mcp-probe-kit 3.6.10 → 3.6.11

package/build/tools/fix_bug.js

@@ -1,199 +1,13 @@
 /**
- * fix_bug 工具
- *
- * 功能：基于 TBP 8 步法输出真因分析与修复指南
- * 模式：指令生成器模式 - 返回详细的 RCA 与修复指南，由 AI 执行实际操作
- *
- * 流程：定义现象 → 复盘时间线 → 排除错误方向 → 找共同模式 → 定位边界 → 陈述真因 → 闭合证据链 → 设计修复方案
+ * fix_bug 工具 — SRC-8 delegated plan + 真因工作表
  */
 import { parseArgs, getString } from "../utils/parseArgs.js";
 import { okStructured } from "../lib/response.js";
-import { renderGuidanceHeader } from "../lib/guidance.js";
+import { renderOrchestrationHeader } from "../lib/orchestration-guidance.js";
 import { handleToolError } from "../utils/error-handler.js";
-const PROMPT_TEMPLATE = `# TBP 8 步 Bug 真因分析与修复指南
-
-## 🐛 Bug 信息
-
-**错误信息**:
-\`\`\`
-{error_message}
-\`\`\`
-
-{stack_trace_section}
-
-{reproduce_section}
-
-{behavior_section}
-
-{code_context_section}
-
----
-
-## TBP-1 现象
-
-先把用户可见的问题定义准确，避免“坏了/卡了/有问题”这类泛化描述。
-
-要求：
-1. 用 1-2 句定义现象
-2. 明确是“慢”“停滞”“失败”“回归”还是“未生效”
-3. 如果提供了期望/实际行为，必须纳入现象定义
-
-## TBP-2 时间线
-
-基于现有信息复盘事件顺序，至少回答：
-1. 什么时候开始
-2. 中间发生了什么
-3. 最后停在什么状态
-
-优先记录：
-- 用户输入 / 复现步骤
-- first progress / first error
-- tool call / stack / stop / timeout
-- 关键文件或模块
-
-## TBP-3 不是这个
-
-列出当前应优先排除的错误方向，并说明需要什么证据排除。
-
-常见排除项：
-- 不是单纯网络慢
-- 不是前端展示截断
-- 不是消息没转发
-- 不是纯粹的超时表象
-- 不是用户输入丢失
-
-## TBP-4 共同模式
-
-对比成功/失败样本，找出从哪一步开始分叉。
-
-重点检查：
-- 传输层
-- 会话状态机
-- 完成态判定
-- 工具执行
-- 文件写入
-- 重试策略
-
-## TBP-5 边界
-
-明确指出问题失控落在哪一层：
-- 上游模型 / SDK
-- 网关状态机
-- session 复用
-- tool 执行层
-- 文件系统
-- 环境配置
-- UI 展示层
-
-## TBP-6 真因
-
-真因必须写成因果句：
-\`A + B 在条件 D 下导致了 C\`
-
-不允许只写：
-- 超时了
-- SDK 有 bug
-- 返回慢
-
-## TBP-7 证据链
-
-说明：
-1. 哪些证据支持真因
-2. 哪些现象若出现会推翻它，但现场没有出现
-3. 为什么其他解释更弱
-
-## TBP-8 修复
-
-只有在证据链闭合后，才开始设计修复。
-
-每个修复方案都要说明：
-- 改哪一层
-- 为什么这是修真因，不是补症状
-- 风险是什么
-- 怎么验证
-
----
-
-## 📤 输出格式要求
-
-请严格按以下 JSON 格式输出修复指南：
-
-\`\`\`json
-{
-  "bug_summary": "Bug 简述（一句话）",
-  "analysis_mode": "tbp8",
-  "analysis": {
-    "error_type": "错误类型",
-    "direct_cause": "直接原因",
-    "root_cause": "根本原因",
-    "affected_scope": "影响范围"
-  },
-  "tbp": {
-    "phenomenon": "TBP-1 现象",
-    "timeline": [],
-    "ruled_out": [],
-    "common_pattern": "TBP-4 共同模式",
-    "boundary": "TBP-5 边界",
-    "root_cause_statement": "A + B 在条件 D 下导致 C",
-    "evidence": [],
-    "repair": []
-  },
-  "location": {
-    "file": "问题文件路径",
-    "line": 42,
-    "function": "问题函数名",
-    "code_snippet": "问题代码片段"
-  },
-  "fix_plan": {
-    "chosen_solution": "选择的修复方案",
-    "reason": "选择理由",
-    "steps": [
-      { "step": 1, "action": "修复步骤", "file": "文件", "change": "变更内容" }
-    ],
-    "code_before": "修改前代码",
-    "code_after": "修改后代码"
-  },
-  "verification": {
-    "test_cases": ["测试用例1", "测试用例2"],
-    "manual_checks": ["手动验证项1", "手动验证项2"]
-  }
-}
-\`\`\`
-
-## ⚠️ 护栏
-
-- 不要一看到新现象就补一个新分支，先判断是不是同一类根因在换表现
-- 不要把“超时”直接当真因，要先分清是没进展、慢进展，还是结果已出现但判定过严
-- 不要只看单一样本就下结论，有对比样本时必须做对比
-- 如果证据不足，要明确说明还缺什么证据
-
----
-
-*指南版本: 2.0.0*
-*工具: MCP Probe Kit - fix_bug*
-`;
-function inferBugType(text) {
-    if (/timeout|latency|slow|卡|慢|超时/i.test(text))
-        return "performance";
-    if (/auth|permission|unauthor|token|登录|权限/i.test(text))
-        return "security";
-    if (/ui|render|white screen|白屏|按钮|页面/i.test(text))
-        return "ui";
-    if (/sql|db|data|schema|缓存|数据/i.test(text))
-        return "data";
-    if (/api|request|network|gateway|integration|调用失败/i.test(text))
-        return "integration";
-    return "functional";
-}
-function inferSeverity(text) {
-    if (/crash|fatal|panic|生产故障|数据丢失|security|权限绕过/i.test(text))
-        return "critical";
-    if (/white screen|白屏|500|无法登录|阻塞|不可用|timeout|超时/i.test(text))
-        return "high";
-    if (/error|exception|失败|异常/i.test(text))
-        return "medium";
-    return "low";
-}
+import { resolveGuidanceCode, trimCodeForPrompt } from "../lib/code-review-input.js";
+import { resolveWorkspaceRoot } from "../lib/workspace-root.js";
+import { ATTRIBUTION_LAYERS, SRC8_METHODOLOGY, ROOT_CAUSE_WORKSHEET_GATES, buildRootCauseWorksheet, buildSrc8Checklist, buildSrc8DelegatedPlan, buildSrc8EvidenceFromInput, renderFixBugAgentPromptBody, resolveAnalysisMode, } from "../lib/src8-guidance.js";
 function extractFilePaths(text) {
     if (!text)
         return [];
@@ -202,13 +16,10 @@ function extractFilePaths(text) {
 }
 function summarizePhenomenon(errorMessage, expectedBehavior, actualBehavior) {
     if (expectedBehavior || actualBehavior) {
-        return `实际表现为“${actualBehavior || errorMessage}”，但期望是“${expectedBehavior || "保持正常行为"}”。`;
+        return `实际表现为「${actualBehavior || errorMessage}」，但期望是「${expectedBehavior || "保持正常行为"}」。`;
     }
-    return `当前可见现象是“${errorMessage}”，需要先区分它属于失败、停滞、未生效还是性能退化。`;
+    return `当前可见现象是「${errorMessage}」，需要先区分它属于失败、停滞、未生效还是性能退化。`;
 }
-/**
- * fix_bug 工具实现
- */
 export async function fixBug(args) {
     try {
         const parsedArgs = parseArgs(args, {
@@ -216,134 +27,171 @@ export async function fixBug(args) {
                 error_message: "",
                 stack_trace: "",
                 code_context: "",
-                analysis_mode: "tbp8",
+                file_path: "",
+                project_root: "",
+                analysis_mode: SRC8_METHODOLOGY,
                 steps_to_reproduce: "",
                 expected_behavior: "",
                 actual_behavior: "",
+                success_sample: "",
+                verification_target: "",
             },
             primaryField: "error_message",
             fieldAliases: {
-                error_message: ["error", "err", "message", "错误", "错误信息"],
+                error_message: ["error", "err", "message", "错误", "错误信息", "description", "desc", "summary", "问题", "bug", "issue"],
                 stack_trace: ["stack", "trace", "堆栈", "调用栈"],
                 code_context: ["code_context", "code", "context", "相关代码", "代码上下文"],
-                analysis_mode: ["analysis_mode", "methodology", "tbp", "rca", "分析方法"],
+                file_path: ["filePath", "filepath", "path", "文件路径"],
+                project_root: ["projectRoot", "project_path", "dir", "directory", "项目路径"],
+                analysis_mode: ["analysis_mode", "methodology", "tbp", "src8", "rca", "分析方法"],
                 steps_to_reproduce: ["steps", "reproduce", "步骤", "复现步骤"],
                 expected_behavior: ["expected", "期望", "期望行为"],
                 actual_behavior: ["actual", "实际", "实际行为"],
+                success_sample: ["success", "working_case", "正常样本", "成功样本", "对比样本"],
+                verification_target: ["target", "acceptance", "验收标准", "验证目标", "done_criteria"],
             },
         });
-        const errorMessage = getString(parsedArgs.error_message);
+        const errorMessage = getString(parsedArgs.error_message) || getString(parsedArgs.description);
         const stackTrace = getString(parsedArgs.stack_trace);
-        const codeContext = getString(parsedArgs.code_context);
-        const analysisMode = (getString(parsedArgs.analysis_mode) || "tbp8").toLowerCase() === "tbp8"
-            ? "tbp8"
-            : "tbp8";
+        const inlineCodeContext = getString(parsedArgs.code_context);
+        const filePath = getString(parsedArgs.file_path);
+        const projectRoot = getString(parsedArgs.project_root);
+        const analysisMode = resolveAnalysisMode(getString(parsedArgs.analysis_mode));
         const stepsToReproduce = getString(parsedArgs.steps_to_reproduce);
         const expectedBehavior = getString(parsedArgs.expected_behavior);
         const actualBehavior = getString(parsedArgs.actual_behavior);
+        const successSample = getString(parsedArgs.success_sample);
+        const verificationTarget = getString(parsedArgs.verification_target);
         if (!errorMessage) {
             throw new Error("缺少必填参数: error_message（错误信息）");
         }
+        const resolvedProjectRoot = projectRoot ? resolveWorkspaceRoot(projectRoot) : undefined;
+        const resolvedCode = resolveGuidanceCode({
+            code: inlineCodeContext,
+            filePath: filePath || undefined,
+            projectRoot: resolvedProjectRoot,
+        });
+        const hasSuccessSample = Boolean(successSample.trim());
+        const rootCauseWorksheet = buildRootCauseWorksheet({ hasComparisonSample: hasSuccessSample });
+        const plan = buildSrc8DelegatedPlan({
+            error_message: errorMessage,
+            stack_trace: stackTrace || undefined,
+            analysis_mode: analysisMode,
+            code_context: resolvedCode.code || inlineCodeContext || undefined,
+            project_root: resolvedProjectRoot,
+            file_path: filePath || resolvedCode.file || undefined,
+        });
+        if (resolvedCode.error) {
+            return okStructured(`❌ fix_bug 无法读取代码上下文: ${resolvedCode.error}`, {
+                mode: "delegated",
+                methodology: analysisMode,
+                src8Checklist: buildSrc8Checklist(),
+                rootCauseWorksheet,
+                metadata: { plan },
+                bugfixInput: {
+                    received: false,
+                    error: resolvedCode.error,
+                    error_message: errorMessage,
+                    file: filePath || null,
+                    analysis_mode: analysisMode,
+                },
+            }, {
+                note: "请修正 file_path/project_root 后重试；执行顺序见 metadata.plan",
+            });
+        }
+        const codeContext = resolvedCode.code;
+        const hasCodeContext = Boolean(codeContext.trim());
+        const promptCodeContext = hasCodeContext ? trimCodeForPrompt(codeContext, 24000, "bugfixInput") : "";
         const stackTraceSection = stackTrace
             ? `**堆栈跟踪**:\n\`\`\`\n${stackTrace}\n\`\`\``
             : "**堆栈跟踪**: 未提供（建议提供以便更准确定位）";
-        const reproduceSection = stepsToReproduce
-            ? `**复现步骤**:\n${stepsToReproduce}`
-            : "";
+        const reproduceSection = stepsToReproduce ? `**复现步骤**:\n${stepsToReproduce}` : "";
         const behaviorSection = [expectedBehavior ? `**期望行为**: ${expectedBehavior}` : "", actualBehavior ? `**实际行为**: ${actualBehavior}` : ""]
             .filter(Boolean)
             .join("\n\n");
-        const codeContextSection = codeContext
-            ? `**代码/图谱上下文**:\n\`\`\`\n${codeContext}\n\`\`\``
+        const comparisonSection = hasSuccessSample
+            ? `**成功样本（SRC-4c 对比分叉）**:\n${successSample}`
+            : "**成功样本**: 未提供（SRC-4c 须在 rootCauseAnalysis.evidenceGaps 说明缺口）";
+        const verificationTargetSection = verificationTarget
+            ? `**验收契约（SRC-3）**:\n${verificationTarget}`
+            : "**验收契约**: 未提供（SRC-3 须设定 SMART 标准，如 failing test 变绿）";
+        const codeContextSection = hasCodeContext
+            ? `**代码/图谱上下文**${resolvedCode.file ? `（来源: ${resolvedCode.file}）` : ""}:\n\`\`\`\n${promptCodeContext}\n\`\`\``
             : "";
-        const header = renderGuidanceHeader({
+        const header = renderOrchestrationHeader({
             tool: "fix_bug",
-            goal: "基于 TBP 8 步法分析 Bug 真因，并提供修复指南。AI 应根据指南执行实际修复。",
-            tasks: ["先定义现象并复盘时间线", "闭合证据链后形成真因", "输出修复步骤与验证方案"],
-            outputs: ["TBP 真因分析指南", "结构化 RCA 数据"],
+            goal: `SRC-8 修复 Bug：${errorMessage.substring(0, 120)}${errorMessage.length > 120 ? "..." : ""}`,
+            tasks: [
+                "严格按 structuredContent.metadata.plan.steps 顺序执行（禁止从 src8-4 跳起）",
+                "src8-4 闭合 rootCauseWorksheet 并输出 rootCauseAnalysis 后才可进入 src8-5/6",
+                "src8-6 前满足复现门禁；src8-8 用 memorize_asset 沉淀",
+            ],
+            notes: [
+                "MCP 为 guidance-only，不自动修 Bug",
+                hasCodeContext ? `已注入代码上下文（${codeContext.split("\n").length} 行）` : "建议补充 code_context 或 file_path + project_root",
+            ],
+        });
+        const guide = `${header}${renderFixBugAgentPromptBody({
+            error_message: errorMessage,
+            stack_trace_section: stackTraceSection,
+            reproduce_section: reproduceSection,
+            behavior_section: behaviorSection,
+            comparison_section: comparisonSection,
+            verification_target_section: verificationTargetSection,
+            code_context_section: codeContextSection,
+            plan,
+        })}`;
+        const evidence = buildSrc8EvidenceFromInput({
+            error_message: errorMessage,
+            stack_trace: stackTrace || undefined,
+            code_context: codeContext || undefined,
+            steps_to_reproduce: stepsToReproduce || undefined,
+            expected_behavior: expectedBehavior || undefined,
+            actual_behavior: actualBehavior || undefined,
+            success_sample: successSample || undefined,
+            verification_target: verificationTarget || undefined,
         });
-        const guide = `${header}${PROMPT_TEMPLATE
-            .replace(/{error_message}/g, errorMessage)
-            .replace(/{stack_trace_section}/g, stackTraceSection)
-            .replace(/{reproduce_section}/g, reproduceSection)
-            .replace(/{behavior_section}/g, behaviorSection)
-            .replace(/{code_context_section}/g, codeContextSection)}`;
-        const evidence = [
-            { type: "symptom", detail: errorMessage, source: "error_message" },
-            ...(stepsToReproduce ? [{ type: "timeline", detail: stepsToReproduce, source: "steps_to_reproduce" }] : []),
-            ...(stackTrace ? [{ type: "stack", detail: stackTrace, source: "stack_trace" }] : []),
-            ...(codeContext ? [{ type: "code", detail: codeContext, source: "code_context" }] : []),
-            ...(expectedBehavior ? [{ type: "comparison", detail: `期望行为: ${expectedBehavior}`, source: "expected_behavior" }] : []),
-            ...(actualBehavior ? [{ type: "comparison", detail: `实际行为: ${actualBehavior}`, source: "actual_behavior" }] : []),
-        ];
         const affectedFiles = [...extractFilePaths(stackTrace), ...extractFilePaths(codeContext)]
             .filter((value, index, array) => array.indexOf(value) === index);
-        const structuredData = {
-            summary: `TBP 8 步真因分析：${errorMessage.substring(0, 60)}${errorMessage.length > 60 ? "..." : ""}`,
-            bugType: inferBugType([errorMessage, actualBehavior, codeContext].filter(Boolean).join("\n")),
-            severity: inferSeverity([errorMessage, actualBehavior].filter(Boolean).join("\n")),
-            analysisMode,
-            rootCause: "待通过 TBP-6 真因分析确认；在证据链闭合前不直接给出确定性修复结论。",
-            affectedComponents: affectedFiles.length > 0 ? affectedFiles.map((item) => item.split("/").slice(0, -1).join("/") || item) : [],
-            affectedFiles,
-            fixPlan: {
-                steps: [
-                    "先完成 TBP-1：精确定义现象，避免泛化描述",
-                    "完成 TBP-2 时间线，明确开始、经过、终止状态",
-                    "完成 TBP-3 与 TBP-4，排除错误方向并对比成功/失败样本",
-                    "在 TBP-5/6 确认问题边界与因果句真因后，再修改代码",
-                ],
-                estimatedTime: "30-90 分钟（取决于证据完整度）",
-                risks: [
-                    "如果跳过 TBP 1-7，容易只修表象不修真因",
-                    "如果没有成功样本对比，TBP-4 共同模式结论可能偏弱",
-                ],
+        return okStructured(guide, {
+            mode: "delegated",
+            methodology: analysisMode,
+            src8Checklist: buildSrc8Checklist(),
+            attributionLayers: ATTRIBUTION_LAYERS,
+            rootCauseWorksheet,
+            agentMustProduce: "BugAnalysis",
+            src8Gate: {
+                blockCodeChangeUntilStep: 4,
+                blockCountermeasuresUntilRootCauseWorksheetClosed: true,
+                implementAtStep: 6,
+                rootCauseWorksheetGates: ROOT_CAUSE_WORKSHEET_GATES,
+                requireReproductionBeforeImplement: true,
+                escalateAfterFailedFixAttempts: 3,
             },
-            testPlan: {
-                unitTests: ["覆盖触发 Bug 的最小输入场景", "覆盖修复后的边界条件"],
-                integrationTests: ["覆盖完整调用链或关键交互路径"],
-                manualTests: ["按原复现步骤再次验证", "验证成功样本与失败样本分叉点已消失"],
-            },
-            preventionMeasures: [
-                "补充针对该边界条件的自动化测试",
-                "把关键状态与错误上下文写入日志，便于 TBP-2 时间线复盘",
-                "对完成态判定、文件写入、重试等边界增加显式校验",
-            ],
-            tbp: {
-                phenomenon: summarizePhenomenon(errorMessage, expectedBehavior, actualBehavior),
-                timeline: [
-                    { order: 1, event: "收到问题现象", evidence: errorMessage },
-                    ...(stepsToReproduce ? [{ order: 2, event: "获得复现步骤", evidence: stepsToReproduce }] : []),
-                    ...(stackTrace ? [{ order: 3, event: "获得堆栈或错误栈", evidence: stackTrace }] : []),
-                    ...(codeContext ? [{ order: 4, event: "获得代码/图谱上下文", evidence: codeContext }] : []),
-                ],
-                ruledOut: [
-                    "尚不能直接归因于“纯超时”或“SDK 有 bug”，需要先完成时间线与边界分析",
-                    "尚不能直接归因于“网络慢”，除非有传输层证据支持",
-                ],
-                commonPattern: "优先比较成功与失败样本，从调用链、状态机、工具执行、文件写入等维度找分叉点。",
-                boundary: "待定位；默认优先排查工具执行层、状态机、文件写入层和环境配置层。",
-                rootCauseStatement: "待形成 “A + B 在条件 D 下导致 C” 的因果句；在证据闭环前不下最终结论。",
+            metadata: { plan },
+            bugfixInput: {
+                received: true,
+                error_message: errorMessage,
+                stack_trace: stackTrace || null,
+                code_context: hasCodeContext ? codeContext : null,
+                file: resolvedCode.file ?? (filePath || null),
+                lineCount: hasCodeContext ? codeContext.split("\n").length : 0,
+                truncatedInPrompt: hasCodeContext && codeContext.length !== promptCodeContext.length,
+                steps_to_reproduce: stepsToReproduce || null,
+                expected_behavior: expectedBehavior || null,
+                actual_behavior: actualBehavior || null,
+                success_sample: successSample || null,
+                verification_target: verificationTarget || null,
+                hasComparisonSample: hasSuccessSample,
+                hasVerificationTarget: Boolean(verificationTarget.trim()),
+                analysis_mode: analysisMode,
+                affectedFiles,
                 evidence,
-                repair: [
-                    {
-                        layer: "analysis",
-                        action: "先补齐现象、时间线、对比样本和边界信息，再进入代码修复",
-                        risk: "证据不足时过早修复，可能导致重复返工",
-                        verification: "确认 TBP-7 证据链能解释全部关键现象",
-                    },
-                    {
-                        layer: "implementation",
-                        action: "只修改真因所在层，不以“兜底分支”替代根因修复",
-                        risk: "补症状会掩盖真正失控边界",
-                        verification: "验证原问题消失且未引入新的回归",
-                    },
-                ],
+                phenomenon: summarizePhenomenon(errorMessage, expectedBehavior, actualBehavior),
             },
-        };
-        return okStructured(guide, structuredData, {
+        }, {
             schema: (await import("../schemas/output/core-tools.js")).BugAnalysisSchema,
-            note: "本工具默认采用 TBP 8 步法输出真因分析与修复指南，AI 应先完成证据闭环，再执行代码修复和验证测试",
+            note: "按 metadata.plan 执行 SRC-8；src8-4 闭合前禁止改代码",
         });
     }
     catch (error) {