Chain of Thought

Problem

An agent often does several things before it answers: reads files, calls tools, reasons through a problem. Users want a quiet summary of that work ("Thought for 4 seconds", "Ran 3 tools") they can open to see the steps. PDS has no pattern for it. This is a net-new pattern.

Proposal

A collapsible list. The always-visible line is a Thinking State with a chevron. The body is a list of steps, each with an optional status icon (done, active, pending, error), a label, quiet secondary text, and optional expandable detail such as a tool call's input and output.

Source

Three reference screenshots from Erika (2026-10-12): a tool-run log ("Ran 3 tools: Read, Edit, Run"), a reasoning panel mixing italic reasoning text with expandable tool calls, and a checked step list ("Thought for 4 seconds"). No Figma or Jira links.