cortico 0.1.0 → 0.1.2

package/src/worlds/minecraft/until.ts

@@ -0,0 +1,75 @@
+/**
+ * 早停名单:find 与 tunnel 声明的 `until`,把类别名摊成方块 id,一路上撞见就收工。
+ *
+ * 只认名单与判定,收不收工由技能自己决定。
+ */
+import type { Bot } from 'mineflayer';
+import { UNTIL_CATEGORIES } from './skills.ts';
+import { matchBlockIds } from './travel.ts';
+import { canSeeBlockAt } from './terrain.ts';
+import { zhName } from './names.ts';
+
+/**
+ * `until` 的早停名单 → 方块 id 集合。
+ *
+ * `#类别` 先探一次 registry 的 tag 数据面(本仓库钉住的版本没有这一面,留着是为了
+ * 将来有了就自动走过去),取不到落到 `UNTIL_CATEGORIES` 那几个内置类别。
+ * 认不出的名字不抛错 —— 早停是这一单的副条件,它认不出来不该把整步判死;
+ * 认不出哪几个由回执点名(静默吃掉参数是这条链上最贵的一类失败)。
+ */
+export function untilBlockIds(bot: Bot, until: readonly string[]): { ids: number[]; unknown: string[] } {
+  const byName = bot.registry.blocksByName as Record<string, { id: number; name: string }>;
+  const tags = (bot.registry as unknown as { blockTags?: Record<string, string[]> }).blockTags;
+  const ids = new Set<number>();
+  const unknown: string[] = [];
+  for (const raw of until) {
+    const name = raw.replace(/^minecraft:/, '');
+    if (name.startsWith('#')) {
+      const bare = name.slice(1);
+      const tagged = tags?.[bare] ?? tags?.[`minecraft:${bare}`] ?? null;
+      if (tagged) {
+        for (const n of tagged) {
+          const b = byName[n.replace(/^minecraft:/, '')];
+          if (b) ids.add(b.id);
+        }
+        continue;
+      }
+      const pred = UNTIL_CATEGORIES[bare];
+      if (!pred) { unknown.push(name); continue; }
+      for (const b of Object.values(byName)) if (pred(b.name)) ids.add(b.id);
+      continue;
+    }
+    const matched = matchBlockIds(bot, name);
+    if (matched.length === 0) { unknown.push(name); continue; }
+    for (const id of matched) ids.add(id);
+  }
+  return { ids: [...ids], unknown };
+}
+
+/** `until` 命中的那一格 */
+export interface UntilHit { x: number; y: number; z: number; what: string }
+
+/**
+ * 周身有没有碰到早停名单里的东西。
+ *
+ * `visible` = 只认看得见的:行军途中用这一档(与 find 自己的感知规则同源)。
+ * 挖通道那一档不设视线闸 —— 铲子下去露出来的那一面本来就贴着脸,视线判据在坑里没意义。
+ */
+export function untilHit(bot: Bot, ids: readonly number[], radius: number, visible: boolean): UntilHit | null {
+  if (ids.length === 0) return null;
+  const found = bot.findBlocks({ matching: [...ids], maxDistance: radius, count: 16 });
+  const p = found.find((q) => !visible || canSeeBlockAt(bot, q));
+  if (!p) return null;
+  return { x: p.x, y: p.y, z: p.z, what: zhName(bot.blockAt(p)?.name ?? 'unknown') };
+}
+
+/** 行军途中每走完一段扫一次早停的半径(格) */
+export const UNTIL_TRAVEL_RADIUS = 16;
+/** 挖通道每挖完一格扫一次早停的半径(格):坑壁上露出来的那一圈 */
+export const UNTIL_DIG_RADIUS = 4;
+
+/** 认不出的早停名字那一句;全认得出返回空串 */
+export function untilUnknownNote(unknown: readonly string[]): string {
+  return unknown.length > 0 ? `(until 里的 ${unknown.join('、')} 认不出来,这几样没算进去)` : '';
+}
+

package/src/worlds/qq/normalize.ts

@@ -178,6 +178,9 @@ export function renderSegmentsPlain(segments: Segment[]): string {
       case 'image':
         parts.push({ text: '[图片]', isToken: true });
         break;
+      case 'record':
+        parts.push({ text: '[语音]', isToken: true });
+        break;
       case 'face': {
         const name = QFACE_NAMES[String(data.id ?? '')];
         parts.push({ text: name ? `[表情:${name}]` : '[QQ表情]', isToken: true });
@@ -253,6 +256,11 @@ export function renderIncoming(
         parts.push({ text: renderImage(data), isToken: true });
         break;
       }
+      // 语音条的音频不随消息到达,取回它要另一次动作调用;这里只说这是一条语音。
+      case 'record': {
+        parts.push({ text: '[语音]', isToken: true });
+        break;
+      }
       case 'face': {
         const name = QFACE_NAMES[String(data.id ?? '')];
         parts.push({ text: name ? `[表情:${name}]` : '[QQ表情]', isToken: true });
@@ -301,14 +309,20 @@ interface OutgoingArgs {
   reply_to_message_id?: number | string;
   /** 图片base64(不含data:前缀);有则编译成image段(OneBot base64://) */
   image_base64?: string;
+  /** 语音base64(不含data:前缀);有则整条消息只有这一个record段 */
+  record_base64?: string;
 }
 
 /**
  * 出站 → segment数组:reply段(如有)在前,然后image段(如有),最后text段
  * (text非空才加——支持只发图不带字)。
+ * 语音条独占一条消息:QQ 的 record 段不与别的段同发,给了它就只发它。
  * 不做 @ 编译：agent 写出的“@某人”作为纯文本原样发出。
  */
 export function buildOutgoing(args: OutgoingArgs): Segment[] {
+  if (args.record_base64) {
+    return [{ type: 'record', data: { file: `base64://${args.record_base64}` } }];
+  }
   const segments: Segment[] = [];
   if (args.reply_to_message_id !== undefined) {
     segments.push({ type: 'reply', data: { id: String(args.reply_to_message_id) } });

package/src/worlds/qq/world.ts

@@ -109,6 +109,8 @@ interface PendingDraft {
   replyMessageId?: number | string;
   /** 要发送的图片的句柄(`log:` 她看见过的 / `mem:` 她收藏的);无=纯文本 */
   imageRef?: string;
+  /** 要发送的语音的句柄;语音条独占一条消息,与 text、imageRef 互斥 */
+  voiceRef?: string;
 }
 
 interface ForwardNode {
@@ -1235,7 +1237,14 @@ export class QQWorld implements World {
   /** 实际发送到目标会话,回录 qq.self,返回回执文本 */
   private async sendToTarget(
     target: Conv,
-    opts: { text: string; replyMessageId?: number | string; imageRef?: string; imageBytes?: Uint8Array },
+    opts: {
+      text: string;
+      replyMessageId?: number | string;
+      imageRef?: string;
+      imageBytes?: Uint8Array;
+      voiceRef?: string;
+      voiceBytes?: Uint8Array;
+    },
   ): Promise<string> {
     const host = this.host!;
     const driver = this.driver!;
@@ -1247,6 +1256,7 @@ export class QQWorld implements World {
       text: opts.text,
       reply_to_message_id: opts.replyMessageId,
       image_base64: imageBase64,
+      ...(opts.voiceBytes ? { record_base64: Buffer.from(opts.voiceBytes).toString('base64') } : {}),
     });
     const action = target.kind === 'group' ? 'send_group_msg' : 'send_private_msg';
     const params =
@@ -1259,7 +1269,13 @@ export class QQWorld implements World {
     // 回录正文:文字 + (如有)图片标记,保证自己发的图在历史里也可见。
     // 行首保留平台消息号,使历史查询结果可引用自身消息。
     const imgMark = opts.imageRef ? `[图片: ${opts.imageRef}]` : '';
-    const selfBody = [opts.text, imgMark].filter(Boolean).join(' ');
+    const voiceMark = opts.voiceRef ? `[语音: ${opts.voiceRef}]` : '';
+    const selfBody = [opts.text, imgMark, voiceMark].filter(Boolean).join(' ');
+    const selfBlob = opts.imageRef
+      ? { handle: opts.imageRef, fallbackText: '你发出的图片' }
+      : opts.voiceRef
+        ? { handle: opts.voiceRef, fallbackText: '你发出的语音' }
+        : null;
     const idTag = messageId !== undefined ? `#${messageId} ` : '';
     const env = await host.pushEvent(
       {
@@ -1267,8 +1283,8 @@ export class QQWorld implements World {
         ts: nowIso(this.timezone),
         source: this.id,
         text: `${idTag}[${this.convLabel(target)} ${shortTime(this.timezone)}] 你: ${selfBody}`,
-        // 发出去的图按句柄附在自己这条记录上:时间线与吃图的模型都看得到发了什么
-        ...(opts.imageRef ? { blobs: [{ handle: opts.imageRef, fallbackText: '你发出的图片' }] } : {}),
+        // 发出去的图和语音按句柄附在自己这条记录上:时间线与吃这些类型的模型都看得到发了什么
+        ...(selfBlob ? { blobs: [selfBlob] } : {}),
         senderKey: String(driver.identity?.selfId ?? 'self'),
         meta: {
           message_id: messageId,
@@ -1320,6 +1336,13 @@ export class QQWorld implements World {
               'Optional. An image to send, given by handle: a log: handle from a [blob ...] line you saw, '
               + 'or a mem: handle of one kept in your workspace.',
           },
+          voice: {
+            type: 'string',
+            description:
+              'Optional. Audio to send as a QQ voice bar, given by handle the same way as image. '
+              + 'A voice bar is a message of its own: leave text, image and reply_to out. '
+              + 'The protocol end converts the audio it accepts (wav, mp3) to what QQ plays.',
+          },
           reply_to: {
             type: 'string',
             description:
@@ -1335,7 +1358,12 @@ export class QQWorld implements World {
         const rawText = typeof args.text === 'string' ? args.text : '';
         const imageArg =
           typeof args.image === 'string' && args.image.trim() ? args.image.trim() : undefined;
-        if (!rawText.trim() && !imageArg) return '[bad input] provide at least one of text or image';
+        const voiceArg =
+          typeof args.voice === 'string' && args.voice.trim() ? args.voice.trim() : undefined;
+        if (!rawText.trim() && !imageArg && !voiceArg) return '[bad input] provide at least one of text, image or voice';
+        if (voiceArg && (rawText.trim() || imageArg || (args.reply_to !== undefined && args.reply_to !== null && String(args.reply_to) !== ''))) {
+          return '[bad input] a voice bar is a message of its own; it carries no text, image or quote-reply';
+        }
         // 纯空白正文视为不带字(避免只发图时混进一个空白 text 段)
         const text = rawText.trim() ? rawText : '';
 
@@ -1373,7 +1401,17 @@ export class QQWorld implements World {
             : `[will send image: ${imageArg}]`;
         }
 
-        this.pendingDraft = { target, text, replyMessageId: rep.mid, imageRef };
+        // 语音:同样按句柄确认取得到,MIME 必须是音频
+        let voiceRef: string | undefined;
+        if (voiceArg) {
+          const got = host.blob(voiceArg);
+          if (!got) return `[send failed] no blob for ${voiceArg}; use a log: handle from a [blob ...] line or a mem: handle from your workspace`;
+          if (!got.mime.startsWith('audio/')) return `[send failed] ${voiceArg} is ${got.mime}, not audio`;
+          voiceRef = voiceArg;
+          previewLine = `[will send voice: ${voiceArg}]`;
+        }
+
+        this.pendingDraft = { target, text, replyMessageId: rep.mid, imageRef, voiceRef };
 
         // 排出当下已积、尚未投递的QQ会话事件(consume-once)。正文不能伪装成
         // draft 工具结果:交回 loop 按常规投递,本轮下一次推理前它们就在眼前,
@@ -1382,7 +1420,7 @@ export class QQWorld implements World {
         ctx.queueExternalEvents?.(drained);
 
         const lines = [`[draft staged → ${this.targetDesc(target)}]`];
-        lines.push(text.trim() ? `Text: ${text}` : '(image only, no text)');
+        lines.push(text.trim() ? `Text: ${text}` : voiceRef ? '(voice bar, no text)' : '(image only, no text)');
         if (previewLine) lines.push(previewLine);
         lines.push(
           '',
@@ -1433,11 +1471,19 @@ export class QQWorld implements World {
               if (!got) return `[send failed] blob ${draft.imageRef} is gone (draft kept)`;
               imageBytes = got.bytes;
             }
+            let voiceBytes: Uint8Array | undefined;
+            if (draft.voiceRef) {
+              const got = host.blob(draft.voiceRef);
+              if (!got) return `[send failed] blob ${draft.voiceRef} is gone (draft kept)`;
+              voiceBytes = got.bytes;
+            }
             const receipt = await this.sendToTarget(draft.target, {
               text: draft.text,
               replyMessageId: draft.replyMessageId,
               imageRef: draft.imageRef,
               imageBytes,
+              voiceRef: draft.voiceRef,
+              voiceBytes,
             });
             this.pendingDraft = null; // 成功才清;失败保留草稿,可再 confirm 重试
             return receipt;

package/src/worlds/terminal/world.ts

@@ -251,6 +251,8 @@ interface TerminalWorldOptions {
 
 const NAME_MAX = 32;
 const TEXT_MAX = 4000;
+/** 按钮标签进事件正文前的长度上限。 */
+const LABEL_MAX = 40;
 
 /** 一条消息最多带几张图,单张解码后的字节上限。与面板侧的上限一致。 */
 const IMAGES_MAX = 8;
@@ -606,8 +608,8 @@ export class TerminalWorld implements World {
         this.sendJson(client.peer, { type: 'sys', text: t.botNotConnected });
         return;
       }
-      // 事件引用单行标签；没有标签时不产生事件。
-      const label = typeof msg.label === 'string' ? msg.label.replace(/\s+/g, ' ').trim() : '';
+      // 标签逐字进事件正文,所以压成一行并截断;没有标签就没有可引用的事实,这一帧作废。
+      const label = typeof msg.label === 'string' ? msg.label.replace(/\s+/g, ' ').trim().slice(0, LABEL_MAX) : '';
       if (!label) return;
       const host = this.host;
       host.pushEvent(
@@ -628,7 +630,7 @@ export class TerminalWorld implements World {
     this.sendJson(client.peer, { type: 'sys', text: t.unknownType(String(msg.type)) });
   }
 
-  /** 已保存的终端事件中是否有发言。 */
+  /** 已保存的终端事件里有没有人说过话。 */
   private spokenBefore(): boolean {
     if (!this.host) return false;
     return this.host.store