orchestrix-yuri 4.8.3 → 4.8.5

package/lib/gateway/engine/messages.js

@@ -52,8 +52,8 @@ const MESSAGES = {
     zh: '🎉 规划完成！{summary}\n\n下一步：检查上面的文档，然后运行 `*develop` 开始开发。',
   },
   dev_complete: {
-    en: '🎉 Development complete! All stories implemented.\n\nNext: run `*test` to validate each epic with smoke tests.',
-    zh: '🎉 开发完成！所有 story 已实现。\n\n下一步：运行 `*test` 对每个 epic 进行冒烟测试。',
+    en: '🎉 Development complete! All stories implemented.\n\nAuto-starting smoke tests...',
+    zh: '🎉 开发完成！所有 story 已实现。\n\n自动启动冒烟测试...',
   },
   test_all_passed: {
     en: '\n🚀 All tests passed! Run `*deploy` when ready.',

package/lib/gateway/engine/phase-orchestrator.js

@@ -887,6 +887,16 @@ class PhaseOrchestrator {
 
     log.engine('Dev phase complete');
     this.onComplete('develop', msg('dev_complete'));
+
+    // Auto-chain: start test phase immediately after dev completes.
+    // This makes "develop all stories then test" a single unattended flow.
+    try {
+      const result = this.startTest(this._projectRoot);
+      log.engine(`Auto-chained to test phase: ${result.slice(0, 80)}`);
+      this.onProgress(result);
+    } catch (err) {
+      log.warn(`Auto-chain to test failed: ${err.message}`);
+    }
   }
 
   // ── Test Phase ──────────────────────────────────────────────────────────────

package/lib/gateway/router.js

@@ -307,6 +307,30 @@ class Router {
         log.router(`Dispatcher unavailable, defaulting to change for: "${msg.text.slice(0, 50)}..."`);
       }
 
+      // Context-aware override: if user sends a short confirmation and context
+      // clearly suggests a specific phase action, override dispatcher's classification.
+      if (classified.action === 'change' && lastReply) {
+        const userText = msg.text.trim().toLowerCase();
+        const isConfirmation = /^(启动|开始|go|ok|好|好的|批准|确认|start|begin|launch|yes|可以|干吧|搞起|冲|执行)$/i.test(userText)
+          || userText.length <= 6;
+        if (isConfirmation) {
+          const ctx = lastReply.toLowerCase();
+          if (ctx.includes('启动开发') || ctx.includes('start dev') || ctx.includes('*develop') || ctx.includes('开始开发') || ctx.includes('agent') && ctx.includes('开始工作')) {
+            classified.action = 'develop';
+            classified.reasoning = 'confirmation of dev-start suggestion from context';
+            log.router(`Override: change → develop (user confirmed dev-start from context)`);
+          } else if (ctx.includes('*test') || ctx.includes('运行测试') || ctx.includes('冒烟测试')) {
+            classified.action = 'test';
+            classified.reasoning = 'confirmation of test suggestion from context';
+            log.router(`Override: change → test (user confirmed test from context)`);
+          } else if (ctx.includes('*deploy') || ctx.includes('部署') || ctx.includes('上线')) {
+            classified.action = 'deploy';
+            classified.reasoning = 'confirmation of deploy suggestion from context';
+            log.router(`Override: change → deploy (user confirmed deploy from context)`);
+          }
+        }
+      }
+
       if (classified.action !== 'conversation') {
         try {
           const intentResult = await this._executeClassifiedIntent(classified, msg);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "orchestrix-yuri",
-  "version": "4.8.3",
+  "version": "4.8.5",
   "description": "Yuri — Meta-Orchestrator for Orchestrix. Drive your entire project lifecycle with natural language.",
   "main": "lib/installer.js",
   "bin": {

package/skill/resources/dispatcher-prompt.txt

@@ -7,15 +7,15 @@ Messages may include "[CONTEXT]" with the previous assistant response for refere
 
 | Action | When to use | Examples |
 |--------|-------------|---------|
-| bugfix | User reports a bug, error, or specific problem that needs fixing | "这个 bug 修一下", "Fix the login error", "页面报错了", "按钮点了没反应", "这个问题处理一下" |
-| change | User wants work done: modify, add, redesign, create a plan, write something, have an agent do work (NOT a bug) | "让 UX 重新设计界面", "Add dark mode", "先做个完整的设计 plan", "写一份方案", "从 PO 开始改" |
-| plan | User explicitly wants to START the planning phase from scratch | "开始规划", "*plan" |
-| develop | User explicitly wants to START development phase | "开始开发", "*develop" |
-| test | User wants to run tests | "跑一下测试", "冒烟测试" |
+| bugfix | User reports a bug, error, or specific problem that needs fixing | "这个 bug 修一下", "Fix the login error", "页面报错了", "按钮点了没反应" |
+| change | User wants NEW work done: modify, add, redesign, create a plan, write something | "让 UX 重新设计界面", "Add dark mode", "写一份方案" |
+| plan | User wants to start or restart planning | "开始规划", "*plan", "重新规划" |
+| develop | User wants to start, resume, or continue development | "开始开发", "*develop", "启动开发", "继续开发", "让 Agent 开始工作" |
+| test | User wants to run tests | "跑一下测试", "冒烟测试", "*test" |
 | deploy | User wants to deploy or release | "部署到线上", "发布", "上线" |
-| status | User asks about progress, previous results, or what happened | "之前的方案呢？", "进展如何？", "怎么样了", "结果出来了吗" |
+| status | User asks about progress or what happened | "进展如何？", "怎么样了", "结果出来了吗" |
 | iterate | User wants a new iteration cycle | "新迭代", "下一轮" |
-| conversation | ONLY for pure questions or opinions with NO actionable request | "你觉得怎么样？", "这个架构好不好？", "解释一下这段代码" |
+| conversation | ONLY for pure questions or opinions with NO actionable request | "你觉得怎么样？", "解释一下这段代码" |
 
 ## Response Format
 
@@ -23,14 +23,31 @@ Reply with EXACTLY this JSON format on a single line, nothing else:
 
 {"action":"<action>","description":"<brief description for the agent>","reasoning":"<one sentence why>"}
 
-## Critical Rules
+## Critical Rules (read ALL before classifying)
+
+### Rule 1: CONTEXT determines the action for confirmations
+If the user sends a short confirmation ("启动", "ok", "开始", "go", "批准", "好的", "确认"), look at [CONTEXT] to determine WHAT they're confirming:
+- Context mentions starting development/coding → "develop"
+- Context mentions running tests → "test"
+- Context mentions deploying → "deploy"
+- Context mentions planning → "plan"
+- Context mentions a specific change/redesign → "change"
+- No clear context → "change" (default)
+
+### Rule 2: Bug vs Change
+Bug, error, problem, something broken → "bugfix". New work, redesign, feature → "change".
+
+### Rule 3: Bias toward action
+If the user wants ANYTHING done, classify as an action (change/bugfix/develop/test/etc). Never let work requests fall into "conversation".
+
+### Rule 4: Agent references
+If the user names an agent (UX, architect, PM, dev, SM, QA) and wants them to do something → "change".
+
+### Rule 5: "conversation" is rare
+Only for messages that require NO action — pure questions, opinions, greetings. When in doubt, prefer an action.
+
+### Rule 6: "develop" is broader than you think
+Any of these mean "develop": 启动, 开始干活, 让他们工作, 跑起来, 开工, start working, begin, launch, kick off development. If user wants agents to start executing stories → "develop".
 
-- **Bug vs Change**: If the user reports a bug, error, problem, or something broken → "bugfix". If the user wants new work, redesign, or feature → "change". When in doubt between bugfix and change, prefer "bugfix" for problems and "change" for new work.
-- **Bias toward action**: If the user wants ANYTHING done (create, write, fix, modify, redesign, evaluate, plan something), classify as "change" or "bugfix". Never let work requests fall into "conversation".
-- If the user is replying to a question with a choice/decision (e.g., "先做 plan" answering "要 A 还是 B？"), that reply IS an instruction — classify as "change", NOT "conversation".
-- If the user references a specific agent role (UX, UI, architect, PM, QA, dev, SM) and wants them to do something, classify as "change".
-- If the user asks about previous results, output, or progress, classify as "status".
-- "conversation" is ONLY for messages that require NO action at all — pure questions, opinions, greetings.
-- When in doubt between "change" and "conversation", ALWAYS choose "change".
 - The "description" field should be a concise summary of what the user wants done.
 - ALWAYS respond with the JSON format above. Never add explanation, markdown, or anything else.