@tea-agent/loop-agent 0.25.4 → 0.25.5

package/dist/workflows/dag/backend-test-case-manifest.js

@@ -44,7 +44,7 @@ const gapSchema = z
 const caseSchema = z
     .object({
     caseId: z.string().regex(BACKEND_TEST_CASE_ID_PATTERN),
-    acIds: z.array(z.string().regex(BACKEND_TEST_AC_ID_PATTERN)).min(1),
+    acIds: z.array(z.string().regex(BACKEND_TEST_AC_ID_PATTERN)),
     title: z.string().min(1),
     category: backendTestCaseCategorySchema,
     automationStatus: backendTestCaseAutomationStatusSchema,
@@ -96,6 +96,44 @@ export const backendTestCaseManifestSchema = z
         requirementIds: z.array(z.string().regex(/^(?:REQ|BR|AC)-[A-Z0-9]+(?:-[A-Z0-9]+)*$/)),
     })
         .strict(),
+    materializationStatus: z.enum(["available", "partial", "unavailable"]).optional(),
+    sourceFactsIssues: z.array(z.string().min(1)).optional(),
+    ruleCoverageSummary: z.object({
+        explicitAcCount: z.number().int().min(0),
+        coveredAcCount: z.number().int().min(0),
+        ruleCount: z.number().int().min(0),
+        coveredRuleCount: z.number().int().min(0),
+        testPointCount: z.number().int().min(0),
+        coveredTestPointCount: z.number().int().min(0),
+        enumValueCount: z.number().int().min(0),
+        coveredEnumValueCount: z.number().int().min(0),
+        invalidEquivalenceClassCount: z.number().int().min(0),
+        coveredInvalidEquivalenceClassCount: z.number().int().min(0),
+        boundaryPointCount: z.number().int().min(0),
+        coveredBoundaryPointCount: z.number().int().min(0),
+        formatClassCount: z.number().int().min(0),
+        coveredFormatClassCount: z.number().int().min(0),
+        businessStateCount: z.number().int().min(0),
+        coveredBusinessStateCount: z.number().int().min(0),
+        gapCount: z.number().int().min(0),
+        conflictCount: z.number().int().min(0),
+    }).strict().optional(),
+    correspondenceSummary: z.object({
+        markdownModuleCount: z.number().int().min(0),
+        exactModuleCount: z.number().int().min(0),
+        markdownCaseCount: z.number().int().min(0),
+        exactCorrespondenceCount: z.number().int().min(0),
+        missingPytestCount: z.number().int().min(0),
+        multiplePytestCount: z.number().int().min(0),
+        extraPytestCount: z.number().int().min(0),
+        scriptMismatchCount: z.number().int().min(0),
+        testPointCount: z.number().int().min(0),
+        mappedTestPointCount: z.number().int().min(0),
+    }).strict().optional(),
+    artifactRefs: z.object({
+        caseCoverageFacts: z.object({ path: z.string().min(1), sha256: z.string().regex(/^[a-f0-9]{64}$/).optional() }).strict(),
+        correspondenceFacts: z.object({ path: z.string().min(1), sha256: z.string().regex(/^[a-f0-9]{64}$/).optional() }).strict(),
+    }).strict().optional(),
     cases: z.array(caseSchema),
     evidenceGaps: z.array(gapSchema).default([]),
     coverageSummary: z

package/dist/workflows/dag/backend-test-markdown-workflow.js

@@ -6,6 +6,7 @@ import { createWriteStream } from "node:fs";
 import path from "node:path";
 import { promisify } from "node:util";
 import { parseJacocoXml, } from "./backend-test-coverage-contract.js";
+import { backendTestCaseManifestSchema, computeCaseManifestCoverageSummary, } from "./backend-test-case-manifest.js";
 const execFileAsync = promisify(execFile);
 // Canonical contract is three digits (BE-MODULE-001). Readers accept legacy/
 // model-drift two-digit suffixes and canonicalize them in memory so one bad
@@ -247,12 +248,20 @@ function extractSourceReferences(input) {
     // rejects them instead of silently accepting a nearby bound prefix.
     return unique([...boundMatches, ...citedPaths]);
 }
+function hasStructuredMarkdownTable(value, headerPattern) {
+    const rows = value.split(/\r?\n/).filter((line) => /^\s*\|.*\|\s*$/.test(line));
+    if (rows.length < 3 || !headerPattern.test(rows[0]))
+        return false;
+    return /^\s*\|(?:\s*:?-{3,}:?\s*\|)+\s*$/i.test(rows[1]) && rows.slice(2).some((row) => row.split("|").some((cell) => cell.trim().length > 0));
+}
 function hasNumberedListItem(value) {
     return (/^\s*\d+[.)]\s+\S+/m.test(value) ||
-        /^\s*\|\s*\d+\s*\|\s*\S+/m.test(value));
+        /^\s*\|\s*\d+\s*\|\s*\S+/m.test(value) ||
+        hasStructuredMarkdownTable(value, /(?:步骤|操作|测试点|参数|step|action|operation|test point|parameter)/i));
 }
 function hasAssertableExpectedResult(value) {
-    return /^\s*(?:\d+[.)]|[-*+])\s+\S+/m.test(value);
+    return (/^\s*(?:\d+[.)]|[-*+])\s+\S+/m.test(value) ||
+        hasStructuredMarkdownTable(value, /(?:预期|期望|断言|状态|结果|expected|assert|status|result)/i));
 }
 function resolveBoundSourceReference(input) {
     if (!input.sourceBinding) {
@@ -879,7 +888,10 @@ function reportOutputLines(value, prefix) {
         const trimmed = line.startsWith(underscorePrefix)
             ? line.slice(underscorePrefix.length)
             : line.slice(spacePrefix.length);
-        return redactBackendTestOutput(trimmed.trim()).slice(0, 4000);
+        // Preserve a complete single-line JSON payload through parsing. Bounding
+        // happens after structural extraction; slicing here can corrupt valid JSON
+        // and make a large passed response appear as an empty body.
+        return redactBackendTestOutput(trimmed.trim());
     });
 }
 /**
@@ -896,9 +908,16 @@ export function normalizeBackendTestHttpLogText(value) {
     while (i < lines.length) {
         const raw = lines[i];
         const line = raw.trim();
-        // Already-contract lines: pass through.
-        if (/^(HTTP_REQUEST|HTTP REQUEST|HTTP_RESPONSE|HTTP RESPONSE)\b/i.test(line)) {
-            out.push(raw);
+        // Canonical and compatible single-line forms. Normalize case, separator,
+        // and an optional colon so downstream parsing only consumes the canonical
+        // HTTP_REQUEST / HTTP_RESPONSE protocol. Examples accepted here include
+        // `HTTP_REQUEST {...}`, `HTTP REQUEST method=...`, and the generated-helper
+        // dialect `HTTP request: {...}`.
+        const protocolLine = line.match(/^HTTP(?:_|\s+)(REQUEST|RESPONSE)\b\s*:?\s*(.*)$/i);
+        if (protocolLine) {
+            const canonicalPrefix = `HTTP_${protocolLine[1].toUpperCase()}`;
+            const payload = protocolLine[2]?.trim() ?? "";
+            out.push(payload ? `${canonicalPrefix} ${payload}` : canonicalPrefix);
             i += 1;
             continue;
         }
@@ -1000,12 +1019,13 @@ function parseHttpLogPayload(line, kind) {
     const trimmed = line.trim();
     if (!trimmed)
         return {};
+    let parsedJson;
     let kv = {};
     // Try JSON first (legacy HTTP_REQUEST {"method":...} shape).
     if (trimmed.startsWith("{")) {
         try {
-            const parsed = JSON.parse(trimmed);
-            for (const [key, value] of Object.entries(parsed)) {
+            parsedJson = JSON.parse(trimmed);
+            for (const [key, value] of Object.entries(parsedJson)) {
                 kv[key.toLowerCase()] = typeof value === "string" ? value : safeStringify(value);
             }
         }
@@ -1017,21 +1037,68 @@ function parseHttpLogPayload(line, kind) {
         kv = extractKvPairs(trimmed);
     }
     if (kind === "request") {
+        const queryEnvelope = asHttpLogRecord(parsedJson?.query);
+        const method = kv.method ?? httpLogScalar(queryEnvelope?.method);
+        const url = kv.url ?? kv.path ?? httpLogScalar(queryEnvelope?.url) ?? httpLogScalar(queryEnvelope?.path);
+        let body = kv.parameters ?? kv.payload ?? kv.data;
+        if (!body && parsedJson) {
+            const parameters = {};
+            if (queryEnvelope) {
+                const businessQuery = Object.fromEntries(Object.entries(queryEnvelope).filter(([key]) => !["method", "url", "path"].includes(key)));
+                parameters.query = businessQuery;
+            }
+            else if (Object.hasOwn(parsedJson, "query")) {
+                parameters.query = parsedJson.query;
+            }
+            for (const key of ["body", "json"]) {
+                if (Object.hasOwn(parsedJson, key))
+                    parameters[key] = parsedJson[key];
+            }
+            if (Object.keys(parameters).length > 0)
+                body = boundedHttpLogBody(parameters);
+        }
         return {
-            method: kv.method,
-            url: kv.url,
-            body: kv.parameters ?? kv.body ?? kv.json ?? kv.data ?? kv.query ?? kv.payload,
+            method,
+            url,
+            body: boundedHttpLogBody(body ?? kv.body ?? kv.json ?? kv.query),
         };
     }
-    const status = kv.status_code ?? kv.status;
+    const nestedJson = asHttpLogRecord(parsedJson?.json);
+    const nestedTransport = !kv.status_code && !kv.status
+        && Boolean(nestedJson && (Object.hasOwn(nestedJson, "status_code") || Object.hasOwn(nestedJson, "body")));
+    const status = kv.status_code ?? kv.status
+        ?? (nestedTransport ? httpLogScalar(nestedJson?.status_code ?? nestedJson?.status) : undefined);
+    const body = kv.result
+        ?? (nestedTransport ? boundedHttpLogBody(nestedJson?.body ?? nestedJson) : kv.json)
+        ?? kv.body
+        ?? kv.text;
     return {
         method: kv.method,
-        url: kv.url,
+        url: kv.url ?? kv.path,
         status,
         statusText: kv.status_text ?? (status ? HTTP_STATUS_TEXT[status] : undefined),
-        body: kv.result ?? kv.body ?? kv.text,
+        body: boundedHttpLogBody(body),
     };
 }
+function asHttpLogRecord(value) {
+    return value !== null && typeof value === "object" && !Array.isArray(value)
+        ? value
+        : undefined;
+}
+function httpLogScalar(value) {
+    return typeof value === "string" || typeof value === "number"
+        ? String(value)
+        : undefined;
+}
+function boundedHttpLogBody(value) {
+    if (value === undefined)
+        return undefined;
+    const text = typeof value === "string" ? value : safeStringify(value);
+    const maxChars = 4000;
+    return text.length <= maxChars
+        ? text
+        : `${text.slice(0, maxChars)}\n… [truncated ${text.length - maxChars} chars]`;
+}
 function extractKvPairs(value) {
     const result = {};
     // Split on top-level spaces but keep {...}/[...] payloads intact.
@@ -1242,7 +1309,9 @@ export function renderBackendTestHtml(input) {
     const passRate = input.parsed.tests > 0 ? input.parsed.passed / input.parsed.tests : 0;
     const catalog = new Map((input.cases ?? []).map((item) => [item.id, item]));
     const caseValidation = parseAdvisorySummary(input.caseValidationSummary);
+    const caseCoverage = parseAdvisorySummary(input.caseCoverageSummary);
     const traceability = parseAdvisorySummary(input.traceabilitySummary);
+    const correspondence = parseAdvisorySummary(input.correspondenceSummary);
     const failures = input.parsed.cases.filter((result) => result.status !== "passed");
     const headColor = failed ? { fg: "#b42318", bg: "linear-gradient(135deg,#fef3f2 0,#fee4e2 100%)", border: "#fda29b", icon: "✗" } : { fg: "#067647", bg: "linear-gradient(135deg,#ecfdf3 0,#d1fadf 100%)", border: "#abefc6", icon: "✓" };
     const orderedCases = input.parsed.cases
@@ -1271,7 +1340,7 @@ export function renderBackendTestHtml(input) {
     const metricCard = (label, value, valueColor = "#172033") => `<div style="padding:14px 16px;border:1px solid #e3e8ef;border-radius:12px;background:#fbfcfe"><span style="color:#667085;font-size:0.78rem">${escapeHtml(label)}</span><div style="font-size:22px;font-weight:700;color:${valueColor}">${value}</div></div>`;
     const metrics = `<div style="display:grid;grid-template-columns:repeat(auto-fit,minmax(130px,1fr));gap:12px;margin-top:14px">${metricCard("用例总数", input.parsed.tests)}${metricCard("通过", input.parsed.passed, "#067647")}${metricCard("失败", input.parsed.failed)}${metricCard("错误", input.parsed.errors)}${metricCard("跳过", input.parsed.skipped)}<div style="padding:14px 16px;border:1px solid #e3e8ef;border-radius:12px;background:#fbfcfe"><span style="color:#667085;font-size:0.78rem">通过率</span><div style="font-size:22px;font-weight:700;color:#067647">${(passRate * 100).toFixed(2)}%</div><span style="color:#667085;font-size:0.78rem">耗时 ${formatDuration(input.parsed.durationMs)}</span></div></div>`;
     const summaryBanner = `<div style="display:flex;align-items:center;gap:16px;padding:16px 18px;border-radius:14px;background:${headColor.bg};border:1px solid ${headColor.border}"><div style="flex:0 0 auto;width:44px;height:44px;border-radius:999px;background:${headColor.fg};display:flex;align-items:center;justify-content:center;color:#fff;font-size:24px;font-weight:900;box-shadow:0 4px 12px ${headColor.fg}4d">${headColor.icon}</div><div><div style="color:${headColor.fg};font-size:1.05rem;font-weight:800;line-height:1.3">${failed ? "本轮测试未通过" : "本轮测试通过"}</div><div style="color:${headColor.fg};font-size:0.88rem;margin-top:2px">${failed ? `${input.parsed.failed + input.parsed.errors} 条用例失败或错误` : `${input.parsed.passed} 条用例全部执行成功`}</div></div></div>`;
-    const qualityBlock = `<div style="margin-top:18px"><div style="color:#17365d;font-size:16px;font-weight:700;margin:0 4px 6px">质量校验</div><div style="display:grid;grid-template-columns:repeat(auto-fit,minmax(240px,1fr));gap:12px;margin-top:8px">${renderQualityCard("Markdown 用例校验", caseValidation, input.caseValidationSummary)}${renderQualityCard("Markdown → pytest 追溯", traceability, input.traceabilitySummary)}</div></div>`;
+    const qualityBlock = `<div style="margin-top:18px"><div style="color:#17365d;font-size:16px;font-weight:700;margin:0 4px 6px">质量校验</div><div style="display:grid;grid-template-columns:repeat(auto-fit,minmax(240px,1fr));gap:12px;margin-top:8px">${renderQualityCard("Markdown 用例校验", caseValidation, input.caseValidationSummary)}${renderQualityCard("用例场景覆盖分析", caseCoverage, input.caseCoverageSummary)}${renderQualityCard("Markdown → pytest 追溯", traceability, input.traceabilitySummary)}${renderQualityCard("Markdown → pytest 一一对应", correspondence, input.correspondenceSummary)}</div></div>`;
     const failureOverview = failures.length > 0
         ? `<div style="margin-top:18px"><div style="color:#17365d;font-size:16px;font-weight:700;margin:0 4px 6px">失败概览</div><div style="display:grid;gap:10px;margin-top:8px">${failures.map((result) => {
             const caseId = resolveReportCaseId(result, catalog);
@@ -1381,6 +1450,118 @@ function dumpJacocoExec(host, port, destPath, connectTimeoutMs) {
         socket.pipe(writeStream);
     });
 }
+/**
+ * Deterministically derive a Backend Test Case Manifest v1 from final
+ * Markdown cases (testcase/md/**) and generated pytest scripts
+ * (testcase/test_*.py). No model involvement: AC IDs come from the case
+ * 验收标准/Acceptance Criteria section, automation status comes from whether
+ * a pytest test function references the Case ID. The output feeds
+ * computeCaseManifestCoverageSummary so the L-5 dashboard shows real
+ * AC/automation coverage instead of unavailable.
+ */
+export async function deriveBackendTestCaseManifest(input) {
+    const files = await markdownFiles(input.workspaceRoot);
+    const manifestCases = [];
+    for (const file of files) {
+        if (path.basename(file).toLowerCase() === "readme.md")
+            continue;
+        const content = await readFile(file, "utf8");
+        for (const testCase of splitCases(content)) {
+            const acIds = unique(testCase.body.match(AC_ID) ?? []);
+            if (acIds.length === 0)
+                continue; // skip cases without AC binding
+            const mappedScripts = extractMappedPytestScripts(testCase.body);
+            const category = inferCaseCategory(testCase.body);
+            let generated = false;
+            let symbol;
+            let scriptFile;
+            for (const script of mappedScripts) {
+                if (!isSafeBackendPytestScript(script, input.workspaceRoot))
+                    continue;
+                const absolute = path.resolve(input.workspaceRoot, script);
+                if (!(await exists(absolute)))
+                    continue;
+                const source = await readFile(absolute, "utf8");
+                const match = findCaseIdInPytest(source, testCase.id);
+                if (match) {
+                    generated = true;
+                    symbol = match;
+                    scriptFile = script;
+                    break;
+                }
+            }
+            manifestCases.push({
+                caseId: testCase.id,
+                acIds,
+                title: inferCaseTitle(testCase.body, testCase.id),
+                category,
+                automationStatus: generated ? "generated" : "planned",
+                ...(scriptFile ? { file: scriptFile } : {}),
+                ...(symbol ? { symbol } : {}),
+                ...(generated ? {} : { gapReason: "pytest symbol not yet generated or not associated" }),
+            });
+        }
+    }
+    const manifest = backendTestCaseManifestSchema.parse({
+        schemaVersion: 1,
+        sourceBinding: {
+            taskId: input.sourceBinding.taskId,
+            requirementPath: input.sourceBinding.requirementPath,
+            requirementSha256: input.sourceBinding.requirementSha256,
+            referencePaths: input.sourceBinding.referencePaths,
+            requirementIds: input.sourceBinding.requirementIds,
+        },
+        cases: manifestCases,
+        evidenceGaps: [],
+        coverageSummary: computeCaseManifestCoverageSummary({
+            cases: manifestCases,
+            evidenceGaps: [],
+            sourceBinding: input.sourceBinding,
+        }),
+    });
+    return manifest;
+}
+function inferCaseCategory(body) {
+    const text = body.toLowerCase();
+    if (/非法|无效|错误|不存在|越权|negative|invalid|unauthorized|forbidden|40[0-9]/.test(text))
+        return "negative";
+    if (/边界|boundary|edge|limit|max|min/.test(text))
+        return "boundary";
+    if (/鉴权|权限|登录|token|auth|permission|role/.test(text))
+        return "auth";
+    if (/状态|流转|state|transition|workflow/.test(text))
+        return "state-transition";
+    if (/超时|timeout/.test(text))
+        return "timeout";
+    if (/并发|concurren/.test(text))
+        return "concurrency";
+    return "positive";
+}
+function inferCaseTitle(body, caseId) {
+    // ## BE-XXX-001｜中文标题 → extract title after the delimiter
+    const heading = body.split(/\r?\n/, 1)[0] ?? "";
+    const delimiterMatch = heading.match(/[｜|]\s*(.+)$/);
+    if (delimiterMatch)
+        return delimiterMatch[1].trim();
+    return caseId;
+}
+function findCaseIdInPytest(source, caseId) {
+    for (const match of source.matchAll(/^(\s*)(?:async\s+)?def\s+(test_[A-Za-z0-9_]+)\s*\([^)]*\)\s*(?:->\s*[^:\r\n]+)?\s*:/gm)) {
+        const symbol = match[2];
+        const region = testFunctionRegion({
+            source,
+            functionIndex: match.index,
+            functionHeaderEnd: match.index + match[0].length,
+        });
+        const ids = new Set(caseIdsInText(region));
+        const fromSymbol = symbolCaseId(symbol);
+        if (fromSymbol)
+            ids.add(fromSymbol);
+        if (ids.has(caseId))
+            return symbol;
+    }
+    return undefined;
+}
 function inferModuleScript(classname) {
     // classname like "test_process_definition_list.py::TestX::test_a" → take the file stem.
     const first = classname.split("::")[0] ?? classname;
@@ -1444,7 +1625,9 @@ export function renderBackendTestL5Dashboard(input) {
         return `<div style="padding:13px 14px;border:1px solid #e4e9f1;border-radius:11px;background:#fbfcfe"><div style="display:flex;justify-content:space-between;align-items:center;gap:8px"><span style="color:#2a3c5a;font-size:12px;font-weight:700;font-family:ui-monospace,SFMono-Regular,Menlo,monospace">${escapeHtml(stat.script)}</span><span style="display:inline-flex;padding:2px 8px;border-radius:999px;font-size:11px;font-weight:800;color:${badge.fg};background:${badge.bg}">${badge.t}</span></div><div style="height:6px;margin:9px 0 6px;background:#edf1f6;border-radius:99px;overflow:hidden;display:flex"><i style="display:block;height:100%;width:${passPct}%;background:#15815c"></i><i style="display:block;height:100%;width:${failPct}%;background:#d34661"></i></div></div>`;
     }).join("");
     const caseValidation = parseAdvisorySummary(input.caseValidationSummary);
+    const caseCoverage = parseAdvisorySummary(input.caseCoverageSummary);
     const traceability = parseAdvisorySummary(input.traceabilitySummary);
+    const correspondence = parseAdvisorySummary(input.correspondenceSummary);
     const qRow = (name, status, extra = "") => {
         const c = status === "PASS" ? { fg: "#15815c", bg: "#e7f7f0" } : status === "FAIL" ? { fg: "#d34661", bg: "#fff0f3" } : { fg: "#718097", bg: "#eef2f7" };
         return `<div style="display:flex;align-items:center;justify-content:space-between;padding:11px 13px;border:1px solid #e4e9f1;border-radius:10px;background:#fbfcfe"><span style="color:#2a3c5a;font-size:13px;font-weight:650">${escapeHtml(name)}</span><span style="font-size:11px;font-weight:800;color:${c.fg};background:${c.bg};padding:2px 8px;border-radius:999px">${status === "Unavailable" ? "UNAVAILABLE" : status}${extra}</span></div>`;
@@ -1496,7 +1679,9 @@ export function renderBackendTestL5Dashboard(input) {
       <div style="display:grid;grid-template-columns:1fr 1fr;gap:8px">
         ${qRow("环境硬门（Shell）", "PASS")}
         ${qRow("Markdown 用例校验（节点4）", caseValidation.status, caseValidation.findings !== undefined ? ` · ${caseValidation.findings}` : "")}
+        ${qRow("用例场景覆盖分析（节点4）", caseCoverage.status, caseCoverage.findings !== undefined ? ` · ${caseCoverage.findings}` : "")}
         ${qRow("Markdown → pytest 追溯（节点6）", traceability.status, traceability.findings !== undefined ? ` · ${traceability.findings}` : "")}
+        ${qRow("Markdown → pytest 一一对应（节点6）", correspondence.status, correspondence.findings !== undefined ? ` · ${correspondence.findings}` : "")}
         ${qRow("HTTP 请求/响应日志（节点6 advisory）", traceability.status)}
         ${qRow("Markdown 敏感信息", "PASS")}
         ${qRow("代码覆盖率", m.lineCoverage.status === "unavailable" && m.lineCoverage.reason !== null ? "Unavailable" : "PASS")}
@@ -1504,14 +1689,16 @@ export function renderBackendTestL5Dashboard(input) {
     </section>
     ${failures ? `<section style="background:#fff;border:1px solid #e0e7f0;border-radius:15px;box-shadow:0 7px 20px rgba(25,53,92,.04);padding:22px;margin-top:20px"><div style="display:flex;align-items:center;justify-content:space-between;gap:14px;margin-bottom:12px"><h2 style="margin:0;color:#17365d;font-size:19px">缺陷记录</h2><span style="color:#718097;font-size:13px">${input.failures?.length ?? 0} 条 · 已关联用例与证据</span></div>${failures}</section>` : ""}
     <section style="background:#fff;border:1px solid #e0e7f0;border-radius:15px;box-shadow:0 7px 20px rgba(25,53,92,.04);padding:22px;margin-top:20px">
-      <p style="margin:0;color:#718097;font-size:12px">报告由 backend-test DAG 节点 7（execute-backend-pytest-and-html-report-shell）基于本轮真实执行产物确定性渲染。L-5 指标由 <code style="color:#17365d;font:11px ui-monospace,SFMono-Regular,Menlo,monospace">computeL5ReportMetrics</code> 机器计算。判定依据：pass=100%、AC=100%、automation≥90%、line≥80%、branch≥70%、skipped=0、无阻断级 Critical 风险。</p>
+      <p style="margin:0;color:#718097;font-size:12px">报告由 backend-test DAG 节点 8（execute-backend-pytest-and-html-report-shell）基于本轮真实执行产物确定性渲染。L-5 指标由 <code style="color:#17365d;font:11px ui-monospace,SFMono-Regular,Menlo,monospace">computeL5ReportMetrics</code> 机器计算。判定依据：pass=100%、AC=100%、automation≥90%、line≥80%、branch≥70%、skipped=0、无阻断级 Critical 风险。</p>
     </section>
   </div></body></html>`;
 }
 export function renderBackendTestFacts(input) {
     const passRate = input.parsed.tests > 0 ? input.parsed.passed / input.parsed.tests : 0;
     const caseValidation = parseAdvisorySummary(input.caseValidationSummary);
+    const caseCoverage = parseAdvisorySummary(input.caseCoverageSummary);
     const traceability = parseAdvisorySummary(input.traceabilitySummary);
+    const correspondence = parseAdvisorySummary(input.correspondenceSummary);
     const catalog = new Map((input.cases ?? []).map((entry) => [entry.id, entry]));
     return ([
         "# 后端自动化测试执行报告",
@@ -1533,7 +1720,9 @@ export function renderBackendTestFacts(input) {
         "| 校验项 | 状态 | Findings | 首要发现 |",
         "|---|---|---:|---|",
         `| Markdown 用例校验 | ${caseValidation.status} | ${caseValidation.findings ?? "不可用"} | ${caseValidation.firstFinding.replaceAll("|", "\\|")} |`,
+        `| 用例场景覆盖分析 | ${caseCoverage.status} | ${caseCoverage.findings ?? "不可用"} | ${caseCoverage.firstFinding.replaceAll("|", "\\|")} |`,
         `| Markdown → pytest 追溯 | ${traceability.status} | ${traceability.findings ?? "不可用"} | ${traceability.firstFinding.replaceAll("|", "\\|")} |`,
+        `| Markdown → pytest 一一对应 | ${correspondence.status} | ${correspondence.findings ?? "不可用"} | ${correspondence.firstFinding.replaceAll("|", "\\|")} |`,
         "",
         "## 失败概览",
         "",
@@ -1565,12 +1754,26 @@ export function renderBackendTestFacts(input) {
         "</details>",
         "",
         "<details>",
+        "<summary>用例场景覆盖分析原文</summary>",
+        "",
+        input.caseCoverageSummary?.trim() ?? "Unavailable",
+        "",
+        "</details>",
+        "",
+        "<details>",
         "<summary>Markdown → pytest 追溯原文</summary>",
         "",
         input.traceabilitySummary?.trim() ?? "Unavailable",
         "",
         "</details>",
         "",
+        "<details>",
+        "<summary>Markdown → pytest 一一对应原文</summary>",
+        "",
+        input.correspondenceSummary?.trim() ?? "Unavailable",
+        "",
+        "</details>",
+        "",
         "## 覆盖率证据",
         "",
         "- Code coverage: unavailable unless a separate validated coverage artifact exists.",

package/dist/workflows/dag/convergence/controller.js

@@ -8,6 +8,16 @@ const DEFAULT_CONVERGENCE_CHAIN_NODE_IDS = [
     "repair-pi",
     "hard-verify-shell",
 ];
+/**
+ * Review chain node ids that the supervised convergence loop may also observe.
+ * When these are part of the active convergence chain (via
+ * `spec.convergence.chainNodeIds`), a legitimate review `request-revision`
+ * re-enters the same bounded repair-reverify-review loop instead of only
+ * blocking closeout. Review protocol-invalid failures keep using their own
+ * protocol recovery; only a safe `request-revision` drives code repair here.
+ */
+const REVIEW_GATE_NODE_ID = "review-gate-shell";
+const REVIEW_VERDICT_NODE_ID = "review-verdict-recovery-pi";
 const CONVERGENCE_NON_RETRY_FAILURES = new Set([
     "timeout",
     "spawn-error",
@@ -17,6 +27,11 @@ const CONVERGENCE_NON_RETRY_FAILURES = new Set([
     "human-rejected",
     "decision-gate-requires-human",
 ]);
+const REVIEW_SOURCE_NON_RETRY_FAILURES = new Set([
+    ...CONVERGENCE_NON_RETRY_FAILURES,
+    "protocol-invalid",
+    "invalid-output",
+]);
 export function shouldEnableDagConvergence(spec) {
     return (process.env.HARNESS_DAG_CONVERGENCE !== "off" &&
         spec.convergence?.enabled === true);
@@ -33,10 +48,22 @@ export async function runConvergencePassController(input) {
         convergence.terminalReason = "unsupported-dag-shape";
         return { retry: false };
     }
+    const chain = getConvergenceChain(input.spec);
+    const observesReview = chain.includes(REVIEW_GATE_NODE_ID);
     const hardVerify = input.state.nodes["hard-verify-shell"];
     if (!hardVerify)
         return { retry: false };
-    if (hardVerify.status === "FINISHED") {
+    if (hardVerify.status === "ERROR") {
+        return handleHardVerifyFailure(input, hardVerify);
+    }
+    if (hardVerify.status !== "FINISHED")
+        return { retry: false };
+    // Hard verification passed. When the chain observes a review gate, success
+    // is gated on the review verdict: a legitimate `request-revision` re-enters
+    // the same bounded repair-reverify-review loop (AC3/AC4). Non-supervised
+    // DAGs without a review gate in the chain keep the original hard-verify-pass
+    // terminal behavior.
+    if (!observesReview) {
         convergence.terminalReason = "hard-verify-pass";
         await appendConvergenceKnowledgePattern({
             cwd: input.cwd,
@@ -44,8 +71,29 @@ export async function runConvergencePassController(input) {
         });
         return { retry: false };
     }
-    if (hardVerify.status !== "ERROR")
+    const reviewGate = input.state.nodes[REVIEW_GATE_NODE_ID];
+    if (!reviewGate)
         return { retry: false };
+    if (reviewGate.status === "FINISHED") {
+        convergence.terminalReason = "review-pass";
+        await appendConvergenceKnowledgePattern({
+            cwd: input.cwd,
+            state: input.state,
+        });
+        return { retry: false };
+    }
+    if (reviewGate.status === "ERROR") {
+        return handleReviewRequestRevision(input, reviewGate);
+    }
+    // Review gate still PENDING/SKIPPED mid-rank: wait for the next loop.
+    return { retry: false };
+}
+/**
+ * Hard verification failed. Apply the existing non-retry / regression /
+ * max-passes guards, then reset the convergence chain for another pass.
+ */
+async function handleHardVerifyFailure(input, hardVerify) {
+    const convergence = input.state.convergence;
     const currentPass = convergence.currentPass || 1;
     const hardFailure = hardVerify.failureCategory ?? "unknown";
     const passRecord = await buildConvergencePassRecord({
@@ -97,6 +145,69 @@ export async function runConvergencePassController(input) {
     await input.persistState();
     return { retry: true };
 }
+/**
+ * Hard verification passed but the review gate blocked closeout with a
+ * legitimate `request-revision` (review-gate-shell ERRORs when the verdict is
+ * not `pass`). Re-enter the same bounded recovery chain so repair can address
+ * the review findings, then re-verify and re-review. Review protocol-invalid
+ * and invalid-output sources, plus safety failures such as auth/write-guard,
+ * stay fail-closed and never enter code repair.
+ */
+async function handleReviewRequestRevision(input, reviewGate) {
+    const convergence = input.state.convergence;
+    const currentPass = convergence.currentPass || 1;
+    const reviewNode = input.state.nodes["review-pi"];
+    const reviewVerdictNode = input.state.nodes[REVIEW_VERDICT_NODE_ID];
+    // The gate commonly reports only nonzero-exit. Inspect the complete review
+    // chain so recovery output cannot launder provider/safety/protocol failures
+    // into automatic code repair.
+    const reviewFailure = [reviewGate, reviewVerdictNode, reviewNode]
+        .map((node) => node?.failureCategory)
+        .find((category) => category && REVIEW_SOURCE_NON_RETRY_FAILURES.has(category));
+    const legitimateRequestRevision = reviewNode?.status === "FINISHED" &&
+        parseProcessVerdict(reviewNode) === "request-revision" &&
+        reviewVerdictNode?.status === "FINISHED" &&
+        parseProcessVerdict(reviewVerdictNode) === "request-revision";
+    if (reviewFailure || !legitimateRequestRevision) {
+        const passRecord = await buildConvergencePassRecord({
+            pass: currentPass,
+            status: "terminal",
+            reason: "non-retry-failure",
+            state: input.state,
+            runDir: input.runDir,
+            spec: input.spec,
+        });
+        convergence.passHistory.push(passRecord);
+        convergence.terminalReason = "non-retry-failure";
+        await input.persistState();
+        return { retry: false };
+    }
+    const passRecord = await buildConvergencePassRecord({
+        pass: currentPass,
+        status: "retrying",
+        reason: "review-request-revision",
+        state: input.state,
+        runDir: input.runDir,
+        spec: input.spec,
+    });
+    if (currentPass >= convergence.maxPasses) {
+        passRecord.status = "terminal";
+        passRecord.reason = "max-passes";
+        convergence.passHistory.push(passRecord);
+        convergence.terminalReason = "max-passes";
+        await input.persistState();
+        return { retry: false };
+    }
+    convergence.passHistory.push(passRecord);
+    convergence.currentPass = currentPass + 1;
+    await resetConvergenceNodesForNextPass({
+        spec: input.spec,
+        state: input.state,
+        tasksById: input.tasksById,
+    });
+    await input.persistState();
+    return { retry: true };
+}
 function getConvergenceChain(spec) {
     return spec.convergence?.chainNodeIds ?? DEFAULT_CONVERGENCE_CHAIN_NODE_IDS;
 }
@@ -107,6 +218,8 @@ function hasConvergenceChain(tasksById, spec) {
 async function buildConvergencePassRecord(input) {
     const hardVerify = input.state.nodes["hard-verify-shell"];
     const processSupervisor = input.state.nodes["process-supervisor-pi"];
+    const reviewVerdictNode = input.state.nodes[REVIEW_VERDICT_NODE_ID];
+    const reviewGate = input.state.nodes[REVIEW_GATE_NODE_ID];
     const verifyEvidence = hardVerify?.verifyEvidence;
     return {
         pass: input.pass,
@@ -123,6 +236,9 @@ async function buildConvergencePassRecord(input) {
         verifyCommandCount: verifyEvidence?.commandCount,
         verifyCommandLabels: verifyEvidence?.commandLabels,
         shellSuccessCount: countSuccessfulShellCommands(hardVerify?.stdout),
+        reviewVerdict: parseProcessVerdict(reviewVerdictNode),
+        reviewGateStatus: reviewGate?.status,
+        reviewFailureCategory: reviewGate?.failureCategory,
         artifactRefs: await preserveConvergencePassArtifacts({
             pass: input.pass,
             state: input.state,
@@ -241,10 +357,17 @@ function extractSupervisorStructuredBlock(node) {
     // Legacy compatibility only. New supervisor prompts must emit REPAIR_ARTIFACT_JSON.
     const text = `${node?.assistantText ?? ""}\n${node?.stdout ?? ""}`;
     const block = {};
-    for (const key of ["FAILURE_CLASS", "FIX_SCOPE", "INVARIANT"]) {
-        const match = text.match(new RegExp(`^${key}:\\s*(.+)$`, "im"));
-        if (match?.[1])
-            block[key] = match[1].trim();
+    const acceptedKeys = new Set(["FAILURE_CLASS", "FIX_SCOPE", "INVARIANT"]);
+    for (const line of text.split(/\r?\n/)) {
+        const separatorIndex = line.indexOf(":");
+        if (separatorIndex <= 0)
+            continue;
+        const key = line.slice(0, separatorIndex).trim().toUpperCase();
+        if (!acceptedKeys.has(key))
+            continue;
+        const value = line.slice(separatorIndex + 1).trim();
+        if (value)
+            block[key] = value;
     }
     return block;
 }
@@ -252,9 +375,11 @@ async function resetConvergenceNodesForNextPass(input) {
     const chain = getConvergenceChain(input.spec);
     const resetIds = new Set(chain);
     for (const id of collectTransitiveDescendantTaskIds(input.spec, "hard-verify-shell")) {
-        const node = input.state.nodes[id];
-        if (node?.status === "SKIPPED")
-            resetIds.add(id);
+        // A new repair pass invalidates every downstream result, including nodes
+        // that already FINISHED (for example authority audit or failure-aware
+        // closeout). Reset the complete controlled descendant closure so the next
+        // pass cannot reuse stale governance or handoff evidence.
+        resetIds.add(id);
     }
     for (const id of resetIds) {
         const task = input.tasksById.get(id);