agentgui 1.0.1056 → 1.0.1058

package/.gm/memories/mem-c4aaec5c4e6f80c2-920.md

@@ -0,0 +1,8 @@
+---
+key: mem-c4aaec5c4e6f80c2-920
+ns: default
+created: 1783884935543
+updated: 1783884935543
+---
+
+{"key":"agentgui-ds-load-local-vendor-detail","value":"agentgui DS (design kit) loading: index.html loads the DS kit from local ./vendor/anentrypoint-design/247420.{js,css}, NOT unpkg. Prior 'moved to unpkg @latest' note (2026-05-29) was a transient state reverted in eb1eab3 (re-vendored with AgentChat), witnessed loading from ./vendor/ again on the 2026-06-04/06-05 runs and reconfirmed live on the 37th (2026-07-12) and 40th (2026-07-12) runs via live network-trace browser dispatch. The vendored copy IS the shipped UI - predictable, no upstream-publish drift. Update flow: edit the kit at /config/workspace/design, node scripts/build.mjs, copy dist/247420.{js,css} into site/app/vendor/anentrypoint-design/, browser-witness, then push the kit so unpkg stays in sync for other consumers. The markdown stack (marked/dompurify/prismjs) still fetches from jsdelivr on first chat render, unaffected by this vendoring."}

package/.gm/memories/mem-eae1e383927854fb-2649.md

@@ -0,0 +1,8 @@
+---
+key: mem-eae1e383927854fb-2649
+ns: default
+created: 1783884912840
+updated: 1783884912840
+---
+
+{"key":"agentgui-40th-run-gui-logic-predictability-sweep","value":"agentgui 40th run (2026-07-12): gui-logic-predictability workflow wf_c0a8d954-547 (20 agents, 196 tool calls, 0 errors) - 11 confirmed findings, all implemented. App (site/app/js/app.js, backend.js): global Escape ladder gained confirmingClearData disarm (settings clear-local-data confirm was the one armed state missing from the ladder); History detail header title/lede now use pathBasename(cwd) fallback instead of raw path, matching every other headline surface; wsCall gained a 15s timeout+cleanup (previously a hung WS RPC left promises pending forever with zero indicator); agy (Antigravity) resumeSid now forwards true once past first turn so its already-implemented --continue path actually fires (was hardcoded to only forward for claude-code), and the 'does not resume' banner now excludes agy since it has real continuity; pastedFilesToChat now calls the newly-exported kit flashComposerNote for visible upload-progress/failure/success feedback plus an undo-history-loss note. Kit (design/src/components/{chat,content}.js, app-shell.css, chat.css, components.js barrel): Row's primary .title span gained a title= tooltip fallback (39th run only fixed .sub); .ds-file-actions resting opacity raised 0->0.35 for desktop-pointer discoverability without hover; chat-composer-hint now visible at rest (was focus/draft-gated) and no longer aria-hidden; insertEmoji gained a one-time sessionStorage-gated undo-loss note matching draft-restore's existing pattern; flashComposerNote exported through the components.js barrel. Kit built+tested clean, rebuilt, re-vendored. Kit pushed 541df4d (npm-publish auto-bump landed as 92422d8), agentgui pushed 53610a4 then 4a5d6e5 (PRD-resolve residual), all 4 CI checks green both repos. Browser-witnessed every finding live on buildesk using a NEW reliable pattern: combine setHTTPCredentials+goto+wait+evaluate into ONE dispatch (splitting across dispatches hits session-id rotation and reads a genuinely-blank document.body despite window.__agentgui resolving), and read results via console.log('WITNESS:'+JSON.stringify(...)) + grep stdout/debug.console rather than the capture prefix's result field (confirmed still returns undefined even though the evaluated code's side effects and console output are real). All witnesses: confirmingClearData Escape before:true/after:false; composer-hint display:block no aria-hidden; .ds-file-actions opacity 0.35; 16/16 Row .title elements carry title= in settings agents panel; deployed backend.js/app.js/247420.js curl-confirmed to contain every shipped fix string; 0 pageErrors throughout."}

package/.gm/prd.yml

@@ -3076,3 +3076,9 @@
   - gm-plugkit/playwriter browser verb dom=/capture eval context defect - requires an upstream tool fix outside agentgui/design
   status: completed
   witness: 'External tool defect, not resolvable from this repo/session: the gm-plugkit/playwriter browser verb''s dom= and capture prefixes evaluate in a Node vm context with no page/document binding (ReferenceError on every attempt this session), unlike the screenshot= and url=<target> paths which do reach the real page. This is a defect in the upstream gm-plugkit/playwriter tool package itself, outside agentgui and design repo scope. Recorded per the false-completion rule as blockedBy:external rather than marked done - screenshots remain the working witness fallback and were used for all browser verification in this sweep.'
+- id: 41st-run-verify-buildesk-live
+  subject: Live-verify all confirmed+implemented findings on https://buildesk.acc.l-inc.co.za/gm via browser witness using the reliable single-dispatch creds+goto+wait+evaluate pattern
+  status: pending
+- id: 41st-run-push-ci-green
+  subject: Push kit and agentgui changes, verify CI green on both repos before COMPLETE
+  status: pending

package/AGENTS.md

@@ -1,5 +1,9 @@
 # AgentGUI — Agent Notes
 
+## GUI logic+predictability sweep (2026-07-12) — forty-first run
+
+`gui-logic-predictability` workflow wf_c645ba95-c6a (23 agents, 192 tool calls, 0 errors) — 10 confirmed findings, all implemented, plus one real kit bug found and fixed mid-verification (not from the workflow's own hunt). App (`site/app/js/app.js`): `runningPanel()`'s title read `'running · N'` with no scoping cue even though the panel always lists every in-flight session app-wide, not just ones related to the current chat — retitled `'all running sessions · N'`, and it now appends `'(refreshing…)'` while a backgrounded tab's poll is catching up after refocus (`refreshActive()` sets a `state._activeStale` flag while `document.hidden` skips the network call, cleared on the next real response) instead of silently painting stale data as fresh. `runningRowTitle()` fell straight to `UNTITLED_CONVERSATION` for a brand-new session with no index row yet, while the Live dashboard's `computeLiveSessions()` already had a 3s "indexing…" carve-out for the identical case — the same session read two contradictory identities in the same 0-3s window depending only on which panel you looked at; gave `runningRowTitle` the matching carve-out. Live dashboard cards sorted by `elapsed`/`activity` re-sort on every tick (both are continuously-changing values), which could reorder a card out from under a mid-click pointer (selecting a checkbox, clicking stop) — added a module-level `_livePointerDown` flag (set/cleared on document `pointerdown`/`pointerup`/`pointercancel`) that freezes the sort to the last-computed sid order while a pointer is held anywhere in the dashboard. agy's real `--continue` resume path (wired last run) had no composer disclosure the way claude-code's resume banner does — added a `'continues most recent conversation'` composerContext bit for agy past its first turn. History's `ConversationList` sessions mapper hardcoded `agent: undefined` while the Running-panel/Live-dashboard rows for the identical finished-vs-in-flight session both show an agent/model badge — traced to ccsniff's `Store.sessions()` never folding per-event `model` into the session record even though `flattenEvent` captures it; fixed upstream (see below) and wired `agent: agentById('claude-code').name + ' · ' + s.model` into the mapper (ccsniff only reads Claude Code's own JSONL, so the agent is always constant). Kit (`design/src/components/{chat,content}.js`, `design/chat.css`, `design/src/components/sessions.js`): per-message chat copy action (`ChatMessage`'s generic `actionRow`) had zero visible click feedback for sighted/mouse users — only an `aria-live` `announce()` — unlike the three sibling copy patterns already in the same file (code-block copy, `CodeNode`/`ToolCallNode` copy) which all self-manage a label/icon flip to "copied" for 1.6s; added the matching flip keyed off `a.label === 'copy'`. Follow-up chips (contextual, derived from the last turn) and empty-state seed chips (generic starter ideas) shared identical DOM shape/class with zero visual signal distinguishing them — `.agentchat-followup` gained a dashed border (same pill shape/size, no new color/shape language). `flashComposerNote`'s single shared DOM node silently overwrote an unread note if a second call landed before the first note's 2.6s timeout — now queues notes and drains them in order instead of dropping the first message. `SessionDashboard`'s filter `SearchInput` never forwarded `resultCount` to its `aria-live` region (the Live dashboard's own filtered session count was available but unused) — wired it through. **That last change surfaced a real, previously-latent kit bug independent of this run's own hunt: `SearchInput` (`design/src/components/content.js`) conditionally returned either a bare `<input>` VElement or a wrapping `<span>` VElement depending on whether `resultCount`/a clear button were present that render — since `resultCount` now toggles from `undefined` to a string as the Live filter's value changes, webjsx's `applyDiff` tried to morph one element type into another at the same keyed toolbar slot, producing a corrupted merged DOM node (a real `<select>` picked up `type=search`/`name=q`/`placeholder=...` attributes from the `SearchInput` render, and the actual `<input>` vanished entirely) — live-witnessed on buildesk before the fix (`hasInput:false` on the toolbar). Fixed by always returning the `.ds-search-input-wrap` shape (confirmed `display:contents` in CSS, so it's layout-transparent and safe for every existing call site).** ccsniff (separate repo, `github:AnEntrypoint/ccsniff#main`, cloned fresh to `/config/workspace/ccsniff-work`): `Store.sessions()` (`src/store.js`) folded in the most-recent-by-timestamp `model` per session (events aren't guaranteed in ts order since JSONL is read file-by-file); `index.js`'s assistant-event ingestion never actually populated `model` on pushed content blocks (only `system`/init events carried `e.model`) — real Claude Code JSONL carries the actually-used model at `message.model` on every assistant turn, so that's now inherited onto every block from that message. `node test.js`: ALL TESTS PASS both commits. Pushed `11f6c3b` then (after a remote-advanced rebase) `26bbfcc`/`beff5e7`, ccsniff CI (Publish/Auto-Declaudeify/Deploy GH Pages) green both pushes. Kit built+tested twice (once per round of edits — findings, then the SearchInput fix), re-vendored both times. Reinstalling the ccsniff dependency required removing `bun.lock`'s pinned commit hash entirely (`bun install ccsniff@github:...#main --force` alone kept resolving to the stale pinned sha even after a `rm -rf node_modules/ccsniff`) — deleting `bun.lock` and re-running `bun install` re-resolved `#main` to the actual latest commit. The local buildesk server process needed an explicit restart (`kill -TERM` on the `bun server.js` PID) for the new `node_modules/ccsniff` to load — Node/Bun caches modules in-process, so an npm/git dependency swap under a running process is invisible until restart; the env's custom Node supervisor (`/opt/gmweb-startup`, not a standard supervisord) auto-respawned it within ~20s of the health-check failing, re-launching via `bunx agentgui@latest` which correctly resolves to the local workspace's own `bin/gmgui.cjs` in this workspace-linked setup (confirmed by the "Workspace mode: skipping npm version checker" startup log line), not the published npm package. Browser-witnessed every finding live on buildesk: `.ds-dash-toolbar` filter input/select intact before AND after typing with the corruption fixed, `aria-live` region reading `'0 results'`; `.chat-msg-action.is-copied`/`.agentchat-followup` CSS rules present in the live CSSOM; History rail rows reading `'agentgui · 3s ago · Claude Code · claude-sonnet-5'` (18/18 sessions carrying a real model after the ccsniff fix + server restart); synthetic `pointerdown` dispatch confirmed no page error from the new listener. 0 pageErrors throughout.
+
 ## GUI logic+predictability sweep (2026-07-12) — fortieth run
 
 `gui-logic-predictability` workflow wf_c0a8d954-547 (20 agents, 196 tool calls, 0 errors) — 11 confirmed findings, all implemented; the lens is still surfacing real gaps at run 40, contrary to the 39th run's thinning-yield note. App (`site/app/js/app.js`, `site/app/js/backend.js`): the global Escape ladder disarmed `confirmingEdit`/`confirmingNewChat`/`confirmingStopAll`/`confirmingStopSelected` but never `confirmingClearData` (settings' arm-then-confirm "clear all local data?" banner) — a keyboard user who armed it and pressed the ASCII-standard cancel gesture this app trains everywhere else got no response; added the matching disarm branch. History detail header title/lede fell back to the raw full `cwd` path instead of `pathBasename(cwd)` like every other headline surface (chat cwd label, files breadcrumb, resume banner) — wrapped both fallback sites; the unabbreviated path stays available via the labeled `SessionMeta` "directory" row's own `title=` tooltip. `wsCall` (the WS RPC helper every `backend.js` call goes through) had no timeout at all — a hung `chat.active`/any other response left its promise pending forever with zero indicator; added a 15s timeout that rejects and clears the pending map, `resolve`/`reject` wrapped to clear the timer on a real reply. agy (Antigravity) has a working `--continue` (most-recent-conversation) resume path server-side (`lib/claude-runner-agents.js`) that the client never triggered — `resumeSid` was hardcoded to only forward for claude-code, so agy always got the flattened-transcript preamble instead of true continuity, AND the "does not resume across turns" banner fired for agy exactly like the genuinely-stateless opencode/kilo/codex, misattributing a missing client wire as an agent limitation; now `resumeSid` forwards `true` for agy once the chat is past its first turn (agy's `--continue` ignores the sid value, boolean-truthy is sufficient) and the banner excludes agy. Pasted/dropped files into the composer only fired a screen-reader-only `announce()` during upload — zero visible feedback for sighted mouse users on progress or failure; now calls the kit's `flashComposerNote` (newly exported) on the real composer element for upload-start/failure/success, and the success announce text now discloses the undo-history caveat inline. Kit (`design/src/components/{chat,content}.js`, `design/app-shell.css`, `design/chat.css`, `design/src/components.js`): `Row`'s primary `.title` span had no `title=` tooltip fallback (only `.sub` got one in the 39th run) — the more commonly-truncated primary label was the bigger gap; fixed at the one shared render site so every `Row` consumer (settings agents panel, etc.) inherits it. `.ds-file-actions` (Files row rename/move/delete icons) were `opacity:0` by default on desktop mouse/trackpad pointers, fully invisible without an accidental hover — raised the resting baseline to `0.35` so the actions are discoverable without hover, full `1` still on hover/focus-within. The Enter-to-send composer hint was CSS-hidden except while focused or carrying a draft (and `aria-hidden`), so a brand-new user looking at an empty, unfocused composer had no in-UI way to learn Enter-vs-Shift+Enter — now visible at rest above 420px (still hidden under 420px to save rows) and no longer `aria-hidden` since it's real content, not decoration. `insertEmoji`'s direct `taEl.value=` assignment discards the native undo stack the same way draft-restore's `.value=` does, but only draft-restore surfaced an explanatory note — added a matching one-time (`sessionStorage`-gated) `flashComposerNote` call on first emoji insert per session. `flashComposerNote` exported from `chat.js` through the `components.js` barrel (33rd-run standing lesson: a kit function must clear the barrel to be consumable by the app). Kit built+tested (`bun test.js` all pass), rebuilt (`node scripts/build.mjs`, 0 lint errors), re-vendored into `site/app/vendor/anentrypoint-design/`. Browser-witnessed live on buildesk via the now-standard `page.context().setHTTPCredentials()` + single-dispatch (creds+goto+wait+evaluate, same session) pattern: `confirmingClearData` Escape-disarm before:true/after:false; composer hint `display:block` with `aria-hidden` absent; `.ds-file-actions` computed `opacity:"0.35"`; settings agents-panel `Row` `.title` elements 16/16 carry a `title=` attribute; deployed `backend.js`/`app.js`/`247420.js` all curl-confirmed to contain the shipped fix strings (`timeoutMs = 15000`, the agy banner-exclusion comment, `pathBasename(sess?.cwd)`, `undoNoteShown`, `ds-file-actions`); 0 pageErrors across every dispatch. **Reconfirmed standing tool defect, worse than previously documented: browser session ids silently rotate across SEPARATE dispatches even against the "same" nominal session, and a dispatch landing on a stale/rotated session reads an entirely blank `document.body` (0 children) despite `window.__agentgui` still resolving — the fix is to always combine credentials-set + goto + wait + evaluate into ONE single dispatch rather than splitting navigation and reading across dispatches.** Also reconfirmed the `capture` prefix's `page.evaluate` return value is NOT propagated back through the tool's `result` field (comes back `undefined` even though the evaluated expression itself runs correctly and its side effects land) — the reliable read pattern is `console.log("WITNESS:" + JSON.stringify(...))` inside the evaluated callback and grep the dispatch's `stdout`/`debug.console`, not the `result` field.
@@ -284,7 +288,7 @@ GUI source keeps typographic product characters — the middot separator `·`, t
 
 ## DS load = LOCAL vendor (resolved — the earlier unpkg note is stale)
 
-**`index.html` loads the DS kit from the local `./vendor/anentrypoint-design/247420.{js,css}`, NOT unpkg.** The prior "moved to unpkg @latest" note described a transient state that was reverted: the kit was re-vendored (with `AgentChat`) in `eb1eab3` and witnessed loading from `./vendor/` again (2026-06-04 and 2026-06-05 runs). The vendored copy IS the shipped UI — predictable, no upstream-publish drift. Update flow: edit the kit at `c:\dev\anentrypoint-design`, `node scripts/build.mjs`, copy `dist/247420.{js,css}` into `site/app/vendor/anentrypoint-design/`, browser-witness, then push the kit so unpkg stays in sync for other consumers. (The markdown stack marked/dompurify/prismjs still fetches from jsdelivr on first chat render.)
+`index.html` loads the DS kit from local `./vendor/anentrypoint-design/247420.{js,css}`, NOT unpkg — reconfirmed live on the 40th run. Full detail in rs-learn (recall "agentgui ds load local vendor detail").
 
 
 @.gm/next-step.md

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agentgui",
-  "version": "1.0.1056",
+  "version": "1.0.1058",
   "description": "Multi-agent ACP client with real-time communication",
   "type": "module",
   "main": "electron/main.js",

package/site/app/js/app.js

@@ -358,10 +358,13 @@ function activeSig(list) {
 async function refreshActive() {
   // A backgrounded tab still holds its interval timer, but polling every ~3.3s
   // while hidden is needless server load across many left-open tabs - skip the
-  // network call (and let the next visible tick catch up).
-  if (typeof document !== 'undefined' && document.hidden) return;
+  // network call (and let the next visible tick catch up). Mark the currently
+  // held state.active as stale so a refocus render doesn't silently paint
+  // minutes-old data as fresh while the re-poll is still in flight.
+  if (typeof document !== 'undefined' && document.hidden) { state._activeStale = true; return; }
   let next;
   try { next = await B.listActiveChats(state.backend); } catch { return; }
+  state._activeStale = false;
   const changed = activeSig(next) !== activeSig(state.active);
   state.active = next;
   if (changed) state._sessionGroupsCache = null;
@@ -845,7 +848,13 @@ function sessionsColumn() {
       // carries only the time (and agent when present).
       project: project === title ? '' : project,
       time: fmtRelTime(s.last),
-      agent: undefined,
+      // ccsniff only reads Claude Code's own JSONL (see AGENTS.md), so the
+      // agent is always claude-code; ccsniff's sessions() now also folds in
+      // the session's most-recent model - surfacing both here matches the
+      // badge the running-panel/live-dashboard rows already show for the
+      // identical session while it's in-flight, instead of going blank the
+      // moment it finishes and drops into History.
+      agent: (agentById('claude-code')?.name || 'Claude Code') + (s.model ? ' · ' + s.model : ''),
       running: runningSids.has(s.sid),
       status: runningSids.has(s.sid) ? liveStatusBySid.get(s.sid) : undefined,
       unread: false,
@@ -1724,6 +1733,14 @@ async function stopAllActive(sessions) {
 // before it is treated as STALE - alive but not making progress, so a stuck
 // agent reads differently from a busy one.
 const STALE_AFTER_MS = 45000;
+// Tracks whether a pointer is currently held down anywhere in the document,
+// used to freeze Live-dashboard card sort order mid-click (see liveMain()).
+let _livePointerDown = false;
+if (typeof document !== 'undefined') {
+  document.addEventListener('pointerdown', () => { _livePointerDown = true; });
+  document.addEventListener('pointerup', () => { _livePointerDown = false; });
+  document.addEventListener('pointercancel', () => { _livePointerDown = false; });
+}
 // Rank for the error-then-stale-first ordering (W3).
 const STATUS_RANK = { error: 0, stale: 1, running: 2 };
 
@@ -1879,15 +1896,26 @@ function liveMain() {
   // external, and a deterministic sid tiebreaker so equal-rank cards never
   // reshuffle with the server's return order.
   const sortKey = lv.sort || 'status';
-  sessions.sort((a, b) => {
-    let d = (a.external ? 1 : 0) - (b.external ? 1 : 0);
-    if (d) return d;
-    if (sortKey === 'elapsed') d = (b.elapsedMs || 0) - (a.elapsedMs || 0);
-    else if (sortKey === 'activity') d = (b.lastTs || 0) - (a.lastTs || 0);
-    else if (sortKey === 'errors') d = (b.errors || 0) - (a.errors || 0);
-    else d = (STATUS_RANK[a.status] ?? 9) - (STATUS_RANK[b.status] ?? 9);
-    return d || String(a.sid).localeCompare(String(b.sid));
-  });
+  // 'elapsed'/'activity' sort keys are continuously-ticking values (they change
+  // every second), so re-sorting on every render can reorder a card out from
+  // under a mid-click pointer (selecting a checkbox, clicking stop). While a
+  // pointer is held down anywhere in the dashboard, freeze the sort to the
+  // last-computed sid order instead of live-recomputing it.
+  if (_livePointerDown && state._liveSortedOrder && state._liveSortedOrder.length) {
+    const rank = new Map(state._liveSortedOrder.map((sid, i) => [sid, i]));
+    sessions.sort((a, b) => (rank.has(a.sid) ? rank.get(a.sid) : 1e9) - (rank.has(b.sid) ? rank.get(b.sid) : 1e9));
+  } else {
+    sessions.sort((a, b) => {
+      let d = (a.external ? 1 : 0) - (b.external ? 1 : 0);
+      if (d) return d;
+      if (sortKey === 'elapsed') d = (b.elapsedMs || 0) - (a.elapsedMs || 0);
+      else if (sortKey === 'activity') d = (b.lastTs || 0) - (a.lastTs || 0);
+      else if (sortKey === 'errors') d = (b.errors || 0) - (a.errors || 0);
+      else d = (STATUS_RANK[a.status] ?? 9) - (STATUS_RANK[b.status] ?? 9);
+      return d || String(a.sid).localeCompare(String(b.sid));
+    });
+    state._liveSortedOrder = sessions.map(s => s.sid);
+  }
   const selected = lv.selected instanceof Set ? lv.selected : new Set();
   // The in-page chat's card accent matches on the real sid, but the kit
   // compares activeSid against card.sid (the ephemeral id for owned cards).
@@ -2230,7 +2258,9 @@ function chatMain() {
           userTurnCount > 0 ? plural(userTurnCount, 'turn') : null,
           (state.chat.resumeSid && state.selectedAgent === 'claude-code')
             ? 'continues conversation (' + resumeSidLabel(state.chat.resumeSid) + ')'
-            : null,
+            : (state.selectedAgent === 'agy' && userTurnCount > 0)
+              ? 'continues most recent conversation'
+              : null,
         ].filter(Boolean),
       } : undefined,
       // W15: contextual follow-up chips derived from the settled last turn
@@ -3164,6 +3194,13 @@ function resumeSidLabel(sid) {
 // everywhere else.
 function runningRowTitle(r) {
   const realSid = r.claudeSessionId || r.sessionId;
+  const arr = Array.isArray(state.sessions) ? state.sessions : [];
+  const sess = arr.find((s) => s.sid === realSid);
+  // A brand-new session (started <3s ago) with no index row yet is still
+  // BEING indexed, not permanently untitled - same carve-out and threshold
+  // computeLiveSessions() uses, so the identical session doesn't read
+  // "Untitled conversation" here and "indexing…" on the Live dashboard.
+  if (!sess && r.startedAt && (Date.now() - r.startedAt < 3000)) return 'indexing…';
   return resumeSidLabel(realSid);
 }
 
@@ -3200,7 +3237,7 @@ function runningPanel() {
   const stopping = state.live.stopping || new Set();
   return Panel({
     key: 'runningPanel',
-    title: 'running · ' + running.length,
+    title: 'all running sessions · ' + running.length + (state._activeStale ? ' (refreshing…)' : ''),
     children: [
       // A discoverable path from "this running chat" to the management surface.
       h('div', { key: 'runall', class: 'resume-banner', role: 'group' },

package/site/app/vendor/anentrypoint-design/247420.css

@@ -7130,6 +7130,7 @@
 .ds-247420 .chat-msg-action-label { font-size: var(--fs-tiny); }
 .ds-247420 .chat-msg-action:hover { background: var(--bg-2); color: var(--fg); }
 .ds-247420 .chat-msg-action:focus-visible { outline: var(--focus-w) solid var(--focus-color); outline-offset: 1px; opacity: 1; }
+.ds-247420 .chat-msg-action.is-copied { color: var(--accent-ink); opacity: 1; }
 
 /* Empty state: a fresh thread is an invitation, not a void. */
 .ds-247420 .agentchat-empty {
@@ -7164,6 +7165,11 @@
 }
 .ds-247420 .agentchat-empty-suggestion:hover, .ds-247420 .chat-empty-suggestion:hover { background: color-mix(in srgb, var(--accent, var(--fg)) 12%, transparent); }
 .ds-247420 .agentchat-empty-suggestion:focus-visible, .ds-247420 .chat-empty-suggestion:focus-visible { outline: 2px solid var(--accent, var(--fg)); outline-offset: var(--focus-offset); }
+/* Follow-up chips (contextual, based on the last turn) vs empty-state seed
+   chips (generic starter ideas) previously shared identical DOM shape with no
+   visual signal distinguishing them - a dashed border reads as "derived from
+   context" without introducing a new color/shape language. */
+.ds-247420 .agentchat-followup { border-style: dashed; }
 
 /* Guided install path in the empty state: copy line + monospaced command rows
    (each with its own copy button) + a recheck button. No animation. */