feat(web): collapse chat tool calls into one agent trace, reverse the activity rail

The chat rendered one card per tool call, so a 20-call turn buried the
answer under 20 stacked cards. Merge them with the "thinking" indicator
into a single collapsible strip above the answer:

- collapsed: the live activity while running, a count once finished
- expanded: the turn's work in humanized language (reuses the activity
  log's toolActivityLabel, so ten identical "run · target: host:strong"
  rows now read as what they actually did)
- per row: the raw args/result, one more click in

Also flip the Activity rail to newest-first with the current step on top:

- follow-mode/auto-scroll re-anchored to the top to match, or it would
  jump to the oldest entry on every new event
- pending plan steps park at the tail rather than sorting above the
  running step and pushing it off the top; the goal anchors the bottom

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

web/src/lib/components/AgentTrace.svelte

@@ -0,0 +1,115 @@
+<script lang="ts">
+  // The agent's working trace for one assistant turn: the live "thinking"
+  // indicator and that turn's tool calls merged into a single collapsible
+  // strip, instead of a stack of one card per call (a 13-call turn buried the
+  // actual answer). Collapsed it's one line — the current activity while
+  // running, a count once finished. Expanded it lists what the agent did, in
+  // humanized language, each row opening to its raw args/result.
+  import type { ToolCallResult } from '$lib/types'
+  import ToolCallCard from './ToolCallCard.svelte'
+  import Spinner from './Spinner.svelte'
+  import ChevronRightIcon from '@lucide/svelte/icons/chevron-right'
+  import CheckIcon from '@lucide/svelte/icons/check'
+  import XIcon from '@lucide/svelte/icons/x'
+  import SparklesIcon from '@lucide/svelte/icons/sparkles'
+
+  let {
+    tools = [],
+    label = null,
+    status = 'idle'
+  }: {
+    tools?: ToolCallResult[]
+    /** Live indicator text — the running step, an error, or "Done". */
+    label?: string | null
+    /** `idle` = no live state; the strip is just this turn's finished trace. */
+    status?: 'running' | 'done' | 'error' | 'idle'
+  } = $props()
+
+  let expanded = $state(false)
+
+  const count = $derived(tools.length)
+  // Collapsed line: prefer the live activity while something is happening,
+  // otherwise summarize the turn so a finished trace still says what it was.
+  const headline = $derived.by(() => {
+    if (status !== 'idle' && label) return label
+    if (count > 0) return count === 1 ? '1 tool call' : `${count} tool calls`
+    return 'No tool calls'
+  })
+</script>
+
+<div class="trace rounded-lg border border-border/60 bg-card/40 transition-colors" class:running={status === 'running'}>
+  <button
+    class="flex w-full items-center gap-2 px-3 py-1.5 text-left transition-colors hover:bg-muted/40"
+    onclick={() => (expanded = !expanded)}
+    aria-expanded={expanded}
+    aria-label={expanded ? 'Hide agent trace' : 'Show agent trace'}
+  >
+    <span class="shrink-0 {status === 'error' ? 'text-destructive' : status === 'idle' ? 'text-muted-foreground' : 'text-primary'}">
+      {#if status === 'running'}
+        <Spinner class="size-3" />
+      {:else if status === 'error'}
+        <XIcon class="size-3" />
+      {:else if status === 'done'}
+        <CheckIcon class="size-3" />
+      {:else}
+        <SparklesIcon class="size-3" />
+      {/if}
+    </span>
+
+    <span
+      class="min-w-0 flex-1 truncate text-xs {status === 'error'
+        ? 'text-destructive'
+        : status === 'running'
+          ? 'text-foreground/80'
+          : 'text-muted-foreground'}"
+    >
+      {headline}
+    </span>
+
+    {#if count > 0 && status !== 'idle'}
+      <span class="shrink-0 text-[10px] tabular-nums text-muted-foreground/60">{count}</span>
+    {/if}
+
+    <ChevronRightIcon
+      class="size-3 shrink-0 text-muted-foreground/50 transition-transform {expanded ? 'rotate-90' : ''}"
+    />
+  </button>
+
+  {#if expanded}
+    <div class="border-t border-border/40 p-1">
+      {#if count > 0}
+        {#each tools as tool (tool.id)}
+          <ToolCallCard {tool} />
+        {/each}
+      {:else}
+        <p class="px-2 py-1.5 text-[11px] text-muted-foreground">Nothing recorded for this turn yet.</p>
+      {/if}
+    </div>
+  {/if}
+</div>
+
+<style>
+  .trace {
+    animation: trace-in 0.2s ease-out;
+  }
+  /* A faint pulse while the agent is mid-turn — the collapsed strip is the
+     only thing on screen then, so it carries the "still working" signal. */
+  .trace.running {
+    border-color: color-mix(in oklab, var(--primary) 35%, var(--border));
+  }
+  @keyframes trace-in {
+    from {
+      opacity: 0;
+      transform: translateY(-2px);
+    }
+    to {
+      opacity: 1;
+      transform: translateY(0);
+    }
+  }
+  @media (prefers-reduced-motion: reduce) {
+    .trace {
+      animation: none;
+    }
+  }
+</style>

web/src/lib/components/ChatThread.svelte

@@ -9,12 +9,9 @@
   import type { Readable } from 'svelte/store'
   import { Button } from '$lib/components/ui/button'
   import { Textarea } from '$lib/components/ui/textarea'
-  import Spinner from './Spinner.svelte'
-  import ToolCallCard from './ToolCallCard.svelte'
+  import AgentTrace from './AgentTrace.svelte'
   import OperatorQuestion from './OperatorQuestion.svelte'
   import CornerDownLeftIcon from '@lucide/svelte/icons/corner-down-left'
-  import CheckIcon from '@lucide/svelte/icons/check'
-  import XIcon from '@lucide/svelte/icons/x'
   import RefreshCwIcon from '@lucide/svelte/icons/refresh-cw'
   import SquareIcon from '@lucide/svelte/icons/square'
   import { marked } from 'marked'
@@ -217,6 +214,16 @@
             </div>
             <div class="max-w-[85%] rounded-2xl rounded-br-sm bg-primary px-4 py-2.5 text-sm text-primary-foreground whitespace-pre-wrap user-msg">{msg.text}</div>
           {:else}
+            {@const isLast = idx === messages.length - 1}
+            {@const traceStatus = !isLast
+              ? 'idle'
+              : error
+                ? 'error'
+                : streaming
+                  ? 'running'
+                  : indicatorDone
+                    ? 'done'
+                    : 'idle'}
             <div class="flex w-full flex-col gap-2">
               <div class="flex items-baseline gap-2 px-1">
                 <span class="text-[10px] font-medium text-muted-foreground/70">Nomos</span>
@@ -224,34 +231,25 @@
                   <span class="text-[9px] text-muted-foreground/50">{formatTime(msg.created_at)}</span>
                 {/if}
               </div>
+              <!-- The working trace sits above the answer: it's what happened
+                   first, and collapsed it keeps a long tool run from burying
+                   the text below it. -->
+              {#if msg.tools.length > 0 || traceStatus !== 'idle'}
+                <AgentTrace
+                  tools={msg.tools}
+                  status={traceStatus}
+                  label={traceStatus === 'idle' ? null : indicatorLabel}
+                />
+              {/if}
               {#if msg.text}
                 <div class="prose-chat max-w-none text-sm leading-relaxed assistant-msg">
                   <!-- eslint-disable-next-line svelte/no-at-html-tags — sanitized via DOMPurify -->
                   {@html render(msg.text)}
-                  {#if idx === messages.length - 1 && streaming}
+                  {#if isLast && streaming}
                     <span class="stream-cursor" aria-hidden="true"></span>
                   {/if}
                 </div>
               {/if}
-              {#if msg.tools.length > 0}
-                <div class="flex flex-col gap-1.5">
-                  {#each msg.tools as tool (tool.id)}
-                    <ToolCallCard {tool} />
-                  {/each}
-                </div>
-              {/if}
-              {#if idx === messages.length - 1 && msg.text === '' && (streaming || indicatorDone || error)}
-                <div class="flex items-center gap-2 py-1 text-xs {error ? 'text-destructive' : indicatorDone ? 'text-primary' : 'text-muted-foreground'}">
-                  {#if error}
-                    <XIcon class="size-3 shrink-0" />
-                  {:else if indicatorDone}
-                    <CheckIcon class="size-3 shrink-0" />
-                  {:else}
-                    <Spinner class="size-3 shrink-0 text-primary" />
-                  {/if}
-                  <span>{indicatorLabel}</span>
-                </div>
-              {/if}
             </div>
           {/if}
         </div>

web/src/lib/components/ToolCallCard.svelte

@@ -1,6 +1,11 @@
 <script lang="ts">
-  import { Check, ChevronRight, Loader2, Wrench, X } from '@lucide/svelte'
+  // One tool call inside AgentTrace's expanded list. Renders as a borderless
+  // row (the trace supplies the container/border) whose own click reveals the
+  // raw args/result — so the trace stays a readable thinking log by default
+  // and the JSON is one more click away, not stacked inline.
+  import { Check, ChevronRight, Loader2, X } from '@lucide/svelte'
   import type { ToolCallResult } from '$lib/types'
+  import { toolActivityLabel } from '$lib/stores/activity'
 
   let { tool }: { tool: ToolCallResult } = $props()
   let expanded = $state(false)
@@ -11,11 +16,7 @@
     return 'done'
   })
 
-  const statusColor = $derived.by(() => {
-    if (status === 'running') return 'text-primary'
-    if (status === 'error') return 'text-destructive'
-    return 'text-primary'
-  })
+  const label = $derived(toolActivityLabel(tool))
 
   const argsSummary = $derived.by(() => {
     if (!tool.args) return ''
@@ -25,61 +26,62 @@
     const val = typeof first[1] === 'string' ? first[1] : JSON.stringify(first[1])
     return `${first[0]}: ${val.length > 60 ? val.slice(0, 60) + '…' : val}`
   })
+
+  const hasDetail = $derived(
+    !!tool.args || (tool.result !== undefined && tool.result !== null) || !!tool.error
+  )
 </script>
 
-<div class="tool-card rounded-lg border border-border/60 bg-card/40 overflow-hidden transition-all">
+<div class="tool-row">
   <button
-    class="flex w-full items-center gap-2 px-3 py-2 text-left hover:bg-muted/40 transition-colors"
+    class="flex w-full items-start gap-2 rounded-md px-2 py-1.5 text-left transition-colors hover:bg-muted/40 disabled:cursor-default"
     onclick={() => (expanded = !expanded)}
     aria-expanded={expanded}
+    disabled={!hasDetail}
   >
-    <ChevronRight class="size-3 shrink-0 text-muted-foreground transition-transform {expanded ? 'rotate-90' : ''}" />
-    <Wrench class="size-3.5 shrink-0 text-muted-foreground" />
-    <span class="font-mono text-xs font-medium text-foreground/80">{tool.name}</span>
-    {#if argsSummary}
-      <span class="ml-1 truncate text-[11px] text-muted-foreground/70">{argsSummary}</span>
-    {/if}
-    <span class="ml-auto shrink-0 {statusColor}">
+    <span class="mt-px shrink-0 {status === 'error' ? 'text-destructive' : 'text-primary'}">
       {#if status === 'running'}
-        <Loader2 class="size-3.5 animate-spin" />
+        <Loader2 class="size-3 animate-spin" />
       {:else if status === 'error'}
-        <X class="size-3.5" />
+        <X class="size-3" />
       {:else}
-        <Check class="size-3.5" />
+        <Check class="size-3" />
       {/if}
     </span>
+    <span class="min-w-0 flex-1">
+      <span class="block truncate text-xs text-foreground/90">{label}</span>
+      {#if argsSummary}
+        <span class="block truncate font-mono text-[10px] text-muted-foreground/60">{argsSummary}</span>
+      {/if}
+    </span>
+    <span class="shrink-0 font-mono text-[10px] text-muted-foreground/50">{tool.name}</span>
+    {#if hasDetail}
+      <ChevronRight
+        class="mt-px size-3 shrink-0 text-muted-foreground/50 transition-transform {expanded ? 'rotate-90' : ''}"
+      />
+    {/if}
   </button>
 
   {#if expanded}
-    <div class="border-t border-border/40 px-3 py-2 space-y-2">
+    <div class="space-y-2 px-2 pb-2 pl-7">
       {#if tool.args}
         <div>
-          <div class="text-[10px] font-semibold uppercase tracking-wider text-muted-foreground mb-1">Args</div>
-          <pre class="tool-pre rounded-md bg-muted/60 p-2 text-[11px] overflow-x-auto max-h-48">{JSON.stringify(tool.args, null, 2)}</pre>
+          <div class="mb-1 text-[10px] font-semibold uppercase tracking-wider text-muted-foreground">Args</div>
+          <pre class="max-h-48 overflow-x-auto rounded-md bg-muted/60 p-2 text-[11px]">{JSON.stringify(tool.args, null, 2)}</pre>
         </div>
       {/if}
       {#if tool.result !== undefined && tool.result !== null}
         <div>
-          <div class="text-[10px] font-semibold uppercase tracking-wider text-muted-foreground mb-1">Result</div>
-          <pre class="tool-pre rounded-md bg-muted/60 p-2 text-[11px] overflow-x-auto max-h-48">{JSON.stringify(tool.result, null, 2)}</pre>
+          <div class="mb-1 text-[10px] font-semibold uppercase tracking-wider text-muted-foreground">Result</div>
+          <pre class="max-h-48 overflow-x-auto rounded-md bg-muted/60 p-2 text-[11px]">{JSON.stringify(tool.result, null, 2)}</pre>
         </div>
       {/if}
       {#if tool.error}
         <div>
-          <div class="text-[10px] font-semibold uppercase tracking-wider text-destructive mb-1">Error</div>
-          <pre class="tool-pre rounded-md bg-destructive/5 border border-destructive/20 p-2 text-[11px] text-destructive overflow-x-auto max-h-48">{tool.error}</pre>
+          <div class="mb-1 text-[10px] font-semibold uppercase tracking-wider text-destructive">Error</div>
+          <pre class="max-h-48 overflow-x-auto rounded-md border border-destructive/20 bg-destructive/5 p-2 text-[11px] text-destructive">{tool.error}</pre>
         </div>
       {/if}
     </div>
   {/if}
 </div>
-
-<style>
-  .tool-card {
-    animation: tool-in 0.2s ease-out;
-  }
-  @keyframes tool-in {
-    from { opacity: 0; transform: translateY(-2px); }
-    to { opacity: 1; transform: translateY(0); }
-  }
-</style>

web/src/lib/components/UnifiedTimeline.svelte

@@ -16,6 +16,9 @@
   import FlagIcon from '@lucide/svelte/icons/flag'
 
   // Merged plan + activity timeline, designed for the narrow rail:
+  // - ordered newest-first: what the agent is doing right now is at the top,
+  //   history flows downward, and the goal sits at the bottom where the task
+  //   began (see the sort in `items`)
   // - one continuous vertical "backbone"; every item owns a segment of it,
   //   colored by state (done = filled primary, running = faint primary,
   //   pending/future = muted) so the line visibly fills in as work completes
@@ -25,7 +28,8 @@
   //   markers on the same backbone
   // - the running step auto-expands and the view auto-scrolls to keep the
   //   current step visible while the agent works (follow mode disengages if
-  //   the operator scrolls up, re-engages when streaming starts again)
+  //   the operator scrolls down into history, re-engages when streaming
+  //   starts again)
   let { entries, planSteps: steps, streaming = false }: {
     entries: ActivityEntry[]
     planSteps: PlanStep[]
@@ -63,9 +67,10 @@
     for (const s of steps) {
       if (s.status === 'pending' && !entries.some((e) => e.stepSeq === s.seq)) {
         // Pending steps with no activity yet still show on the timeline so
-        // the operator sees what's coming — but only if a plan exists.
+        // the operator sees what's coming — but only if a plan exists. ts 0
+        // parks them at the tail of the newest-first sort below (see there).
         if (steps.length > 0) {
-          out.push({ kind: 'step', step: s, tools: [], ts: Number.MAX_SAFE_INTEGER - s.seq })
+          out.push({ kind: 'step', step: s, tools: [], ts: 0 })
         }
         continue
       }
@@ -73,8 +78,11 @@
         (e) => e.stepSeq === s.seq && (e.type === 'tool_running' || e.type === 'tool_done' || e.type === 'tool_error')
       )
       const stepEntry = entries.find((e) => e.id === s.id)
+      // Timed from the step's own entry, else its earliest tool — so a step
+      // is placed by when it started, not by its latest activity.
       const ts = stepEntry?.timestamp ?? tools[0]?.timestamp ?? Date.now()
-      out.push({ kind: 'step', step: s, tools, ts })
+      // Tools inside a step run newest-first too, matching the outer order.
+      out.push({ kind: 'step', step: s, tools: [...tools].reverse(), ts })
     }
 
     for (const e of entries) {
@@ -84,7 +92,15 @@
       out.push({ kind: 'entry', entry: e, ts: e.timestamp })
     }
 
-    out.sort((a, b) => a.ts - b.ts)
+    // Newest first: whatever the agent is doing right now sits at the top of
+    // the rail, with history flowing downward. The two ts-0 groups fall to
+    // the bottom for free, which is where both belong in this order: the goal
+    // (timestamp 0 — where the task started) and not-yet-run plan steps.
+    // Sorting the latter by their future position would put them *above* the
+    // running step and push it off the top, which is exactly what this
+    // ordering exists to prevent. Array.sort is stable, so each group keeps
+    // its insertion order (plan steps in seq order).
+    out.sort((a, b) => b.ts - a.ts)
     return out
   })
 
@@ -100,9 +116,11 @@
   let container = $state<HTMLDivElement | null>(null)
   let follow = $state(true)
 
+  // Newest-first, so "following the agent" means being parked at the top —
+  // the mirror of the bottom-anchored follow this had when it ran oldest-first.
   function onScroll() {
     if (!container) return
-    follow = container.scrollHeight - container.scrollTop - container.clientHeight < 80
+    follow = container.scrollTop < 80
   }
 
   // A new turn re-engages follow mode even if the operator had scrolled up.
@@ -129,7 +147,7 @@
       ? container.querySelector(`[data-tl-id="${CSS.escape(currentId)}"]`)
       : null
     if (target) target.scrollIntoView({ behavior: 'auto', block: 'nearest' })
-    else container.scrollTop = container.scrollHeight
+    else container.scrollTop = 0
   })
 
   // ── Presentation helpers ──────────────────────────────────────────────────

web/src/lib/stores/activity.ts

@@ -185,7 +185,10 @@ export function activityLogFor(sessionId: string): Readable<ActivityEntry[]> {
   return derived([chat.messages, ws.planSteps, task], ([$msgs, $steps, $task]) => computeActivityLog($msgs, $steps, $task))
 }
 
-function toolActivityLabel(t: ToolCallResult): string {
+// Humanized, past/present-tense description of what a tool call is doing
+// ("Check execution", "Research: …") rather than its raw wire name. Exported
+// so the chat's agent trace can read as a thinking log instead of an API log.
+export function toolActivityLabel(t: ToolCallResult): string {
   const args = t.args ?? {}
   const str = (v: unknown): string => typeof v === 'string' ? v : ''
   switch (t.name) {