@kodaris/krubble-app-components 1.0.100 → 1.0.102

package/dist/krubble-app.umd.js

@@ -5629,6 +5629,24 @@
            * in place; call requestUpdate() after changing it to re-render.
            */
           this._globalSkillPending = new Set();
+          /**
+           * The live Web Speech recognizer while dictating, else null. Not reactive —
+           * `_isDictating` drives the UI; this is just the handle used to stop it.
+           */
+          this._speechRecognition = null;
+          /**
+           * The composer text as it stood when dictation started. Each `result` event
+           * re-reports the whole utterance (final + interim), so the value is rebuilt as
+           * this base + the transcript rather than appended to — appending would repeat
+           * the interim text on every event.
+           */
+          this._speechBaseValue = '';
+          /**
+           * Transcript of utterances the recognizer has marked final this session.
+           * Held separately because `results` is pruned in continuous mode on some
+           * browsers; without this, earlier sentences would vanish mid-dictation.
+           */
+          this._speechFinalTranscript = '';
           // --- @property (public API) ---
           /**
            * Callback function for sending messages. Receives the user's message and
@@ -5778,6 +5796,33 @@
           this._skillBrowseSource = 'org';
           /** Search text in the browse dialog, routed to the active tab's search. */
           this._skillBrowseQuery = '';
+          /** True while the mic is live and speech is being transcribed into the composer. */
+          this._isDictating = false;
+          /**
+           * Close the `/` skill menu when the click lands outside it.
+           *
+           * Bound at the document because a click anywhere — including outside this
+           * component entirely — should dismiss the menu. `composedPath()` rather than
+           * `e.target` since the menu lives in this component's shadow root, where
+           * `target` on a document listener is retargeted to the host element and every
+           * inside click would look like an outside one.
+           */
+          this._handleDocumentClick = (e) => {
+              if (!this._skillMenuOpened) {
+                  return;
+              }
+              // Excluded: the textarea, since clicking it keeps editing the same `/` command
+              // and dismissing there would fight the user still typing it; and the plus
+              // button, which owns the menu's open/close itself. The plus exclusion matters
+              // in both directions — this listener runs after the button's own handler, so
+              // without it a click that just opened the menu would immediately close it
+              // again, and a click meant to toggle it shut would be double-handled.
+              const path = e.composedPath();
+              if (path.some((el) => el instanceof HTMLElement && (el.classList.contains('skill-menu') || el.classList.contains('input') || el.classList.contains('input-plus')))) {
+                  return;
+              }
+              this._skillMenuOpened = false;
+          };
           this._handleMouseMove = (e) => {
               if (this._isDragging) {
                   let newLeft = this._dragStartLeft + (e.clientX - this._dragStartX);
@@ -5921,6 +5966,7 @@
               this.opened = true;
           }
           this._restoreState();
+          document.addEventListener('click', this._handleDocumentClick);
       }
       disconnectedCallback() {
           super.disconnectedCallback();
@@ -5930,6 +5976,15 @@
           }
           document.removeEventListener('mousemove', this._handleMouseMove);
           document.removeEventListener('mouseup', this._handleMouseUp);
+          document.removeEventListener('click', this._handleDocumentClick);
+          // Release the mic if the chatbot is torn down mid-dictation — otherwise the
+          // recognizer keeps the browser's recording indicator lit with nothing to
+          // transcribe into. abort() (not stop()) since the discarded text has nowhere to go.
+          if (this._speechRecognition) {
+              this._speechRecognition.abort();
+              this._speechRecognition = null;
+              this._isDictating = false;
+          }
           // Release any object URLs still held by staged attachments.
           this._pendingConversationFiles.forEach((p) => {
               if (p.previewUrl) {
@@ -6143,7 +6198,12 @@
               }
               const state = JSON.parse(raw);
               // Draft text and the debug drawer are persisted for all variants.
-              if (typeof state.message === 'string') {
+              // A lone `/` is dropped — it's the in-progress half of a menu interaction,
+              // not a draft, and restoring it would revive a stale menu the user never
+              // used. Anything with an actual command after the slash (`/code`, `/code do
+              // the thing`) is a real draft and restores normally. Trimmed before testing
+              // so a stray space around the slash doesn't defeat the check.
+              if (typeof state.message === 'string' && state.message.trim() !== '/') {
                   this._inputValue = state.message;
               }
               this._debugPanelOpen = state.debugPanelOpen === 'true';
@@ -6536,6 +6596,92 @@
           this._skillMenuOpened = true;
           this._searchInvocableSkills('');
       }
+      /**
+       * The mic button toggles dictation. Speech is transcribed straight into the
+       * composer as editable text — the user still reviews and presses send, so a
+       * misheard word is fixable and nothing is sent by voice alone.
+       *
+       * Recognition runs entirely in the browser's Web Speech API (Chrome/Edge send
+       * audio to Google's service, Safari to Apple's). No audio reaches Kodaris, and
+       * there's no backend call or cost on our side.
+       */
+      _handleMicClick() {
+          // Second click stops. `stop()` flushes the pending utterance as a final
+          // result, so the last words aren't lost; onend clears the state.
+          if (this._speechRecognition) {
+              this._speechRecognition.stop();
+              return;
+          }
+          // Prefixed in Chrome/Edge/Safari; the unprefixed name is checked first for
+          // whenever they drop the prefix. The mic button isn't rendered without one of
+          // these, so reaching here unsupported means a mid-session change — bail quietly.
+          const SpeechRecognitionCtor = window.SpeechRecognition || window.webkitSpeechRecognition;
+          if (!SpeechRecognitionCtor) {
+              return;
+          }
+          const recognition = new SpeechRecognitionCtor();
+          // Keep listening through natural pauses — a chat message is often several
+          // sentences, and the default stops at the first one.
+          recognition.continuous = true;
+          // Show words as they're recognized rather than only at end of utterance.
+          recognition.interimResults = true;
+          // `lang` is deliberately left unset so the recognizer follows the page/OS
+          // language, which is correct for every consumer of this component.
+          // Dictation ADDS to whatever is already typed rather than replacing it, so
+          // the user can type half a message and speak the rest. The separating space
+          // is part of the base so the transcript never runs into the existing word.
+          this._speechBaseValue = this._inputValue;
+          if (this._speechBaseValue && !this._speechBaseValue.endsWith(' ')) {
+              this._speechBaseValue += ' ';
+          }
+          this._speechFinalTranscript = '';
+          recognition.onresult = (e) => {
+              // Everything from resultIndex on is new since the last event: finalized
+              // pieces move to _speechFinalTranscript permanently, while non-final ones
+              // are rebuilt each event because the browser revises its guess as it hears
+              // more. Rebuilding the whole value (base + final + interim) is what keeps
+              // those revisions from stacking up as duplicated text.
+              let interim = '';
+              for (let i = e.resultIndex; i < e.results.length; i++) {
+                  if (e.results[i].isFinal) {
+                      this._speechFinalTranscript += e.results[i][0].transcript;
+                  }
+                  else {
+                      interim += e.results[i][0].transcript;
+                  }
+              }
+              this._inputValue = this._speechBaseValue + this._speechFinalTranscript + interim;
+              this._saveState();
+              // Grow the textarea as dictated text wraps — the same clamp the typing path
+              // uses, since @input doesn't fire for programmatic value changes.
+              if (this._inputEl) {
+                  this._inputEl.style.height = 'auto';
+                  this._inputEl.style.height = Math.min(this._inputEl.scrollHeight, 120) + 'px';
+              }
+          };
+          recognition.onerror = (e) => {
+              // Denied mic permission is the one error worth explaining — the user has to
+              // fix it in browser settings, and the button would otherwise look broken.
+              if (e.error === 'not-allowed' || e.error === 'service-not-allowed') {
+                  this._error = 'Microphone access is blocked. Allow it in your browser settings to use voice input.';
+                  return;
+              }
+              // `no-speech` (silence) and `aborted` (user stopped) are normal; anything
+              // else is a transient service hiccup. onend still runs, so the mic resets.
+          };
+          // Fires on stop(), on error, and when the browser ends recognition on its own
+          // (e.g. a long silence) — the single place the live state is torn down.
+          recognition.onend = () => {
+              this._speechRecognition = null;
+              this._isDictating = false;
+          };
+          // start() is what triggers the browser's mic permission prompt.
+          this._speechRecognition = recognition;
+          this._isDictating = true;
+          this._error = '';
+          recognition.start();
+          this._inputEl?.focus();
+      }
       /** Insert `/code ` into the composer and focus it, ready for the message. */
       _handleSkillSelect(skill) {
           // A global skill must be installed before it can be chosen — inserting /code
@@ -6605,6 +6751,15 @@
           // the menu interaction, and the picker is a separate window so the menu would
           // otherwise linger open behind it (and after a file is chosen).
           this._skillMenuOpened = false;
+          // A lone `/` was only the means of reaching this action, so it shouldn't be
+          // left behind in the composer once the picker opens. Nothing else is touched:
+          // a real command (`/code`) stays because `/code` plus an attachment is a valid
+          // invocation, and the plus button reaches this same method with an arbitrary
+          // draft in the composer, which must survive.
+          if (this._inputValue.trim() === '/') {
+              this._inputValue = '';
+              this._saveState();
+          }
           this._pendingAttachAnalyze = aiAnalyze;
           this._fileInputEl?.click();
       }
@@ -6806,6 +6961,13 @@
                   previewUrl: p.previewUrl,
               }));
           }
+          // The message is on its way, so stop the mic if it's still live — otherwise
+          // it would keep transcribing into the now-cleared composer. abort() rather
+          // than stop() because a trailing pending utterance belongs to the message
+          // just sent; flushing it would drop stray words into the next one.
+          if (this._speechRecognition) {
+              this._speechRecognition.abort();
+          }
           this.messages = [...this.messages, userMessage];
           this._inputValue = '';
           this._saveState();
@@ -7159,6 +7321,122 @@
           }
       }
       // --- Render ---
+      /**
+       * The composer's controls: attach, the textarea, the mic, and send/stop.
+       *
+       * While dictating, only the mic swaps — to a stop button — so the textarea
+       * stays visible with the transcript streaming into it and send stays reachable.
+       */
+      _renderComposerControls() {
+          return b `
+      <button
+        class="input-plus"
+        @click=${this._handlePlusClick}
+      >+</button>
+      <textarea
+        class="input"
+        .value=${this._inputValue}
+        placeholder=${this.placeholder}
+        rows="1"
+        @input=${this._handleInputChange}
+        @keydown=${this._handleKeyDown}
+        @paste=${this._handlePaste}
+      ></textarea>
+      ${window.SpeechRecognition || window.webkitSpeechRecognition ? b `
+        ${this._isDictating ? b `
+          <button
+            class="mic-stop"
+            title="Stop dictation"
+            aria-label="Stop dictation"
+            @click=${this._handleMicClick}
+          >
+            <svg
+              xmlns="http://www.w3.org/2000/svg"
+              fill="currentColor"
+              viewBox="0 0 24 24"
+              class="mic-stop-icon"
+              aria-hidden="true"
+            >
+              <rect
+                x="6"
+                y="6"
+                width="12"
+                height="12"
+                rx="2"
+              />
+            </svg>
+          </button>
+        ` : b `
+          <button
+            class="mic"
+            title="Dictate a message"
+            aria-label="Dictate a message"
+            @click=${this._handleMicClick}
+          >
+            <svg
+              xmlns="http://www.w3.org/2000/svg"
+              viewBox="0 0 24 24"
+              fill="none"
+              stroke="currentColor"
+              stroke-width="1.75"
+              stroke-linecap="round"
+              stroke-linejoin="round"
+              class="mic-icon"
+              aria-hidden="true"
+            >
+              <path d="M12 15a3 3 0 0 0 3-3V6a3 3 0 0 0-6 0v6a3 3 0 0 0 3 3Z"/>
+              <path d="M19 11a7 7 0 0 1-14 0"/>
+              <path d="M12 18v3"/>
+            </svg>
+          </button>
+        `}
+      ` : A}
+      ${this._isStreaming ? b `
+        <button
+          class="stop"
+          @click=${this._handleStop}
+          title="Stop generating"
+        >
+          <svg
+            xmlns="http://www.w3.org/2000/svg"
+            fill="currentColor"
+            viewBox="0 0 24 24"
+            class="stop-icon"
+          >
+            <rect
+              x="6"
+              y="6"
+              width="12"
+              height="12"
+              rx="2"
+            />
+          </svg>
+        </button>
+      ` : b `
+        <button
+          class=${e({
+            'send': true,
+            'send--disabled': !this._inputValue.trim() && !this._pendingConversationFiles.length,
+        })}
+          @click=${this._handleSubmit}
+          title="Send message"
+        >
+          <svg
+            width="16"
+            height="16"
+            viewBox="0 0 24 24"
+            fill="none"
+            stroke="currentColor"
+            stroke-width="2.5"
+            stroke-linecap="round"
+            class="send-icon"
+          >
+            <path d="M12 19V5M5 12l7-7 7 7"/>
+          </svg>
+        </button>
+      `}
+    `;
+      }
       /** The `/` skill menu shown above the composer while typing a slash command. */
       _renderSkillMenu() {
           const skills = this._invocableSkills;
@@ -7865,63 +8143,7 @@
           />
           ${this._skillMenuOpened ? this._renderSkillMenu() : A}
           <div class="input-wrapper">
-            <button
-              class="input-plus"
-              @click=${this._handlePlusClick}
-            >+</button>
-            <textarea
-              class="input"
-              .value=${this._inputValue}
-              placeholder=${this.placeholder}
-              rows="1"
-              @input=${this._handleInputChange}
-              @keydown=${this._handleKeyDown}
-              @paste=${this._handlePaste}
-            ></textarea>
-            ${this._isStreaming ? b `
-              <button
-                class="stop"
-                @click=${this._handleStop}
-                title="Stop generating"
-              >
-                <svg
-                  xmlns="http://www.w3.org/2000/svg"
-                  fill="currentColor"
-                  viewBox="0 0 24 24"
-                  class="stop-icon"
-                >
-                  <rect
-                    x="6"
-                    y="6"
-                    width="12"
-                    height="12"
-                    rx="2"
-                  />
-                </svg>
-              </button>
-            ` : b `
-              <button
-                class=${e({
-            'send': true,
-            'send--disabled': !this._inputValue.trim() && !this._pendingConversationFiles.length,
-        })}
-                @click=${this._handleSubmit}
-                title="Send message"
-              >
-                <svg
-                  width="16"
-                  height="16"
-                  viewBox="0 0 24 24"
-                  fill="none"
-                  stroke="currentColor"
-                  stroke-width="2.5"
-                  stroke-linecap="round"
-                  class="send-icon"
-                >
-                  <path d="M12 19V5M5 12l7-7 7 7"/>
-                </svg>
-              </button>
-            `}
+            ${this._renderComposerControls()}
           </div>
         </div>
         ${this._skillBrowseOpened ? this._renderSkillBrowse() : A}
@@ -8721,6 +8943,71 @@
       color: #b0b0b0;
     }
 
+    /* margin-right rather than a wider .input-wrapper gap, which would also push
+       the plus button and textarea apart. Applied to both mic states so the
+       spacing doesn't shift when dictation swaps one for the other. */
+    .mic,
+    .mic-stop {
+      margin-right: 4px;
+    }
+
+    .mic {
+      width: 34px;
+      height: 34px;
+      border: none;
+      background: transparent;
+      color: var(--kr-chatbot-text);
+      border-radius: 50%;
+      cursor: pointer;
+      display: flex;
+      align-items: center;
+      justify-content: center;
+      flex-shrink: 0;
+      transition: background 0.15s, color 0.15s;
+    }
+
+    .mic:hover {
+      background: #f8f9fc;
+    }
+
+    .mic-icon {
+      width: 21px;
+      height: 21px;
+    }
+
+    /* Dictation swaps the mic for this stop button, sitting in the mic's slot so
+       the composer's layout is unchanged and the textarea stays visible with the
+       transcript streaming into it. Structurally identical to .stop; only the
+       colors differ — dictation isn't an error state, so it takes a neutral grey
+       instead of the error red, and that also keeps it distinct from the solid
+       send button sitting right beside it. The icon is dark rather than white
+       because white on this grey wouldn't meet contrast. */
+    .mic-stop {
+      width: 34px;
+      height: 34px;
+      border-radius: 50%;
+      background: #eae7e7;
+      color: #1f1f1f;
+      border: none;
+      cursor: pointer;
+      display: flex;
+      align-items: center;
+      justify-content: center;
+      flex-shrink: 0;
+      transition: all 0.15s;
+    }
+
+    .mic-stop:hover {
+      background: #ded9d9;
+    }
+
+    /* Larger than .stop-icon's 14px: the rect fills only half its 24-unit
+       viewBox, so the drawn square is half this value. */
+    .mic-stop-icon {
+      width: 20px;
+      height: 20px;
+    }
+
     .send {
       width: 34px;
       height: 34px;
@@ -9348,6 +9635,9 @@
   __decorate([
       r()
   ], exports.Chatbot.prototype, "_skillBrowseQuery", void 0);
+  __decorate([
+      r()
+  ], exports.Chatbot.prototype, "_isDictating", void 0);
   __decorate([
       e$3('.panel')
   ], exports.Chatbot.prototype, "_panelEl", void 0);