fix(story-mode): what the cold readers believed that was not so (review round 8)

Two agents read only the comic of the eleven recorded sessions (one who
has never written code, one engineering manager skimming), wrote down
what they believed, then compared with the transcripts. Both were misled
in the same places, and the cause was the same: the final panel kept
Claude's first sentence or two, and what a reader must not lose sat
further into its last message.

- Claude's last words now keep key sentences from anywhere in its final
  message (read in full, past the label's clip): an open problem or a
  quirk, something not yet done, checked or confirmed, work still out,
  what you asked for or expected, a correction of an earlier warning
  (with the sentence it refers to). A bare "I didn't" counts only when
  nothing stronger is there; table rows, list intros and questions never.
  "I built the pomodoro timer. Opening the page: it won't work if you
  double-click index.html…"; "…That note makes the current behaviour
  clear, but it doesn't fix it."; "As you expected, git describe --tags
  then failed because the repo has no tags."
- A turn that ended with helpers or a background command still out says
  so: "Done for now", and "Claude's part is done in 36 seconds. Its 2
  helpers finished later and reported back." (`finishedLater`).
- A failure you asked for, or that Claude said it expected, reads as the
  plan: "Test first: the first run fails as planned, the second
  passes.", "The tests don't even load, as planned." A loop that failed
  once on a test and once on another check says both, and the commits
  made inside a loop are part of its story; the final panel counts the
  turn's commits ("5 commits, uploaded").
- One or two files read are named ("Claude checks README.md." in a
  turn a report started); a tool lookup reads "Claude gets the tools it
  needs ready." again (it was overwritten) without its query; the
  advisor is "a stronger AI model", and one that answered with an error
  "gets no answer"; a commit message loses its `feat:` prefix in the
  plain caption.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

src/story-mode.ts

@@ -127,7 +127,7 @@ export type StoryEntry =
       /** Teammates (helpers started as a team) reporting back, matched to their helper by name. */
       teammates?: StoryTeammateReport[];
     }
-  | { k: 'text'; at: string; text: string }
+  | { k: 'text'; at: string; text: string; keyPoints?: string[] }
   | { k: 'tool'; at: string; id: string; name: string; input: ReducedInput }
   | {
       k: 'result';
@@ -269,11 +269,75 @@ export function balloonText(text: string): string {
   return clip(two && REPORTS_PROBLEM.test(one) && two.length <= 140 ? `${one} ${two}` : one, 260);
 }
 
-/** Claude's last words: the first sentence, and the next when it carries a caveat. */
-export function closingText(text: string): string {
+/**
+ * A sentence a reader of Claude's last words must not lose. STRONG: an open problem or quirk,
+ * something not yet done, checked or confirmed, work still out, a word on what you asked for or
+ * expected, a correction of an earlier warning. WEAK: a bare negation ("I didn't push"), kept only when
+ * nothing strong is there.
+ */
+const KEY_POINT_STRONG = new RegExp(
+  [
+    String.raw`\b(problem|known (issue|quirk)s?|quirks?|caveat|limitation|heads up|warning|not yet|haven't|have not|never|unfixed|left (open|unfixed|out)|still (fails?|failing|broken|missing|open|running|working)|must be (served|run)|(won't|doesn't|does not|will not) work|can't tell|isn't (confirmed|committed|pushed|tested|done|checked)|not (checked|tested|verified|confirmed|committed|pushed|opened))\b`,
+    String.raw`\b(as you (asked|expected|said|wanted|suggested)|you asked|your (--?[\w-]+|request|option|flag|change))\b`,
+    String.raw`\b(out of date|looked too early|turned out|was wrong)\b`,
+  ].join('|'),
+  'i'
+);
+const KEY_POINT_WEAK = /\b(didn't|did not|couldn't|could not|wasn't able|won't|doesn't)\b/i;
+
+/**
+ * Sentences from further into a long message that carry a key point, at most two strong ones (or one
+ * weak one when none is strong), in order. Read on the whole message (the event's label is clipped),
+ * its first 8,000 characters. Table rows, list intros ending in `:` and questions are skipped: none of
+ * them reads on its own in a balloon.
+ */
+function keyPointsOf(text: string): string[] {
+  if (text.length <= 160) return [];
+  const spans: string[] = [];
+  const plain = plainMarkdown(text.slice(0, 8000), spans);
+  const strong: string[] = [];
+  let weak: string | undefined;
+  let first = true;
+  let previous = '';
+  for (const line of plain.split('\n')) {
+    if (strong.length >= 2) break;
+    const trimmed = line.trim();
+    if (trimmed.startsWith('|') || (trimmed.match(/\|/g)?.length ?? 0) >= 2) continue; // a table row
+    let rest = trimmed.replace(/\s+/g, ' ');
+    while (rest && strong.length < 2) {
+      const m = SENTENCE_END.exec(rest);
+      const sentence = (m ? m[1] : rest).trim();
+      rest = m ? rest.slice(m[1].length).trim() : '';
+      // The first sentence is the balloon's anyway.
+      if (first) {
+        first = false;
+        continue;
+      }
+      const before = previous;
+      previous = sentence;
+      if (sentence.length < 12 || /[:?]$/.test(sentence)) continue;
+      // "That's out of date" needs what "that" was: the sentence before comes along.
+      const context = /^(that|this|it)\b/i.test(sentence) && before && !strong.includes(before) ? `${before} ` : '';
+      const shown = clip(restoreSpans(context + sentence, spans), context ? 300 : 180);
+      if (KEY_POINT_STRONG.test(sentence)) strong.push(shown);
+      else if (!weak && KEY_POINT_WEAK.test(sentence)) weak = shown;
+    }
+  }
+  return strong.length ? strong : weak ? [weak] : [];
+}
+
+/**
+ * Claude's last words: the first sentence, the next when it carries a caveat, and the key points from
+ * further in (an open problem, what was or was not done), so the final panel never reads as a clean
+ * finish when Claude said otherwise.
+ */
+export function closingText(text: string, keyPoints: string[] = []): string {
   const [one, two] = firstTwoSentences(text);
   // A very short first sentence ("Not done.") says too little alone.
-  return clip(two && (CAVEAT.test(two) || one.length < 24) && two.length <= 200 ? `${one} ${two}` : one, 300);
+  const parts = [one];
+  if (two && (CAVEAT.test(two) || one.length < 24) && two.length <= 200) parts.push(two);
+  for (const point of keyPoints) if (!parts.some((p) => p.includes(point) || point.includes(p))) parts.push(point);
+  return clip(parts.join(' '), parts.length > 2 ? 420 : 300);
 }
 
 /** The first sentence of a message, without markdown, for a speech balloon. */
@@ -1283,14 +1347,17 @@ export function createStoryReducer({ sidechain = false }: { sidechain?: boolean
   function reduceAssistant(entry: Record<string, unknown>, at: string): StoryEntry[] {
     const message = isRecord(entry.message) ? entry.message : {};
     const content = message.content;
-    if (typeof content === 'string')
-      return content.trim() ? [{ k: 'text', at, text: clip(content, STORY_LIMITS.textChars) }] : [];
+    const textEntry = (text: string): StoryEntry => {
+      const keyPoints = keyPointsOf(text);
+      return { k: 'text', at, text: clip(text, STORY_LIMITS.textChars), ...(keyPoints.length ? { keyPoints } : {}) };
+    };
+    if (typeof content === 'string') return content.trim() ? [textEntry(content)] : [];
     if (!Array.isArray(content)) return [];
     const out: StoryEntry[] = [];
     for (const block of content) {
       if (!isRecord(block)) continue;
       if (block.type === 'text' && typeof block.text === 'string' && block.text.trim()) {
-        out.push({ k: 'text', at, text: clip(block.text, STORY_LIMITS.textChars) });
+        out.push(textEntry(block.text));
       } else if (block.type === 'tool_use' && typeof block.id === 'string' && typeof block.name === 'string') {
         const input = reduceInput(block.name, block.input, str(entry.cwd));
         remember(block.id, block.name, input);
@@ -1581,6 +1648,12 @@ export function buildStoryTurns(entries: StoryEntry[]): StoryTurn[] {
   /** The same helpers by name: a teammate reports back by name, not by call. */
   const helpersByName = new Map<string, StoryEvent>();
   let say = 0;
+  /** The turn each helper or command call belongs to: a report after that turn ended came back later. */
+  const turnOfCall = new WeakMap<StoryEvent, StoryTurn>();
+  const cameBackLater = (ev: StoryEvent) => {
+    const own = turnOfCall.get(ev);
+    return !!own && (own !== cur || own.ended);
+  };
   const teammatesBack = (reports: StoryTeammateReport[], prompt: StoryEvent) => {
     const told = reports.filter((r) => r.result || r.summary);
     if (told.length) {
@@ -1593,6 +1666,7 @@ export function buildStoryTurns(entries: StoryEntry[]): StoryTurn[] {
       const ev = helpersByName.get(r.from);
       if (!ev?.helper) continue;
       const helper: StoryHelper = { ...ev.helper, running: undefined };
+      if (helper.background && cameBackLater(ev)) ev.finishedLater = true;
       if (r.result) helper.result = r.result;
       ev.helper = helper;
     }
@@ -1614,6 +1688,7 @@ export function buildStoryTurns(entries: StoryEntry[]): StoryTurn[] {
     if (ev?.running && !ev.helper) {
       // A background command: done now, and its summary carries the exit code.
       ev.running = undefined;
+      if (cameBackLater(ev)) ev.finishedLater = true;
       const code = /\(exit code (\d+)\)/.exec(prompt.label)?.[1];
       ev.ok =
         code !== undefined ? code === '0' : !report.status || /^(completed|done|finished|success)/i.test(report.status);
@@ -1622,6 +1697,7 @@ export function buildStoryTurns(entries: StoryEntry[]): StoryTurn[] {
     }
     if (!ev?.helper) return;
     const helper: StoryHelper = { ...ev.helper, running: undefined };
+    if (helper.background && cameBackLater(ev)) ev.finishedLater = true;
     if (report.result) helper.result = report.result;
     if (report.toolUses !== undefined) helper.toolUses = report.toolUses;
     if (report.durationMs !== undefined) helper.durationMs = report.durationMs;
@@ -1712,6 +1788,7 @@ export function buildStoryTurns(entries: StoryEntry[]): StoryTurn[] {
           tool: 'Text',
           label: e.text,
           intent: 'say',
+          ...(e.keyPoints ? { keyPoints: e.keyPoints } : {}),
         });
         break;
       case 'tool': {
@@ -1721,6 +1798,7 @@ export function buildStoryTurns(entries: StoryEntry[]): StoryTurn[] {
         if (ev.helper || ev.tool === 'Bash') {
           // A helper, or a command that may turn out to run in the background: both report back later.
           helpers.set(e.id, ev);
+          turnOfCall.set(ev, ensure(e.at));
           if (helpers.size > 2000) helpers.delete(helpers.keys().next().value as string);
           if (ev.helper) {
             helpersByName.set(ev.helper.name, ev);
@@ -1916,11 +1994,11 @@ const ANNOUNCES = /\b(I'll|I will|let me|let's|I'm going to|next,? I|now I)\b/i;
 function groupPanels(events: StoryEvent[]): { panels: DraftPanel[]; trailingSay?: string } {
   const panels: DraftPanel[] = [];
   let pendingSay: string | undefined;
-  let lastSaid: string | undefined;
+  let lastSaid: StoryEvent | undefined;
   events.forEach((ev, idx) => {
     if (ev.intent === 'say' && !ev.todos) {
       pendingSay ??= balloonText(ev.label);
-      lastSaid ??= ev.label;
+      lastSaid ??= ev;
       return;
     }
     lastSaid = undefined;
@@ -1959,7 +2037,7 @@ function groupPanels(events: StoryEvent[]): { panels: DraftPanel[]; trailingSay?
     pendingSay = undefined;
   });
   // The closing words keep a caveat that follows them ("Nothing is committed: this is no git repo.").
-  return { panels, trailingSay: lastSaid !== undefined ? closingText(lastSaid) : pendingSay };
+  return { panels, trailingSay: lastSaid !== undefined ? closingText(lastSaid.label, lastSaid.keyPoints) : pendingSay };
 }
 
 /** Fix → check → fix → check (4 panels or more) is one "loop" panel: try, test, repeat. */
@@ -2076,6 +2154,16 @@ function montageSummary(evs: StoryEvent[]): string {
   return parts.length ? parts.join(', ') : 'a few more steps';
 }
 
+/** You asked for a failing run first (test first), or Claude said the failure was expected. */
+const PLANNED_IN_PROMPT =
+  /\b(test[- ]first|tdd|fail(s|ing)? (first|on purpose)|failing tests? first|on purpose|red first|before (the )?(code|module|function|implementation) exists)\b/i;
+const EXPECTED_BY_CLAUDE =
+  /\b(as expected|as planned|as intended|expected (failure|to fail|red)|deliberately fail\w*|on purpose|test[- ]first|the red state)\b/i;
+function plannedFailure(turn: StoryTurn, events: StoryEvent[]): boolean {
+  const prompt = events.find((e) => e.intent === 'prompt')?.label ?? turn.prompt;
+  return PLANNED_IN_PROMPT.test(prompt) || events.some((e) => e.intent === 'say' && EXPECTED_BY_CLAUDE.test(e.label));
+}
+
 function decorate(draft: DraftPanel, events: StoryEvent[], turn: StoryTurn, index: number): StoryPanel {
   const evs = draft.events.map((i) => events[i]);
   const panel: StoryPanel = { kind: draft.kind, events: draft.events, caption: { story: '', director: '' } };
@@ -2139,15 +2227,23 @@ function decorate(draft: DraftPanel, events: StoryEvent[], turn: StoryTurn, inde
         .filter((e) => e.intent !== 'read')
         .map((e) => e.description || shortCommand(e))
         .slice(0, 4);
-      if (evs.every((e) => e.tool === 'ToolSearch')) story = 'Claude gets the tools it needs ready.';
       if (evs.every((e) => e.tool === 'Bash')) director = `${tools} · ${shortCommand(evs[0])}`;
       if (errors) director += ` · ${plural(errors, 'error')}`;
       const shown = files.filter((f) => !/(^|\/)\.codeman[-.]/.test(f));
-      story = shown.length
-        ? `Claude reads ${plural(shown.length, 'file')} to find its way around.`
-        : evs.every((e) => e.intent === 'search')
-          ? 'Claude searches the code.'
-          : 'Claude looks around the project.';
+      // A file or two by name; in a turn a report started, reading is checking what came back.
+      const checking = turn.origin === 'task' || turn.origin === 'peer';
+      const named = shown.map((f) => basename(f));
+      story = evs.every((e) => e.tool === 'ToolSearch')
+        ? 'Claude gets the tools it needs ready.'
+        : shown.length && shown.length <= 2
+          ? `Claude ${checking ? 'checks' : 'reads'} ${named.join(' and ')}.`
+          : shown.length
+            ? `Claude reads ${plural(shown.length, 'file')} to find its way around.`
+            : evs.every((e) => e.intent === 'search')
+              ? 'Claude searches the code.'
+              : 'Claude looks around the project.';
+      // A tool lookup's query (`select:EnterPlanMode,ExitPlanMode`) means nothing to a reader.
+      if (evs.every((e) => e.tool === 'ToolSearch')) panel.notes = [];
       break;
     }
     case 'research':
@@ -2188,7 +2284,9 @@ function decorate(draft: DraftPanel, events: StoryEvent[], turn: StoryTurn, inde
       const advisor = evs.filter((e) => e.tool === 'Advisor').length;
       story = helpers.length
         ? advisor === helpers.length
-          ? 'Claude asks the advisor for a second opinion.'
+          ? evs.every((e) => e.tool !== 'Advisor' || e.ok === false)
+            ? 'Claude asks a stronger AI model for a second opinion, but gets no answer.'
+            : 'Claude asks a stronger AI model for a second opinion.'
           : behind === helpers.length
             ? `Claude sends ${plural(helpers.length, 'helper')} off to work in the background.`
             : `Claude calls in ${plural(helpers.length, 'helper')}.`
@@ -2246,20 +2344,32 @@ function decorate(draft: DraftPanel, events: StoryEvent[], turn: StoryTurn, inde
           (e) => !!e.tests && e.tests.passed > 0 && !e.tests.failed && !e.tests.errors
         ).length;
         const n = panel.rounds;
+        const testFails = fails - lintFails;
+        // A failure you asked for (test first), or that Claude said it expected, is the plan working.
+        const planned = plannedFailure(turn, events) && passed(checks[0]) === false;
         story =
           panel.ok === true && !fails
             ? `Build and test, ${plural(n, 'round')}. Every check passes.`
             : panel.ok === true && fails === lintFails
-              ? `Build and test, ${plural(n, 'round')}. A check fails ${fails === 1 ? 'once' : `${fails} times`}, then everything passes.`
-              : panel.ok === true && checks.length === 2 && fails === 1
-                ? 'Fix and test: the first run fails, the second passes.'
-                : panel.ok === true
-                  ? `Fix and test, ${plural(n, 'round')}: ${fails === 1 ? 'one run fails' : `${fails} runs fail`}, then it passes.`
-                  : panel.ok === false && tests && tests.passed && !tests.failed && !tests.errors
-                    ? `Fix and test, ${plural(n, 'round')}. The tests pass, but another check fails.`
-                    : panel.ok === false
-                      ? `Fix and test, ${plural(n, 'round')}. Still failing.`
-                      : `Fix and test, ${plural(n, 'round')}.`;
+              ? `Build and test, ${plural(n, 'round')}. The tests pass every time; another check fails ${fails === 1 ? 'once' : `${fails} times`}, then passes.`
+              : panel.ok === true && planned && checks.length === 2
+                ? 'Test first: the first run fails as planned, the second passes.'
+                : panel.ok === true && planned
+                  ? `Test first, ${plural(n, 'round')}: the first run fails as planned, then it passes.`
+                  : panel.ok === true && checks.length === 2 && fails === 1
+                    ? 'Fix and test: the first run fails, the second passes.'
+                    : panel.ok === true && testFails && lintFails
+                      ? `Fix and test, ${plural(n, 'round')}: ${testFails === 1 ? 'a test run fails' : `${testFails} test runs fail`}, then another check, then everything passes.`
+                      : panel.ok === true
+                        ? `Fix and test, ${plural(n, 'round')}: ${fails === 1 ? 'one run fails' : `${fails} runs fail`}, then it passes.`
+                        : panel.ok === false && tests && tests.passed && !tests.failed && !tests.errors
+                          ? `Fix and test, ${plural(n, 'round')}. The tests pass, but another check fails.`
+                          : panel.ok === false
+                            ? `Fix and test, ${plural(n, 'round')}. Still failing.`
+                            : `Fix and test, ${plural(n, 'round')}.`;
+        // Commits made inside the loop (a test one-liner that ends in one) are part of its story.
+        const saved = evs.reduce((k, e) => k + (e.commits ?? 0), 0);
+        if (saved) story += ` Claude saves the work${saved > 1 ? ` in ${saved} commits` : ''} along the way.`;
         director = `${plural(checks.length, 'check')} · ${plural(edits.length, 'edit')} · ${fmtCounts(panel.added, panel.removed)} · last: ${
           lastCheck ? outcome(lastCheck) : 'running'
         }`;
@@ -2291,6 +2401,7 @@ function decorate(draft: DraftPanel, events: StoryEvent[], turn: StoryTurn, inde
             ? `${tests.failed} tests fail.`
             : 'A test fails.'
           : 'A check fails.';
+      if (ok === false && plannedFailure(turn, events)) story = `${story.replace(/[.]$/, '')}, as planned.`;
       // The same command with the same outcome twice in a row reads once, with a count.
       const runs: Array<{ text: string; n: number }> = [];
       for (const e of checks) {
@@ -2355,7 +2466,9 @@ function decorate(draft: DraftPanel, events: StoryEvent[], turn: StoryTurn, inde
       const uploadOnly = ships === 'push' && !commits;
       // One commit: its message says what was saved.
       const subject = commits === 1 ? evs.find((e) => e.subject)?.subject : undefined;
-      const said = subject ? `: "${clip(subject, 60).replace(/[.]$/, '')}"` : '';
+      // Its conventional prefix (`feat:`, `chore(deps):`) is for git, not for the reader.
+      const plainSubject = subject?.replace(/^[a-z]+(\([^)]*\))?!?:\s*/i, '');
+      const said = plainSubject ? `: "${clip(plainSubject, 60).replace(/[.]$/, '')}"` : '';
       story = savedAnyway
         ? uploadOnly
           ? 'Claude uploads the work, but a step after that fails.'
@@ -2420,6 +2533,9 @@ function decorate(draft: DraftPanel, events: StoryEvent[], turn: StoryTurn, inde
 function completePanel(turn: StoryTurn, page: Omit<StoryPage, 'panels'>, trailingSay?: string): StoryPanel {
   const running = page.helpersRunning;
   const commands = page.events.filter((e) => e.running && !e.helper).length;
+  const later = page.events.filter((e) => e.finishedLater && e.intent !== 'prompt');
+  const laterHelpers = later.filter((e) => e.helper).length;
+  const laterCommands = later.length - laterHelpers;
   const done = turn.durationMs !== undefined ? `Done in ${roughDuration(turn.durationMs)}.` : 'Done.';
   const story = turn.interrupted
     ? 'You stop Claude here.'
@@ -2436,7 +2552,18 @@ function completePanel(turn: StoryTurn, page: Omit<StoryPage, 'panels'>, trailin
             ]
               .filter(Boolean)
               .join(' ')}`
-          : done;
+          : later.length
+            ? `Claude's part is done${turn.durationMs !== undefined ? ` in ${roughDuration(turn.durationMs)}` : ''}. ${[
+                laterHelpers
+                  ? `${laterHelpers === 1 ? 'Its helper' : `Its ${laterHelpers} helpers`} finished later and reported back.`
+                  : '',
+                laterCommands
+                  ? `${laterCommands === 1 ? 'Its background command' : `${laterCommands} background commands`} finished later.`
+                  : '',
+              ]
+                .filter(Boolean)
+                .join(' ')}`
+            : done;
   const director = [
     turn.endedBy === 'next prompt' ? 'superseded by the next prompt' : (turn.endedBy ?? 'turn_duration'),
     turn.durationMs !== undefined ? exactDuration(turn.durationMs) : '',
@@ -2501,6 +2628,9 @@ export function buildStoryPage(turn: StoryTurn): StoryPage {
     // The advisor is a second opinion, not a helper that does work.
     helpers: events.filter((e) => e.helper && e.tool !== 'Advisor' && e.intent !== 'prompt').length,
     helpersRunning: events.filter((e) => e.helper?.running).length,
+    ...(events.some((e) => e.finishedLater && e.intent !== 'prompt')
+      ? { finishedLater: events.filter((e) => e.finishedLater && e.intent !== 'prompt').length }
+      : {}),
   };
   const calls = events.filter((e) => !NOT_TOOL_CALLS.has(e.intent));
   if (calls.length && calls.every((e) => e.ok === false)) base.noneWorked = true;

src/types/story.ts

@@ -130,6 +130,13 @@ export interface StoryEvent {
   running?: boolean;
   /** A command that ran in the background (its result came back later, as a report of its own). */
   background?: boolean;
+  /** Background work (this command, or this helper call's helper) that finished after its turn ended. */
+  finishedLater?: boolean;
+  /**
+   * Sentences from further into a long message that a reader must not lose: an open problem, a step
+   * not taken, what you asked for. From the whole message, past the clip of `label`.
+   */
+  keyPoints?: string[];
   /** A save that ships the work: `gh pr create` (`pr`) or `git push` (`push`). */
   ships?: 'pr' | 'push';
   /** How many commits a Bash call makes (`git commit` segments). */
@@ -201,6 +208,8 @@ export interface StoryPage extends StoryTurnSummary {
   helpers: number;
   /** Background helpers that had not reported back when this page was drawn. */
   helpersRunning: number;
+  /** Helpers and background commands of this turn that finished only after it ended. */
+  finishedLater?: number;
   /** Every tool call of the turn failed (a blocked command, nothing that worked): no celebration. */
   noneWorked?: boolean;
 }

src/web/public/story-mode.js

@@ -2481,8 +2481,9 @@ Object.assign(CodemanApp.prototype, {
           ? 'To be continued…'
           : page.noneWorked
             ? 'Turn over'
-            : page.helpersRunning || page.events.some((e) => e.running && !e.helper)
-              ? 'To be continued…'
+            : page.helpersRunning || page.finishedLater || page.events.some((e) => e.running && !e.helper)
+              ? // Work was still out when the turn ended (and maybe still is): not a clean finish.
+                'Done for now'
               : 'Mission complete!';
       const chips = rt.splash.querySelector('.story-chips');
       const items = [];
@@ -2493,6 +2494,18 @@ Object.assign(CodemanApp.prototype, {
           open: 'diff:page',
         });
       items.push({ text: `${page.toolCalls} tool call${page.toolCalls === 1 ? '' : 's'}`, open: 'list:page' });
+      // What was saved, wherever in the turn: commits inside a loop or a test one-liner count too.
+      const commits = page.events.reduce((n, e) => n + (e.commits || 0), 0);
+      // An upload that went through, even when a step after it in the same command failed.
+      const shipped = page.events.some(
+        (e) => e.ships === 'push' && (e.ok !== false || /\[new branch\]|\S+\.\.\S+\s+\S+\s+->/.test(e.output || ''))
+      );
+      if (commits || shipped)
+        items.push({
+          text: [commits ? `${commits} commit${commits === 1 ? '' : 's'}` : '', shipped ? 'uploaded' : '']
+            .filter(Boolean)
+            .join(', '),
+        });
       if (page.helpers)
         items.push({
           text: `${page.helpers} helper${page.helpers === 1 ? '' : 's'}${