agentgui 1.0.1053 → 1.0.1054

package/.gm/memories/mem-c7c5bc263850c332-1844.md

@@ -0,0 +1,8 @@
+---
+key: mem-c7c5bc263850c332-1844
+ns: default
+created: 1783854998213
+updated: 1783854998213
+---
+
+agentgui 38th run (2026-07-12): re-ran gui-logic-predictability workflow wf_186e2a2a-dfc (32 agents) one cycle after the 37th run, 18 confirmed findings all implemented. Key app fixes: stopAllActive now clears failed sids from state.live.stopping (was permanently stuck), copyText announces clipboard failures, title-fallback chains unified via projectLabel()+sid across search/history/rail, Live nav-badge stale-flame fixed via stream carve-out, computeLiveSessions gained a 5s reconnect grace window against status flapping, streamHistory EventSource gained exponential backoff mirroring the WS client. Kit fixes: follow-up chips lost aria-hidden (real buttons were AT-invisible), MdNode markdown-settle now preserves active text selection via a deferred selectionchange-gated innerHTML swap, SearchInput gained a visible clear button + working resultCount aria-live wiring. Kit pushed bed4036 (after a rebase with one dist-only conflict, resolved by rebuilding dist fresh rather than hand-merging generated code), agentgui pushed 159030f then b9d9d70 (docs), all CI green both repos. NEW STANDING TOOL DEFECT this run: the gm browser verb's dom= and capture prefixes both throw ReferenceError (no document/page binding at all) in this tool version - a regression beyond the previously-known 'dom= ignores trailing script' issue. Only screenshot= and url=<target>\n<literal-non-DOM-expr> dispatches actually reach the real page; browser session ids silently rotate across dispatches with no state carryover, and one dispatch hit a dead extension-bridge session requiring an unprompted CDP relay restart. Filed as PRD row tool-defect-browser-dom-capture-eval-broken with blockedBy:external since it's a gm-plugkit/playwriter defect, not fixable from agentgui/design repos. Screenshots remain the reliable witness fallback until fixed upstream.

package/AGENTS.md

@@ -1,5 +1,9 @@
 # AgentGUI — Agent Notes
 
+## GUI logic+predictability sweep (2026-07-12) — thirty-ninth run
+
+`gui-logic-predictability` workflow wf_88c431e5-931 (26 agents, 202 tool calls, 0 errors) — 6 confirmed findings after hunt+adversarial-verify, all implemented; a much thinner yield than the 37th/38th runs, consistent with the same lens converging after two straight passes. App (`site/app/js/app.js`): `runningPanel()`'s inline running-chat rows previously showed only agent+model+elapsed+cwd with no session title, reading nameless while the identical session shows a resolved title on the sessions rail/history rail/Live SessionCard — a new `runningRowTitle()` helper (wrapping the existing `resumeSidLabel()`) prepends the same resolved identity string; `stopActiveChat()` could wedge a stop button permanently disabled if the backend accepted the cancel but the underlying process outlived the immediate post-cancel `refreshActive()` call (the next 3s poll would silently re-add the sid to `state.active` while `stopping` never cleared) — a bounded 9s timeout now clears `stopping` and surfaces the same "conversation did not stop - try again" retry banner used on outright rejection; `computeLiveSessions()`'s per-sid SSE tally accumulated `tokens` but never `cost`, so every live/external session showed a blank cost field except the one open in-page in Chat — the tally now accumulates `total_cost_usd` from `result` events too, with the same tally-fallback pattern tokens already had; WorkspaceRail's footer gained a visible shortcuts-help `IconButton` (previously the `?`-overlay was 100% keyboard-only tribal knowledge with zero discovery path outside Settings). Kit (`design/src/components/{chat,content,shell}.js`): `CodeNode.refSink`'s syntax-highlight swap gained the same in-progress-text-selection guard `MdNode.refSink` already had (the two paths had drifted — code blocks still wiped an active selection mid-stream after the markdown path was fixed for the same bug); `Row.sub` gained a `title=` tooltip fallback for string values, matching every other long-text field in the app (agent-chat cwd, sessions rail title/sub/cwd/agent) — `Row` is used everywhere (history sessions rail, search results, context-pane cwd fact, agents panel) so this closes the gap at every call site at once; added a `help` icon path to `ICON_PATHS`/`Glyph` for the new rail affordance. Kit built+tested (`bun test.js` all pass), rebased cleanly on a concurrent `v0.0.316` release-bump commit (no source conflicts), dist rebuilt fresh post-rebase, pushed `00dae4c`. agentgui re-vendored, pushed `64f4466`, all 3 CI checks (Auto-Declaudeify/Publish and Release/Deploy GH Pages) green. Browser-witnessed live on buildesk via `page.context().setHTTPCredentials()` + `page.evaluate()`: help button found in `.ws-rail-foot`, clicking it opened the shortcuts overlay (`shortcutsTextPresent:true`), zero pageErrors; history rail `Row.sub` confirmed rendering `title="server default"` live. **New tool-defect note this run: this env's nginx Basic-Auth password contains a literal comma (`123,slam,123,slam` — see the 32nd-run standing gotcha), and embedding it directly in a `page.goto` URL (`https://user:pass@host/...`) throws `net::ERR_INVALID_AUTH_CREDENTIALS` at the Chromium URL-parser level even when percent-encoded (`%2C`) — this is a stricter failure than the previously-documented "breaks the app's own fetch()" caveat for comma-free creds. Working fix: `await page.context().setHTTPCredentials({username, password})` before `page.goto()` with a plain (no-embedded-creds) URL — this is the durable workaround for any future witness against this deploy, comma-password or not.** Also reconfirmed the 38th-run tool defect (`capture` prefix scripts run in a bare Node vm with no `document`/`page` binding unless the script itself calls `page.evaluate(() => {...})` — a raw top-level `document.querySelector(...)` throws `ReferenceError: document is not defined`; wrapping every DOM read/write in `page.evaluate()` inside the capture script is the reliable pattern, not a tool-side fix).
+
 ## GUI logic+predictability sweep (2026-07-12) — thirty-eighth run
 
 `gui-logic-predictability` workflow wf_186e2a2a-dfc (32 agents, 205 tool calls, 0 errors) — 18 confirmed findings, all implemented, re-running the same lens one cycle after the 37th run to catch residual/newly-introduced gaps. App (`site/app/js/app.js`, `site/app/js/backend.js`): `stopAllActive` now clears every attempted sid (both ok and failed) from `state.live.stopping` after `Promise.allSettled`, so a failed bulk-cancel is retryable instead of leaving that row's stop button permanently disabled; `copyText`'s clipboard-write catch now calls `announce('copy failed')` instead of swallowing the error silently (mirrors `copySid`'s existing pattern); search-result row titles now route through `projectLabel()` (`r.snippet || projectLabel(r.title) || projectLabel(r.project) || UNTITLED_CONVERSATION`) matching the rail's contract; the History detail header's title-fallback chain gained `|| state.selectedSid` before the generic label, so a metadata-sparse session shows the same identity in History as it does in the rail; the Live nav-badge error flame no longer goes stale while off the history tab (stream stays connected whenever `state.active` is non-empty); `computeLiveSessions()` gained a 5s `reconnectGraceUntil` window so an SSE reconnect mid-session no longer flaps a running session's status to stale; a brand-new session between the `chat.active` poll and history-index catch-up now shows `'indexing…'` instead of reading identically to a permanently-failed lookup; the Live dashboard's sort/filter prefs (`hydratePrefs()`) now reconcile across browser tabs on focus/visibilitychange, alongside the existing `refreshActive()` call; `streamHistory`'s EventSource (`backend.js`) gained manual close+reopen exponential backoff (500ms start, doubles, capped 30s, resets on `hello`) mirroring the WS client's `scheduleReconnect`, instead of relying on native EventSource's uncapped retry; the sessions rail empty-state and History event-list both gained next-action affordances (a start-from-chat CTA, and a "load all (N hidden)" button for very large sessions); a draft restored from localStorage now surfaces a one-time note that undo history doesn't carry over a reload. Kit (`design/src/components/{agent-chat,chat,content,sessions}.js`, `app-shell.css`): follow-up suggestion chips' container lost its `aria-hidden="true"` — the chips are real interactive `<button>`s that were completely invisible to screen-reader users despite `role="group"` being present; `MdNode`'s markdown-settle `refSink` now defers its `innerHTML` swap via a one-time `selectionchange` listener when the user's active text selection is anchored inside the element (was wiping any in-progress text selection every streaming tick); `ChatComposer`'s paste handler now surfaces a "pasted N characters" note via `flashComposerNote` for large plain-text pastes; `SearchInput` gained a visible clear (X) button (was Escape-only, undiscoverable to mouse/touch users) at a 44px touch target, sharing the same clear path as Escape; `ConversationList` now forwards a `resultCount` prop through to `SearchInput`'s existing (previously unpopulated) `aria-live` result-count region, wired from agentgui's search-results branch as `hits.length + ' result(s)'`. Kit built+tested (`bun test.js` all pass), a concurrent upstream release (`5cdd667`, v0.0.315) required a rebase with one dist-only merge conflict (resolved by rebuilding `dist/247420.js` fresh from the merged sources rather than hand-merging generated code), re-vendored, browser-witnessed live on buildesk (chat tab shows `connected` status + visible follow-up chip buttons, history tab loads 52 conversations cleanly, 0 visible errors across both screenshots). Kit pushed `bed4036`, agentgui pushed `159030f` (after a second push-rebase for a PRD-state-only commit), all 4 CI checks green both repos. **New tool defect this run (recorded, not fixable from this repo — filed as `blockedBy:external`): the `gm browser` verb's `dom=` and `capture` prefixes both throw `ReferenceError` (no `document`/page binding at all) rather than evaluating in page context — a regression beyond the 37th run's already-documented "`dom=` ignores its trailing script" caveat. Only `screenshot=` and `url=<target>\n<literal-non-DOM-expr>` dispatches actually reach the real page in this tool version; browser sessions also silently rotate ids across dispatches with no state carryover, and one dispatch hit a dead extension-bridge session requiring an unprompted CDP relay restart. Screenshots are the reliable fallback for browser witnessing until this is fixed upstream in gm-plugkit/playwriter.**

package/CHANGELOG.md

@@ -1,3 +1,9 @@
+## [Unreleased] - 39th-run gui-logic-predictability sweep: running-panel identity, stop-button wedge, live cost fallback, shortcuts affordance (kit 00dae4c, app 64f4466)
+
+- app: `runningPanel()` rows now show the same resolved session title used on every other surface (sessions rail, history rail, Live SessionCard) instead of reading nameless; `stopActiveChat()` gains a bounded 9s timeout so a cancel that succeeds server-side but whose process lingers past the immediate refresh no longer wedges the stop button permanently disabled; `computeLiveSessions()`'s per-sid SSE tally now accumulates cost alongside tokens, so every live/external session shows a real cost figure, not just the one open in Chat; WorkspaceRail's footer gains a visible shortcuts-help icon button (the `?` overlay was previously 100% keyboard-only tribal knowledge with no discovery path outside Settings). Why: 39th `gui-logic-predictability` workflow (wf_88c431e5-931) confirmed these as real operating-logic gaps.
+- kit: `CodeNode.refSink`'s syntax-highlight swap gains the same in-progress-selection guard `MdNode.refSink` already had (the two paths had drifted); `Row.sub` gains a `title=` tooltip fallback matching every other long-text field in the kit; new `help` icon path for the rail affordance.
+- browser-witnessed live on buildesk via `page.context().setHTTPCredentials()` (embedded-URL Basic-Auth broke on this env's comma-containing password even percent-encoded) — help button + shortcuts overlay + Row.sub tooltip all confirmed working, zero pageErrors.
+
 ## [Unreleased] - 34th-run gui-ux-craft sweep: kit font-token fallbacks, focus-ring tokenization, agentsLoading state (kit 3b1080c, app 9c8094b)
 
 - kit: `--fs-2`/`--fs-0` shadow font tokens (never defined in the real type scale) remapped to `--fs-lg`/`--fs-tiny`; stale `--lh-base` fallback fixed from 1.4 to 1.55; `.glyph`/`.ds-field-count`/`.ds-row-detail` hardcoded literals tokenized; ~29 more `outline: 2px solid var(--accent-ink)` focus rules converted to `var(--focus-w) solid var(--focus-color)`; `.chat-tool-copy` gained the coarse-pointer 44px touch target; `.ds-select` gained a base `min-height:32px` floor; `Btn`'s `className` prop renamed to `class` for consistency with `Panel`/`Heading`; `AgentControls`/`AgentChat` gained an `agentsLoading` disabled-select state. Why: 34th `gui-ux-craft` workflow (wf_919164f1-648) confirmed these as real, un-filed gaps.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agentgui",
-  "version": "1.0.1053",
+  "version": "1.0.1054",
   "description": "Multi-agent ACP client with real-time communication",
   "type": "module",
   "main": "electron/main.js",