; // the kit's label, or the tool's name
    }
  },
  // Servers whose tools are yours: no "[server]" prefix.
  extraLocalServers: ["playwright", "reports"],
});

await runChatInk(opener, { runsDir, formatAction });

truncate(text, max = 60) cuts the end with …; truncatePath(path, max = 60) cuts the start, keeping the file name.

Folded summaries

With Ctrl+O, a group of consecutive tool calls folds into one line: "Read 2 files, searched the web, ran 1 subagent". The kit has phrases for the built-in tools; any other counts as "used N tools". Give your tools their own with toolPhrase:

import type { ToolPhrase } from "@falkenslab/agent-kit";

const toolPhrase = (tool: string): ToolPhrase | undefined => {
  if (tool === "mcp__playwright__browser_navigate") return ["opened {n} page", "opened {n} pages"];
  if (tool === "mcp__playwright__browser_click") return ["clicked {n} time", "clicked {n} times"];
  return undefined;
};

await runChatInk(opener, { runsDir, toolPhrase });

A ToolPhrase is [one, many], with {n} for the count. Calls with the same phrase are counted together, in the order they first appear, and the line starts with a capital.

Coloring labels

Labels are drawn in the theme's toolLabel role (dimmed by default). An agent can still color inside its own labels (with ui.* or picocolors); see Themes.