speccore 8.3.158 → 8.3.160

package/dist/commands/iteration/split.js

@@ -309,6 +309,7 @@ async function iterationSplitCommand(options) {
             if (hasPipeline) {
                 const { createSplitPipeline } = await Promise.resolve().then(() => __importStar(require('../../core/pipeline-engine')));
                 const { engine } = await createSplitPipeline(iter);
+                // v8.3.160+: 默认步骤隔离模式
                 const result = await engine.advance();
                 if (result.isComplete) {
                     logger_1.logger.success('🎉 Split Pipeline 完成!');
@@ -319,14 +320,47 @@ async function iterationSplitCommand(options) {
                     logger_1.logger.info('');
                     logger_1.logger.info(`🔄 Split Pipeline 推进到: ${result.nextStepName || result.nextStepId}`);
                     logger_1.logger.info('');
+                    // v8.3.160+: 步骤隔离模式输出标记 + 子 Agent 激活
+                    const nextCmd = `speccore iteration split -I ${iter} --resume --pipeline`;
+                    const prompt = await (0, prompt_builder_1.buildPrompt)('split', {
+                        iteration: iter,
+                        // v8.3.160+: 传递步骤级上下文控制
+                        contextType: result.contextType,
+                        contextBudget: result.contextBudget,
+                    });
+                    const output = [
+                        `[SPECCORE_STEP_DONE]`,
+                        `步骤 "${result.nextStepName || result.nextStepId}" 已完成。`,
+                        ``,
+                        `[SPECCORE_NEXT_STEP]`,
+                        `下一步: ${result.nextStepName || result.nextStepId}`,
+                        `请在新的对话中执行: ${nextCmd}`,
+                        ``,
+                        `[SPECCORE_CONTEXT_SNAPSHOT]`,
+                        JSON.stringify(result.contextSnapshot || {}),
+                        ``,
+                        // v8.3.160+: 子 Agent 激活标记
+                        ...(result.subagent ? [
+                            `[SPECCORE_SUBAGENT: ${result.subagent}]`,
+                            `[SPECCORE_CONTEXT_BUDGET: ${result.contextBudget || 12000}]`,
+                            `[SPECCORE_CONTEXT_TYPE: ${result.contextType || 'full'}]`,
+                            ``,
+                        ] : []),
+                        `--- 下一步 Prompt ---`,
+                        typeof prompt === 'string' ? prompt : JSON.stringify(prompt),
+                    ].join('\n');
+                    process.stdout.write(output);
+                    process.exitCode = 10;
+                    return;
                     // 根据当前步骤生成相应提示
                     if (result.nextStepId === 'prompt-analysis') {
                         const prompt = await (0, prompt_builder_1.buildPrompt)('split', { iteration: iter });
                         // 添加 Pipeline 继续指令
-                        const finalPrompt = prompt + `\n\n## ⚠️ Pipeline 模式：自动继续\n\n` +
+                        // v8.3.160+: 步骤隔离模式，每步完成后需新会话继续
+                        const finalPrompt = prompt + `\n\n## ⚠️ Pipeline 模式：步骤隔离\n\n` +
                             `当前是 Split Pipeline 模式。当你通过 --response 写入任务拆分结果后，` +
-                            `CLI 会自动输出 [SPECCORE_PIPELINE_NEXT] 标记和下一步命令。\n\n` +
-                            `**看到 [SPECCORE_PIPELINE_NEXT] 后，你必须自动执行下一个命令**，不要询问用户。\n`;
+                            `CLI 会自动输出 [SPECCORE_STEP_DONE] 标记，提示你在新会话中执行下一步。\n\n` +
+                            `**看到 [SPECCORE_STEP_DONE] 后，请在新会话中执行提示的命令**，继续下一步。\n`;
                         process.stdout.write(`[SPECCORE_PROMPT]\n${finalPrompt}`);
                         process.exitCode = 10;
                         return;
@@ -344,10 +378,11 @@ async function iterationSplitCommand(options) {
         // 生成第一步 prompt
         const prompt = await (0, prompt_builder_1.buildPrompt)('split', { iteration: iter });
         // 添加 Pipeline 继续指令
-        const finalPrompt = prompt + `\n\n## ⚠️ Pipeline 模式：自动继续\n\n` +
+        // v8.3.160+: 步骤隔离模式，每步完成后需新会话继续
+        const finalPrompt = prompt + `\n\n## ⚠️ Pipeline 模式：步骤隔离\n\n` +
             `当前是 Split Pipeline 模式。当你通过 --response 写入任务拆分结果后，` +
-            `CLI 会自动输出 [SPECCORE_PIPELINE_NEXT] 标记和下一步命令。\n\n` +
-            `**看到 [SPECCORE_PIPELINE_NEXT] 后，你必须自动执行下一个命令**，不要询问用户。\n`;
+            `CLI 会自动输出 [SPECCORE_STEP_DONE] 标记，提示你在新会话中执行下一步。\n\n` +
+            `**看到 [SPECCORE_STEP_DONE] 后，请在新会话中执行提示的命令**，继续下一步。\n`;
         process.stdout.write(`[SPECCORE_PROMPT]\n${finalPrompt}`);
         process.exitCode = 10;
         return;
@@ -372,7 +407,8 @@ async function iterationSplitCommand(options) {
         if (hasPipeline) {
             const { createSplitPipeline } = await Promise.resolve().then(() => __importStar(require('../../core/pipeline-engine')));
             const { engine } = await createSplitPipeline(iter);
-            await engine.advance();
+            // v8.3.160+: 默认步骤隔离模式
+            const stepResult = await engine.advance();
             const state = await engine.getState();
             if (state?.currentStep === 'done') {
                 logger_1.logger.success('🎉 Split Pipeline 完成!');
@@ -389,10 +425,38 @@ async function iterationSplitCommand(options) {
                     nextPrompt = `任务拆分已完成，正在创建任务目录结构...\n\n请继续执行后续操作。`;
                 }
                 else {
-                    const promptResult = await (0, prompt_builder_1.buildPrompt)('split', { iteration: iter });
+                    // v8.3.160+: 传递步骤级上下文控制，确保 Pipeline 配置的 budget/type 生效
+                    const promptResult = await (0, prompt_builder_1.buildPrompt)('split', {
+                        iteration: iter,
+                        contextType: stepResult.contextType,
+                        contextBudget: stepResult.contextBudget,
+                    });
                     nextPrompt = typeof promptResult === 'string' ? promptResult : JSON.stringify(promptResult);
                 }
-                process.stdout.write(`[SPECCORE_PIPELINE_NEXT]\n${nextPrompt}`);
+                // v8.3.160+: 步骤隔离模式输出标记 + 子 Agent 激活
+                const nextCmd = `speccore iteration split -I ${iter} --resume --pipeline`;
+                const output = [
+                    `[SPECCORE_STEP_DONE]`,
+                    `步骤 "${state.currentStep}" 已完成。`,
+                    ``,
+                    `[SPECCORE_NEXT_STEP]`,
+                    `下一步: ${state.currentStep}`,
+                    `请在新的对话中执行: ${nextCmd}`,
+                    ``,
+                    `[SPECCORE_CONTEXT_SNAPSHOT]`,
+                    JSON.stringify(stepResult.contextSnapshot || {}),
+                    ``,
+                    // v8.3.160+: 子 Agent 激活标记
+                    ...(stepResult.subagent ? [
+                        `[SPECCORE_SUBAGENT: ${stepResult.subagent}]`,
+                        `[SPECCORE_CONTEXT_BUDGET: ${stepResult.contextBudget || 12000}]`,
+                        `[SPECCORE_CONTEXT_TYPE: ${stepResult.contextType || 'full'}]`,
+                        ``,
+                    ] : []),
+                    `--- 下一步 Prompt ---`,
+                    nextPrompt,
+                ].join('\n');
+                process.stdout.write(output);
                 process.exitCode = 10;
             }
             return;