@bolloon/bolloon-agent 0.3.15 → 0.3.17

package/dist/agents/pi-sdk.js

@@ -41,7 +41,8 @@ import { onPostToolUse } from '../bootstrap/lifecycle-hooks.js';
 import { budgetReduce, snip, microcompact } from '../context-compaction/index.js';
 // React Harness: 8-gate + 4-guard (防越权 / 防 prompt 注入)
 import { ReactHarness } from '../security/react-harness.js';
-import { parseToolCall as parseToolCallImpl, isFinalResponse as isFinalResponseImpl, extractFinalAnswer as extractFinalAnswerImpl } from './parse-tool-call.js';
+import { parseToolCall as parseToolCallImpl, parseAllToolCalls, isFinalResponse as isFinalResponseImpl, extractFinalAnswer as extractFinalAnswerImpl } from './parse-tool-call.js';
+import { buildObservation, buildReflection, formatObservationWithReflection } from './error-classifier.js';
 import { sessionStore as defaultSessionStore } from './session-store.js';
 import { ToolRegistry } from './tool-registry.js';
 import { decideMaxIterations, decideContextOverflow, shouldCompactBeforeIteration } from './react-loop.js';
@@ -1127,303 +1128,235 @@ ${toolDefs}
             // Bug 5 (2026-07-17): 优先用 LLM 的 native tool_calls (response.toolCalls), 再回退到文本解析
             //   deepseek-v4-flash 用 OpenAI 协议 tools 时, 会真返回结构化 tool_calls 数组
             //   之前 nativeToolCalls 被读了不用, 只查 reply 文本, 导致 LLM 明明选了工具但代码找不到
-            let toolCall = null;
+            // 2026-07-28: 修复多工具调用 — 收集 ALL tool calls, 顺序执行后一次性返回
+            let toolCalls = [];
+            // 路径 A: native OpenAI 协议 tool_calls (可能多个)
             if (nativeToolCalls && nativeToolCalls.length > 0) {
-                const nc = nativeToolCalls[0];
-                // OpenAI 协议: { id, type: 'function', function: { name, arguments: JSON string } }
-                // 转换成 internal { name, args, id }
-                try {
-                    const args = typeof nc.function?.arguments === 'string'
-                        ? JSON.parse(nc.function.arguments)
-                        : (nc.function?.arguments || {});
-                    toolCall = {
-                        name: nc.function?.name,
-                        args,
-                        id: nc.id || `call_${Date.now()}_${Math.random().toString(36).slice(2, 8)}`,
-                    };
-                    console.log(`[PiAgent] 用 native tool_call: ${toolCall.name} (id=${toolCall.id})`);
-                }
-                catch (err) {
-                    console.warn(`[PiAgent] 解析 native tool_call 失败, 回退到文本解析: ${err.message?.slice(0, 100)}`);
-                    toolCall = null;
+                for (const nc of nativeToolCalls) {
+                    try {
+                        const args = typeof nc.function?.arguments === 'string'
+                            ? JSON.parse(nc.function.arguments)
+                            : (nc.function?.arguments || {});
+                        toolCalls.push({
+                            name: nc.function?.name,
+                            args,
+                            id: nc.id || `call_${Date.now()}_${Math.random().toString(36).slice(2, 8)}`,
+                        });
+                    }
+                    catch (err) {
+                        console.warn(`[PiAgent] 解析 native tool_call 失败: ${err.message?.slice(0, 100)}`);
+                    }
                 }
             }
-            if (!toolCall) {
-                toolCall = this.parseToolCall(reply);
-            }
-            // 2026-06-30 修: 给 toolCall 分配稳定 id, 让后续 tool result 能引用同一个 id
-            //   OpenAI 协议要求 messages 里 tool result 必须有对应的 tool_call_id, 否则 400
-            if (toolCall && !toolCall.id) {
-                toolCall.id = `call_${Date.now()}_${Math.random().toString(36).slice(2, 8)}`;
+            // 路径 B: 文本解析 (parseAllToolCalls 收集全部)
+            if (toolCalls.length === 0) {
+                const knownTools = new Set(Array.from(this.tools.keys()));
+                toolCalls = parseAllToolCalls(reply, { tools: knownTools });
+            }
+            // 回退路径 C: 原生 parseToolCall (单个)
+            if (toolCalls.length === 0) {
+                const single = this.parseToolCall(reply);
+                if (single)
+                    toolCalls.push(single);
+            }
+            // 给每个 toolCall 分配稳定 id
+            for (const tc of toolCalls) {
+                if (!tc.id) {
+                    tc.id = `call_${Date.now()}_${Math.random().toString(36).slice(2, 8)}`;
+                }
             }
-            if (toolCall) {
+            if (toolCalls.length > 0) {
+                // 把原始 LLM 回复 push 进 history (仅一次)
                 this.messageHistory.push({
                     role: 'assistant',
                     content: reply,
-                    toolCall
+                    toolCalls: toolCalls.length > 1 ? toolCalls : [toolCalls[0]],
                 });
-                // 通知前端：检测到工具调用
-                if (onStream) {
-                    onStream({ type: 'tool', content: `🔧 调用工具: ${toolCall.name}`, tool: toolCall.name });
-                    if (toolCall.args && Object.keys(toolCall.args).length > 0) {
-                        onStream({ type: 'status', content: `📋 参数: ${JSON.stringify(toolCall.args)}`, tool: toolCall.name });
-                    }
-                    // 2026-06-15: step-timeline 状态机 — 开新节点
-                    onStream({
-                        type: 'step_start',
-                        content: `调用 ${toolCall.name}`,
-                        tool: toolCall.name,
-                        args: toolCall.args || {},
-                    });
-                }
-                const tool = this.tools.get(toolCall.name);
-                if (!tool) {
-                    consecutiveErrors++;
-                    // 2026-06-16 新增: 未知工具也要累计 (LLM 幻觉高频场景)
-                    totalErrors++;
-                    const errorResult = { success: false, error: `未知工具: ${toolCall.name}` };
-                    this.messageHistory.push({ role: 'tool', content: JSON.stringify(errorResult), toolResult: errorResult });
-                    this.logToHarness(toolCall.name, toolCall.args, errorResult);
-                    console.warn(`[PiAgent] 未知工具: ${toolCall.name} (累计 ${totalErrors}/${this.MAX_TOTAL_ERRORS})，跳过并继续`);
-                    continue;
-                }
-                // Bootstrap PreToolUse hook: 调工具前校验 (危险命令拦截)
-                // 失败静默 — hook 自身挂掉 = 放行
-                // P2: 透传 permissionMode (从 BootstrapOptions / env BOLLOON_PERM_MODE 解析)
-                let toolToExecute = tool;
-                try {
-                    const pre = await onPreToolUse({
-                        tool: toolCall.name,
-                        args: toolCall.args || {},
-                        permissionMode: this.currentPermissionMode,
-                    });
-                    if (!pre.allowed) {
-                        const deniedResult = {
-                            success: false,
-                            error: `PreToolUse 拒绝: ${pre.reason || '未通过安全校验'}`,
-                        };
-                        this.messageHistory.push({
-                            role: 'tool',
-                            content: JSON.stringify(deniedResult),
-                            toolResult: deniedResult,
-                        });
-                        this.logToHarness(toolCall.name, toolCall.args, deniedResult);
-                        if (onStream) {
-                            onStream({
-                                type: 'error',
-                                content: `🛡️ PreToolUse 拒绝 ${toolCall.name}: ${pre.reason || '安全校验失败'}`,
-                                tool: toolCall.name,
-                            });
-                            // 2026-06-15: step-timeline — 拦在 PreToolUse, 标 step_error
-                            onStream({
-                                type: 'step_error',
-                                content: `PreToolUse 拒绝 ${toolCall.name}`,
-                                tool: toolCall.name,
-                                error: pre.reason || '安全校验失败',
-                            });
+                // 顺序执行每个工具
+                for (let ti = 0; ti < toolCalls.length; ti++) {
+                    const toolCall = toolCalls[ti];
+                    const isMulti = toolCalls.length > 1;
+                    // 通知前端
+                    if (onStream) {
+                        onStream({ type: 'tool', content: `🔧 调用工具 (${ti + 1}/${toolCalls.length}): ${toolCall.name}`, tool: toolCall.name });
+                        if (toolCall.args && Object.keys(toolCall.args).length > 0) {
+                            onStream({ type: 'status', content: `📋 参数: ${JSON.stringify(toolCall.args)}`, tool: toolCall.name });
                         }
-                        console.warn(`[PiAgent] PreToolUse denied ${toolCall.name}: ${pre.reason}`);
-                        // 不调 tool.execute, 也不计 consecutiveErrors (这是用户级拒绝, 不是工具错)
-                        continue;
-                    }
-                }
-                catch (err) {
-                    console.warn('[PiAgent] onPreToolUse failed (non-fatal, allowing):', err);
-                }
-                // React Harness: 8-gate + builtin-guards 校验 (在 PreToolUse 之后, 串接双层)
-                // 失败静默, 拒绝时不调 tool.execute
-                try {
-                    const pre = await this.reactHarness.preToolCall(toolCall.name, toolCall.args || {}, this.currentChannelId || undefined);
-                    if (!pre.allowed) {
-                        const deniedResult = {
-                            success: false,
-                            error: `Harness gate 拒绝 (${pre.details.rejectedBy}): ${pre.reason || '未通过安全校验'}`,
-                        };
-                        this.messageHistory.push({
-                            role: 'tool',
-                            content: JSON.stringify(deniedResult),
-                            toolResult: deniedResult,
+                        onStream({
+                            type: 'step_start',
+                            content: `调用 ${toolCall.name}${isMulti ? ` (${ti + 1}/${toolCalls.length})` : ''}`,
+                            tool: toolCall.name,
+                            args: toolCall.args || {},
                         });
-                        this.logToHarness(toolCall.name, toolCall.args, deniedResult);
-                        if (onStream) {
-                            onStream({
-                                type: 'error',
-                                content: `🛡️ Harness ${pre.details.rejectedBy} 拒绝 ${toolCall.name}: ${pre.reason || '安全校验失败'}`,
-                                tool: toolCall.name,
-                            });
-                            // 2026-06-15: step-timeline — Harness gate 拒绝, 标 step_error
-                            onStream({
-                                type: 'step_error',
-                                content: `Harness 拒绝 ${toolCall.name}`,
-                                tool: toolCall.name,
-                                error: pre.reason || '安全校验失败',
-                            });
-                        }
-                        console.warn(`[PiAgent] Harness denied ${toolCall.name} (${pre.details.rejectedBy}): ${pre.reason}`);
+                    }
+                    const tool = this.tools.get(toolCall.name);
+                    if (!tool) {
+                        consecutiveErrors++;
+                        totalErrors++;
+                        const errorResult = { success: false, error: `未知工具: ${toolCall.name}` };
+                        this.messageHistory.push({ role: 'tool', content: JSON.stringify(errorResult), toolResult: errorResult });
+                        this.logToHarness(toolCall.name, toolCall.args, errorResult);
+                        // 2026-07-28: 注入 Reflection 帮助 LLM 理解错误
+                        const obs = buildObservation(toolCall.name, toolCall.args, errorResult);
+                        const ref = buildReflection(toolCall.name, errorResult.error, totalErrors, lastFailedToolCount);
+                        this.messageHistory.push({ role: 'system', content: formatObservationWithReflection(obs, ref) });
+                        if (onStream)
+                            onStream({ type: 'status', content: `💡 Reflection: ${obs.summary}`, tool: 'system' });
+                        console.warn(`[PiAgent] 未知工具: ${toolCall.name} (累计 ${totalErrors}/${this.MAX_TOTAL_ERRORS})，跳过并继续`);
                         continue;
                     }
-                }
-                catch (err) {
-                    console.warn('[PiAgent] reactHarness.preToolCall failed (non-fatal, allowing):', err);
-                }
-                try {
-                    const toolStart = Date.now();
-                    let result = await tool.execute(toolCall.args);
-                    const toolDurationMs = Date.now() - toolStart;
-                    console.log(`[PiAgent] 工具 ${toolCall.name} 执行完成: success=${result.success} (${toolDurationMs}ms)`);
-                    // PostToolUse 审计 hook: 写 audit log, 默认 continue
+                    // Bootstrap PreToolUse hook: 调工具前校验 (危险命令拦截)
+                    // 失败静默 — hook 自身挂掉 = 放行
+                    // P2: 透传 permissionMode (从 BootstrapOptions / env BOLLOON_PERM_MODE 解析)
+                    let toolToExecute = tool;
                     try {
-                        await onPostToolUse({
+                        const pre = await onPreToolUse({
                             tool: toolCall.name,
                             args: toolCall.args || {},
-                            result: {
-                                success: result.success,
-                                output: result.output?.substring(0, 500),
-                                error: result.error,
-                            },
-                            durationMs: toolDurationMs,
+                            permissionMode: this.currentPermissionMode,
                         });
+                        if (!pre.allowed) {
+                            const deniedResult = {
+                                success: false,
+                                error: `PreToolUse 拒绝: ${pre.reason || '未通过安全校验'}`,
+                            };
+                            this.messageHistory.push({ role: 'tool', content: JSON.stringify(deniedResult), toolResult: deniedResult });
+                            this.logToHarness(toolCall.name, toolCall.args, deniedResult);
+                            if (onStream) {
+                                onStream({ type: 'error', content: `🛡️ PreToolUse 拒绝 ${toolCall.name}: ${pre.reason || '安全校验失败'}`, tool: toolCall.name });
+                                onStream({ type: 'step_error', content: `PreToolUse 拒绝 ${toolCall.name}`, tool: toolCall.name, error: pre.reason || '安全校验失败' });
+                            }
+                            console.warn(`[PiAgent] PreToolUse denied ${toolCall.name}: ${pre.reason}`);
+                            continue;
+                        }
                     }
-                    catch (postErr) {
-                        console.warn('[PiAgent] onPostToolUse failed (non-fatal):', postErr);
-                    }
-                    // Context router: 拿最近一次 preToolCall 算的 hint, 拼到 tool result messageHistory
-                    // (LLM 下次看到 tool result 时, 能"记得"这次调用的安全约束)
-                    const routeHint = this.reactHarness.getLastRouteHint();
-                    if (routeHint && routeHint.systemAddition) {
-                        this.messageHistory.push({
-                            role: 'system',
-                            content: `[Harness Router Hint: ${routeHint.reason}]\n${routeHint.systemAddition}`,
-                        });
-                        this.reactHarness.clearRouteHint();
+                    catch (err) {
+                        console.warn('[PiAgent] onPreToolUse failed (non-fatal, allowing):', err);
                     }
-                    // React Harness: post-tool call (output 审计: secret leak 等)
-                    // 拒绝时 result.output 含敏感 → 替换为 generic message, 不污染 messageHistory
+                    // React Harness: 8-gate + builtin-guards 校验 (串接双层)
                     try {
-                        const post = await this.reactHarness.postToolCall(toolCall.name, String(result.output || ''), this.currentChannelId || undefined);
-                        if (!post.allowed) {
+                        const pre = await this.reactHarness.preToolCall(toolCall.name, toolCall.args || {}, this.currentChannelId || undefined);
+                        if (!pre.allowed) {
+                            const deniedResult = { success: false, error: `Harness gate 拒绝 (${pre.details.rejectedBy}): ${pre.reason || '未通过安全校验'}` };
+                            this.messageHistory.push({ role: 'tool', content: JSON.stringify(deniedResult), toolResult: deniedResult });
+                            this.logToHarness(toolCall.name, toolCall.args, deniedResult);
                             if (onStream) {
-                                onStream({
-                                    type: 'error',
-                                    content: `🛡️ Harness output 拒绝 ${toolCall.name}: ${post.reason || '输出含敏感信息'}`,
-                                    tool: toolCall.name,
-                                });
+                                onStream({ type: 'error', content: `🛡️ Harness ${pre.details.rejectedBy} 拒绝 ${toolCall.name}: ${pre.reason || '安全校验失败'}`, tool: toolCall.name });
+                                onStream({ type: 'step_error', content: `Harness 拒绝 ${toolCall.name}`, tool: toolCall.name, error: pre.reason || '安全校验失败' });
                             }
-                            console.warn(`[PiAgent] Harness output denied ${toolCall.name}: ${post.reason}`);
-                            // 替换 result: success 仍保留 (tool 本身没错), 但 output 改成 generic
-                            // 这样 LLM 下轮看 output 不会拿到秘密, 但 success 标志让它知道 "工具执行了"
-                            result = {
-                                ...result,
-                                output: `[harness output gate: output 含敏感内容, 已屏蔽. 原因: ${post.reason || 'unknown'}]`,
-                                _harnessDenied: true,
-                            };
+                            console.warn(`[PiAgent] Harness denied ${toolCall.name} (${pre.details.rejectedBy}): ${pre.reason}`);
+                            continue;
                         }
                     }
                     catch (err) {
-                        console.warn('[PiAgent] reactHarness.postToolCall failed (non-fatal, allowing):', err);
+                        console.warn('[PiAgent] reactHarness.preToolCall failed (non-fatal, allowing):', err);
                     }
-                    this.messageHistory.push({ role: 'tool', content: JSON.stringify(result), toolResult: result, toolCallId: toolCall.id || `call_${Date.now()}_${Math.random().toString(36).slice(2, 8)}` });
-                    this.logToHarness(toolCall.name, toolCall.args, result);
-                    // 通知前端工具执行结果
-                    if (onStream) {
-                        if (result.success) {
-                            onStream({ type: 'status', content: `✅ ${toolCall.name} 执行成功`, tool: toolCall.name });
-                            if (result.output) {
-                                const outputPreview = result.output.substring(0, 200);
-                                onStream({ type: 'tool', content: `📤 结果: ${outputPreview}${result.output.length > 200 ? '...' : ''}`, tool: toolCall.name });
-                            }
-                            // 2026-06-15: step-timeline 状态机 — 关闭当前节点 (成功)
-                            onStream({
-                                type: 'step_done',
-                                content: `${toolCall.name} 执行成功`,
-                                tool: toolCall.name,
-                                success: true,
-                                output: result.output,
-                            });
-                        }
-                        else {
-                            onStream({ type: 'error', content: `❌ ${toolCall.name} 执行失败: ${result.error}`, tool: toolCall.name });
-                            // 2026-06-15: step-timeline 状态机 — 关闭当前节点 (失败)
-                            onStream({
-                                type: 'step_error',
-                                content: `${toolCall.name} 执行失败`,
-                                tool: toolCall.name,
-                                error: result.error,
-                            });
+                    try {
+                        const toolStart = Date.now();
+                        let result = await tool.execute(toolCall.args);
+                        const toolDurationMs = Date.now() - toolStart;
+                        console.log(`[PiAgent] 工具 ${toolCall.name} 执行完成: success=${result.success} (${toolDurationMs}ms)`);
+                        try {
+                            await onPostToolUse({ tool: toolCall.name, args: toolCall.args || {}, result: { success: result.success, output: result.output?.substring(0, 500), error: result.error }, durationMs: toolDurationMs });
                         }
-                    }
-                    if (result.success) {
-                        consecutiveErrors = 0; // 重置连续错误计数
-                        // 2026-06-19: 记录成功结果, 用于 LLM 失败退出时汇总给用户
-                        if (result.output) {
-                            this.successfulToolResults.push({
-                                tool: toolCall.name,
-                                outputPreview: result.output.substring(0, 200) + (result.output.length > 200 ? '...' : '')
-                            });
+                        catch (postErr) {
+                            console.warn('[PiAgent] onPostToolUse failed (non-fatal):', postErr);
                         }
-                        else {
-                            this.successfulToolResults.push({ tool: toolCall.name, outputPreview: '(无输出)' });
+                        const routeHint = this.reactHarness.getLastRouteHint();
+                        if (routeHint && routeHint.systemAddition) {
+                            this.messageHistory.push({ role: 'system', content: `[Harness Router Hint: ${routeHint.reason}]\n${routeHint.systemAddition}` });
+                            this.reactHarness.clearRouteHint();
                         }
-                        // 检查工具执行质量
-                        lastQualityScore = this.estimateToolResultQuality(result);
-                        if (lastQualityScore < this.QUALITY_THRESHOLD && refineAttempts < this.MAX_REFINE_ATTEMPTS) {
-                            refineAttempts++;
-                            console.log(`[PiAgent] 工具结果质量低，自动重试 (${refineAttempts}/${this.MAX_REFINE_ATTEMPTS})`);
+                        try {
+                            const post = await this.reactHarness.postToolCall(toolCall.name, String(result.output || ''), this.currentChannelId || undefined);
+                            if (!post.allowed) {
+                                if (onStream) {
+                                    onStream({ type: 'error', content: `🛡️ Harness output 拒绝 ${toolCall.name}: ${post.reason || '输出含敏感信息'}`, tool: toolCall.name });
+                                }
+                                console.warn(`[PiAgent] Harness output denied ${toolCall.name}: ${post.reason}`);
+                                result = { ...result, output: `[harness output gate: 输出含敏感内容, 已屏蔽. 原因: ${post.reason || 'unknown'}]`, _harnessDenied: true };
+                            }
                         }
-                        else {
-                            console.log(`[PiAgent] 工具执行成功，质量评分: ${(lastQualityScore * 10).toFixed(1)}/10`);
+                        catch (err) {
+                            console.warn('[PiAgent] reactHarness.postToolCall failed (non-fatal, allowing):', err);
                         }
-                        // 工具执行成功后，继续循环获取下一个 LLM 响应
+                        this.messageHistory.push({ role: 'tool', content: JSON.stringify(result), toolResult: result, toolCallId: toolCall.id || `call_${Date.now()}_${Math.random().toString(36).slice(2, 8)}` });
+                        this.logToHarness(toolCall.name, toolCall.args, result);
                         if (onStream) {
-                            onStream({ type: 'status', content: `🔄 工具执行完成，继续循环...`, tool: 'loop' });
-                        }
-                        // 不 break，继续下一次循环
-                    }
-                    else {
-                        consecutiveErrors++;
-                        // 2026-06-16 新增: 累计错误 (跨工具, 兜底防 LLM 轮换工具名死循环)
-                        totalErrors++;
-                        // 跟踪同一工具连续失败次数
-                        if (toolCall.name === lastFailedTool) {
-                            lastFailedToolCount++;
-                        }
-                        else {
-                            lastFailedTool = toolCall.name;
-                            lastFailedToolCount = 1;
+                            if (result.success) {
+                                onStream({ type: 'status', content: `✅ ${toolCall.name} 执行成功`, tool: toolCall.name });
+                                if (result.output) {
+                                    onStream({ type: 'tool', content: `📤 结果: ${result.output.substring(0, 200)}${result.output.length > 200 ? '...' : ''}`, tool: toolCall.name });
+                                }
+                                onStream({ type: 'step_done', content: `${toolCall.name} 执行成功`, tool: toolCall.name, success: true, output: result.output });
+                            }
+                            else {
+                                onStream({ type: 'error', content: `❌ ${toolCall.name} 执行失败: ${result.error}`, tool: toolCall.name });
+                                onStream({ type: 'step_error', content: `${toolCall.name} 执行失败`, tool: toolCall.name, error: result.error });
+                            }
                         }
-                        console.warn(`[PiAgent] 工具 ${toolCall.name} 执行失败 (${lastFailedToolCount}/${MAX_SAME_TOOL_FAILURES}, 累计 ${totalErrors}/${this.MAX_TOTAL_ERRORS}): ${result.error}`);
-                        // 同一工具连续失败达到上限, 不再重试, 强制 LLM 给出最终答案
-                        if (lastFailedToolCount >= MAX_SAME_TOOL_FAILURES) {
-                            console.log(`[PiAgent] 工具 ${toolCall.name} 连续 ${MAX_SAME_TOOL_FAILURES} 次失败, 放弃并要求直接回答`);
-                            this.messageHistory.push({
-                                role: 'system',
-                                content: `[注意] 工具 ${toolCall.name} 在这个上下文中不可用 (连续 ${MAX_SAME_TOOL_FAILURES} 次失败: ${result.error}). 请不要再次调用它, 直接用你已知的信息回答用户, 并在回答开头标记 <final gen>.`
-                            });
-                            lastFailedTool = '';
-                            lastFailedToolCount = 0;
+                        if (result.success) {
                             consecutiveErrors = 0;
-                            continue; // 让 LLM 看到系统提示后再决定
+                            if (result.output) {
+                                this.successfulToolResults.push({ tool: toolCall.name, outputPreview: result.output.substring(0, 200) + (result.output.length > 200 ? '...' : '') });
+                            }
+                            else {
+                                this.successfulToolResults.push({ tool: toolCall.name, outputPreview: '(无输出)' });
+                            }
+                            lastQualityScore = this.estimateToolResultQuality(result);
+                            if (lastQualityScore < this.QUALITY_THRESHOLD && refineAttempts < this.MAX_REFINE_ATTEMPTS) {
+                                refineAttempts++;
+                            }
+                            if (onStream) {
+                                onStream({ type: 'status', content: `🔄 工具执行完成，继续循环...`, tool: 'loop' });
+                            }
                         }
-                        // 连续错误达到上限(混合不同工具), 尝试换一种方式
-                        if (consecutiveErrors >= MAX_CONSECUTIVE_ERRORS) {
-                            console.log(`[PiAgent] 连续 ${MAX_CONSECUTIVE_ERRORS} 次错误，尝试换一种方式处理`);
-                            this.messageHistory.push({
-                                role: 'system',
-                                content: `[注意] 前面的工具调用连续失败。请尝试其他工具或换一种方式完成用户请求, 或用 <final gen> 给出最终回答.`
-                            });
-                            consecutiveErrors = 0;
+                        else {
+                            consecutiveErrors++;
+                            totalErrors++;
+                            if (toolCall.name === lastFailedTool) {
+                                lastFailedToolCount++;
+                            }
+                            else {
+                                lastFailedTool = toolCall.name;
+                                lastFailedToolCount = 1;
+                            }
+                            console.warn(`[PiAgent] 工具 ${toolCall.name} 执行失败 (${lastFailedToolCount}/${MAX_SAME_TOOL_FAILURES}, 累计 ${totalErrors}/${this.MAX_TOTAL_ERRORS}): ${result.error}`);
+                            // 2026-07-28: 注入 Observation + Reflection 替代旧 hardcode 提示
+                            const obs = buildObservation(toolCall.name, toolCall.args, { success: false, error: result.error });
+                            const ref = buildReflection(toolCall.name, result.error, totalErrors, lastFailedToolCount);
+                            this.messageHistory.push({ role: 'system', content: formatObservationWithReflection(obs, ref) });
+                            if (onStream)
+                                onStream({ type: 'status', content: `💡 Reflection: ${obs.summary} → ${ref[0]?.action || '放弃'}`, tool: 'system' });
+                            if (lastFailedToolCount >= MAX_SAME_TOOL_FAILURES) {
+                                this.messageHistory.push({ role: 'system', content: `[注意] 工具 ${toolCall.name} 在这个上下文中不可用 (连续 ${MAX_SAME_TOOL_FAILURES} 次失败: ${result.error}). 请不要再次调用它, 直接用你已知的信息回答用户, 并在回答开头标记 <final gen>.` });
+                                lastFailedTool = '';
+                                lastFailedToolCount = 0;
+                                consecutiveErrors = 0;
+                                continue;
+                            }
+                            if (consecutiveErrors >= MAX_CONSECUTIVE_ERRORS) {
+                                this.messageHistory.push({ role: 'system', content: `[注意] 前面的工具调用连续失败。请尝试其他工具或换一种方式完成用户请求, 或用 <final gen> 给出最终回答.` });
+                                consecutiveErrors = 0;
+                            }
                         }
                     }
-                }
-                catch (execError) {
-                    consecutiveErrors++;
-                    // 2026-06-16 新增: 异常分支也要累计
-                    totalErrors++;
-                    const errorResult = { success: false, error: String(execError) };
-                    this.messageHistory.push({ role: 'tool', content: JSON.stringify(errorResult), toolResult: errorResult });
-                    this.logToHarness(toolCall.name, toolCall.args, errorResult);
-                    console.error(`[PiAgent] 工具执行异常 (累计 ${totalErrors}/${this.MAX_TOTAL_ERRORS}): ${execError}`);
-                }
+                    catch (execError) {
+                        consecutiveErrors++;
+                        totalErrors++;
+                        const errorResult = { success: false, error: String(execError) };
+                        this.messageHistory.push({ role: 'tool', content: JSON.stringify(errorResult), toolResult: errorResult });
+                        this.logToHarness(toolCall.name, toolCall.args, errorResult);
+                        const obs = buildObservation(toolCall.name, toolCall.args, errorResult);
+                        const ref = buildReflection(toolCall.name, errorResult.error, totalErrors, lastFailedToolCount);
+                        this.messageHistory.push({ role: 'system', content: formatObservationWithReflection(obs, ref) });
+                        if (onStream)
+                            onStream({ type: 'status', content: `💡 Reflection: ${obs.summary}`, tool: 'system' });
+                        console.error(`[PiAgent] 工具执行异常 (累计 ${totalErrors}/${this.MAX_TOTAL_ERRORS}): ${execError}`);
+                    }
+                } // end for (ti)
+                // 所有工具执行完毕后, continue while 循环, 让 LLM 看到结果
+                continue;
             }
             else {
                 // LLM 返回的不是 tool call 格式