refactor: name the agent run breakdown a trace

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

frontend/src/lib/components/AgentResultDisplay.svelte

@@ -2,13 +2,13 @@
 	import type { Snippet } from 'svelte'
 	import { Badge } from '$lib/components/common'
 	import GfmMarkdown from './GfmMarkdown.svelte'
-	import AgentActions from './AgentActions.svelte'
+	import AgentTrace from './AgentTrace.svelte'
 	import { formatTokenCount, summarizeAgentResult, type AgentResult } from './aiAgentResult'
 
 	interface Props {
 		result: AgentResult
 		/** The answer, or what the agent did to get there; JSON is the viewer's own toggle. */
-		view: 'answer' | 'actions'
+		view: 'answer' | 'trace'
 		workspaceId?: string
 		/**
 		 * How to render an answer that is not text. An `output_schema` makes `output`
@@ -27,8 +27,8 @@
 </script>
 
 <div class="flex flex-col gap-2 w-full pt-1">
-	{#if view === 'actions'}
-		<AgentActions messages={result.messages} {workspaceId} />
+	{#if view === 'trace'}
+		<AgentTrace messages={result.messages} {workspaceId} />
 	{:else if textOutput !== undefined}
 		{#if textOutput === ''}
 			<span class="text-tertiary text-xs">The agent returned no answer</span>

frontend/src/lib/components/AgentTrace.svelte

@@ -8,7 +8,7 @@
 	import ToolContentDisplay from './copilot/chat/ToolContentDisplay.svelte'
 	import WebSearchSourcesDisplay from './copilot/chat/WebSearchSourcesDisplay.svelte'
 	import GfmMarkdown from './GfmMarkdown.svelte'
-	import { buildAgentActions, type AgentActionEntry } from './agentActions'
+	import { buildAgentTrace, type AgentTraceEntry } from './agentTrace'
 	import type { AgentMessage } from './aiAgentResult'
 	import { SvelteMap, SvelteSet } from 'svelte/reactivity'
 
@@ -19,7 +19,7 @@
 
 	let { messages, workspaceId }: Props = $props()
 
-	const entries = $derived(buildAgentActions(messages))
+	const entries = $derived(buildAgentTrace(messages))
 
 	let expanded = new SvelteSet<number>()
 	// A tool's own job holds what the envelope does not: its logs, how long it
@@ -45,7 +45,7 @@
 		}
 	}
 
-	function toggle(index: number, entry: AgentActionEntry) {
+	function toggle(index: number, entry: AgentTraceEntry) {
 		if (expanded.delete(index)) {
 			return
 		}
@@ -61,7 +61,7 @@
 		return typeof job === 'object' ? job : undefined
 	}
 
-	function toolLabel(entry: Extract<AgentActionEntry, { kind: 'tool' }>): string {
+	function toolLabel(entry: Extract<AgentTraceEntry, { kind: 'tool' }>): string {
 		const job = jobOf(entry.jobId)
 		const duration = job?.['duration_ms']
 		return duration === undefined

frontend/src/lib/components/DisplayResult.svelte

@@ -58,7 +58,7 @@
 	import type { MarkupTrust } from './apps/markupTrust'
 	import AgentResultDisplay from './AgentResultDisplay.svelte'
 	import AgentStreamDisplay from './AgentStreamDisplay.svelte'
-	import AgentActions from './AgentActions.svelte'
+	import AgentTrace from './AgentTrace.svelte'
 	import { isAgentStream, parseAgentErrorMessages, parseAgentResult } from './aiAgentResult'
 
 	const TABLE_MAX_SIZE = 5000000
@@ -165,7 +165,7 @@
 	}: Props = $props()
 	let s3FileDisplayRawMode = $state(false)
 	/** Which half of an agent result is showing; JSON is `forceJson`, as for any kind. */
-	let agentView: 'answer' | 'actions' = $state('answer')
+	let agentView: 'answer' | 'trace' = $state('answer')
 	/** The partial conversation a max-iterations failure carries, if this is one. */
 	let agentErrorMessages = $derived(parseAgentErrorMessages(result))
 
@@ -308,7 +308,7 @@
 				}
 
 				// Classified before the size caps below: an agent's answer stays small
-				// however long its conversation grows, so a run with many actions
+				// however long its conversation grows, so a run with a long trace
 				// must not fall back to the JSON tree that hides the answer inside it.
 				// `largeObject` is still set honestly, so switching to JSON gets the
 				// same too-big handling as any other oversized result.
@@ -818,15 +818,15 @@
 					{#if !hideAsJson && !['json', 's3object'].includes(resultKind ?? '') && typeof result === 'object'}<ToggleButtonGroup
 							selected={forceJson
 								? 'json'
-								: agentView === 'actions' && resultKind === 'aiagent'
-									? 'actions'
+								: agentView === 'trace' && resultKind === 'aiagent'
+									? 'trace'
 									: resultKind?.startsWith('table-')
 										? 'table'
 										: 'pretty'}
 							on:selected={(ev) => {
 								forceJson = ev.detail === 'json'
-								if (ev.detail === 'actions' || ev.detail === 'pretty') {
-									agentView = ev.detail === 'actions' ? 'actions' : 'answer'
+								if (ev.detail === 'trace' || ev.detail === 'pretty') {
+									agentView = ev.detail === 'trace' ? 'trace' : 'answer'
 								}
 							}}
 						>
@@ -835,7 +835,7 @@
 									<ToggleButton size="sm" value="table" label="Table" icon={Table2} {item} />
 								{:else if resultKind === 'aiagent'}
 									<ToggleButton size="sm" value="pretty" label="Answer" icon={Bot} {item} />
-									<ToggleButton size="sm" value="actions" label="Actions" icon={ListTree} {item} />
+									<ToggleButton size="sm" value="trace" label="Trace" icon={ListTree} {item} />
 								{:else}
 									<ToggleButton size="sm" value="pretty" label="Pretty" icon={Highlighter} {item} />
 								{/if}
@@ -1022,8 +1022,8 @@
 						     whole reason to look at such a run, so it is added under the error
 						     rather than replacing it. -->
 						<div class="flex flex-col gap-1 pt-4 w-full min-w-0">
-							<span class="text-emphasis text-xs font-semibold">Actions</span>
-							<AgentActions messages={agentErrorMessages} {workspaceId} />
+							<span class="text-emphasis text-xs font-semibold">Trace</span>
+							<AgentTrace messages={agentErrorMessages} {workspaceId} />
 						</div>
 					{/if}
 					{#if !isTest && language === 'bun'}

frontend/src/lib/components/agentTrace.test.ts

@@ -1,5 +1,5 @@
 import { describe, expect, it } from 'vitest'
-import { buildAgentActions } from './agentActions'
+import { buildAgentTrace } from './agentTrace'
 import type { AgentMessage } from './aiAgentResult'
 
 // The worker splits one tool call across two messages: the assistant message
@@ -34,9 +34,9 @@ const messages: AgentMessage[] = [
 	{ role: 'assistant', content: 'eu-central-1 is down.', agent_action: { type: 'message' } }
 ]
 
-describe('buildAgentActions', () => {
+describe('buildAgentTrace', () => {
 	it('joins a tool call to the arguments on the message that requested it', () => {
-		expect(buildAgentActions(messages)).toEqual([
+		expect(buildAgentTrace(messages)).toEqual([
 			{
 				kind: 'tool',
 				name: 'query_metrics',
@@ -49,7 +49,7 @@ describe('buildAgentActions', () => {
 	})
 
 	it('keeps an MCP call, whose arguments live on the action itself', () => {
-		const entries = buildAgentActions([
+		const entries = buildAgentTrace([
 			{
 				role: 'tool',
 				content: 'sunny',
@@ -74,7 +74,7 @@ describe('buildAgentActions', () => {
 	})
 
 	it('carries web search citations onto the entry', () => {
-		const entries = buildAgentActions([
+		const entries = buildAgentTrace([
 			{
 				role: 'assistant',
 				content: 'Postgres 17 changed the default.',
@@ -94,9 +94,9 @@ describe('buildAgentActions', () => {
 	// The prompt and the question are the step's inputs, shown as inputs. A replayed
 	// turn comes back from memory without its tag, and crediting this run with an
 	// answer a previous one gave would be a lie about what happened.
-	it('keeps only what this run did', () => {
+	it('traces only what this run did', () => {
 		expect(
-			buildAgentActions([
+			buildAgentTrace([
 				{ role: 'system', content: 'You are an SRE assistant.' },
 				{ role: 'user', content: 'Which region is broken?' },
 				{ role: 'assistant', content: 'Answered in an earlier turn, replayed from memory.' },

frontend/src/lib/components/agentTrace.ts

@@ -2,17 +2,16 @@ import type { WebSearchSource } from './copilot/chat/shared'
 import type { AgentMessage } from './aiAgentResult'
 
 /**
- * One thing an agent did. Built from the envelope alone: the tool arguments come
- * from the assistant message that asked for the call, and the result from the
- * `tool` message that answered it, so the list renders without waiting on any
- * request. A tool's child job is enrichment (logs, duration, whether it
- * succeeded), not what makes the row.
+ * One entry in the trace of an agent run. Built from the envelope alone: the tool
+ * arguments come from the assistant message that asked for the call, and the
+ * result from the `tool` message that answered it, so the trace renders without
+ * waiting on any request. A tool's child job is enrichment (logs, duration,
+ * whether it succeeded), not what makes the row.
  *
  * The prompt and the user's question are deliberately absent. They are inputs to
- * the step, shown as inputs, and a run is not a conversation the viewer is part
- * of — it is a record of what the agent did with them.
+ * the step and are shown as inputs; the trace is what the agent did with them.
  */
-export type AgentActionEntry =
+export type AgentTraceEntry =
 	| { kind: 'assistant'; content: string; sources?: WebSearchSource[] }
 	| { kind: 'search'; content: string; sources?: WebSearchSource[] }
 	| {
@@ -54,7 +53,7 @@ function sourcesOf(message: AgentMessage): WebSearchSource[] | undefined {
 	return sources.length > 0 ? sources : undefined
 }
 
-export function buildAgentActions(messages: AgentMessage[]): AgentActionEntry[] {
+export function buildAgentTrace(messages: AgentMessage[]): AgentTraceEntry[] {
 	// The arguments live on the assistant message that requested the call, while
 	// the action tag and the result live on the `tool` message answering it, so
 	// the two are joined by `tool_call_id`.
@@ -67,7 +66,7 @@ export function buildAgentActions(messages: AgentMessage[]): AgentActionEntry[]
 		}
 	}
 
-	const entries: AgentActionEntry[] = []
+	const entries: AgentTraceEntry[] = []
 	for (const message of messages) {
 		const action = message.agent_action
 		if (action?.type === 'tool_call') {