@yemi33/minions 0.1.1754 → 0.1.1756

package/CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## 0.1.1756 (2026-05-06)
+
+### Other
+- perf(doc-chat): cache locked answer in stream stripper, drop 20K doc truncation
+
+## 0.1.1755 (2026-05-06)
+
+### Fixes
+- keep queued strips pinned below progress UX and answers
+
 ## 0.1.1754 (2026-05-06)
 
 ### Fixes

package/dashboard/js/modal-qa.js

@@ -30,13 +30,23 @@ const QA_QUEUE_CAP = 10; // max queued messages
 const QA_STREAM_STALL_MS = 6 * 60 * 1000; // allow the full doc-chat timeout before treating heartbeat-only streams as stalled
 let _qaSessionKey = ''; // key for current conversation (title or filePath)
 
+// Insert html at the bottom of the thread but above any pending "Queued: ..."
+// strips, so queued messages always remain visually below the active progress
+// UX / answer / errors.
+function _qaInsertBeforeQueued(container, html) {
+  if (!container) return;
+  const firstQueued = container.querySelector && container.querySelector('.qa-queued-item');
+  if (firstQueued) firstQueued.insertAdjacentHTML('beforebegin', html);
+  else container.insertAdjacentHTML('beforeend', html);
+}
+
 function _renderQaUserMessage(thread, message, selection) {
   let qHtml = '<div class="modal-qa-q">' + escHtml(message);
   if (selection) {
     qHtml += '<span class="selection-ref">Re: "' + escHtml(selection.slice(0, 100)) + ((selection.length > 100) ? '...' : '') + '"</span>';
   }
   qHtml += '</div>';
-  thread.insertAdjacentHTML('beforeend', qHtml);
+  _qaInsertBeforeQueued(thread, qHtml);
   thread.scrollTop = thread.scrollHeight;
   _showThreadWrap();
 }
@@ -442,7 +452,7 @@ async function _processQaMessage(message, selection, opts) {
   const loadingId = 'chat-loading-' + Date.now();
   const loadingHtml = _qaBuildLoadingHtml(loadingId, runtime.queue.length);
   const startThreadHtml = _qaMutateThreadHtml(sessionKey, tmp => {
-    tmp.insertAdjacentHTML('beforeend', loadingHtml);
+    _qaInsertBeforeQueued(tmp, loadingHtml);
   });
   _qaPersistSession(sessionKey, {
     threadHtml: startThreadHtml,
@@ -606,8 +616,8 @@ async function _processQaMessage(message, selection, opts) {
         let updatedThreadHtml = _qaMutateThreadHtml(sessionKey, tmp => {
           const loadingEl = tmp.querySelector('#' + loadingId);
           if (loadingEl) loadingEl.remove();
-          tmp.insertAdjacentHTML('beforeend', answerHtml);
-          if (rawErrorHtml) tmp.insertAdjacentHTML('beforeend', rawErrorHtml);
+          _qaInsertBeforeQueued(tmp, answerHtml);
+          if (rawErrorHtml) _qaInsertBeforeQueued(tmp, rawErrorHtml);
         });
 
         runtime.history.push({ role: 'user', text: message });
@@ -621,7 +631,7 @@ async function _processQaMessage(message, selection, opts) {
         } else if (evt.actionParseError) {
           const warning = '<div class="modal-qa-a" style="color:var(--red)">Actions block emitted but JSON could not be parsed — no actions were executed. Resend or rephrase. (' + escHtml(String(evt.actionParseError).slice(0, 200)) + ')</div>';
           updatedThreadHtml = _qaMutateThreadHtml(sessionKey, tmp => {
-            tmp.insertAdjacentHTML('beforeend', warning);
+            _qaInsertBeforeQueued(tmp, warning);
           });
         }
 
@@ -698,7 +708,7 @@ async function _processQaMessage(message, selection, opts) {
     const updatedThreadHtml = _qaMutateThreadHtml(sessionKey, tmp => {
       const loadingEl = tmp.querySelector('#' + loadingId);
       if (loadingEl) loadingEl.remove();
-      tmp.insertAdjacentHTML('beforeend', messageHtml);
+      _qaInsertBeforeQueued(tmp, messageHtml);
     });
     _qaPersistSession(sessionKey, {
       threadHtml: updatedThreadHtml,
@@ -733,7 +743,7 @@ async function _processQaMessage(message, selection, opts) {
     const nextThreadHtml = _qaMutateThreadHtml(sessionKey, tmp => {
       const queuedEl = tmp.querySelector('.qa-queued-item');
       if (queuedEl) queuedEl.remove();
-      tmp.insertAdjacentHTML('beforeend', _qaBuildUserMessageHtml(next.message, next.selection));
+      _qaInsertBeforeQueued(tmp, _qaBuildUserMessageHtml(next.message, next.selection));
     });
     _qaPersistSession(sessionKey, {
       threadHtml: nextThreadHtml,

package/dashboard.js

@@ -2653,10 +2653,40 @@ function _recoverPartialDocChatResponse(result, sessionKey) {
 }
 
 
+// Wraps the streaming onChunk so that once the document delimiter is observed
+// in the growing text, subsequent chunks reuse the locked answer instead of
+// re-scanning the tail. The model emits "<explanation> ---DOCUMENT--- <full file>"
+// — the answer portion can't grow after the delimiter, so re-parsing every
+// chunk through the regenerated file body is wasted O(n²) work.
+//
+// Also dedups identical post-strip answers: the upstream accumulator dedups
+// against the raw growing text, but that text keeps changing as the document
+// body streams in even though the visible answer is locked. Without this
+// guard the SSE writer fires a duplicate `chunk` event for every doc-body
+// delta, which triggers a client DOM rerender and localStorage write each time.
+function _makeDocChatStreamStripper(onChunk) {
+  if (!onChunk) return undefined;
+  let lockedAnswer = null;
+  let lastSent;
+  return (text) => {
+    let answer;
+    if (lockedAnswer !== null) {
+      answer = lockedAnswer;
+    } else {
+      const parsed = _parseDocChatResultText(text);
+      if (parsed.content !== null) lockedAnswer = parsed.answer;
+      answer = parsed.answer;
+    }
+    if (answer === lastSent) return;
+    lastSent = answer;
+    onChunk(answer);
+  };
+}
+
 // Doc-specific wrapper — adds document context, parses ---DOCUMENT---
 async function ccDocCall({ message, document, title, filePath, selection, canEdit, isJson, model, freshSession, onAbortReady }) {
   const sessionKey = filePath || title;
-  const docSlice = document.slice(0, 20000);
+  const docSlice = String(document || '');
 
   // freshSession: true → discard any prior session for this key so the call starts clean.
   // Used by one-shot generation flows (e.g. Create Plan from meeting) that must not
@@ -2720,7 +2750,8 @@ async function ccDocCall({ message, document, title, filePath, selection, canEdi
 
 async function ccDocCallStreaming({ message, document, title, filePath, selection, canEdit, isJson, model, freshSession, onAbortReady, onChunk, onToolUse, onRetry }) {
   const sessionKey = filePath || title;
-  const docSlice = document.slice(0, 20000);
+  const docSlice = String(document || '');
+  const streamStripper = _makeDocChatStreamStripper(onChunk);
 
   if (freshSession && sessionKey) {
     docSessions.delete(sessionKey);
@@ -2742,7 +2773,7 @@ async function ccDocCallStreaming({ message, document, title, filePath, selectio
       systemPrompt: DOC_CHAT_SYSTEM_PROMPT,
       ...(model ? { model } : {}),
       onAbortReady,
-      onChunk: (text) => { if (onChunk) onChunk(_docChatDisplayText(text)); },
+      onChunk: streamStripper,
       onToolUse,
       onRetry,
     });
@@ -7464,6 +7495,7 @@ module.exports = {
   parsePinnedEntries,
   _parseDocChatResultText,
   _formatDocChatContext,
+  _makeDocChatStreamStripper,
   _docChatErrorMessage,
   _docChatPartialWarning,
   _docChatFailureResponse,

package/engine/copilot-models.json

@@ -1,5 +1,5 @@
 {
   "runtime": "copilot",
   "models": null,
-  "cachedAt": "2026-05-06T21:43:07.445Z"
+  "cachedAt": "2026-05-06T22:29:15.651Z"
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@yemi33/minions",
-  "version": "0.1.1754",
+  "version": "0.1.1756",
   "description": "Multi-agent AI dev team that runs from ~/.minions/ — five autonomous agents share a single engine, dashboard, and knowledge base",
   "bin": {
     "minions": "bin/minions.js"