@tea-agent/loop-agent 0.33.5 → 0.33.6-beta.0

package/CHANGELOG.md

@@ -2,6 +2,14 @@
 
 ## [Unreleased]
 
+## [0.33.6-beta.0] - 2026-08-11
+
+### 修复
+
+- 修复 backend-test correspondence scanner 将下一 Case 的分隔注释误归入前一个 pytest function region，造成虚假的 `MULTIPLE_PYTEST`、script mismatch 和额外 Test Point；现在按函数实际缩进边界隔离 body，并只从精确 `Case-ID:` 元数据或 primary symbol 绑定 Case
+- 修复 backend-test 将 README 其他区域的 Markdown 链接误提取为业务模块、允许 `a401606` 等 opaque/hash-like stem 和异常多模块进入 map 的问题；现在只解析唯一 `## Module Index`，非法、mixed 或超过 8 个模块时 README-only 有界重试并由 manifest fail-closed
+- 修复 backend-test map 总 token 预算耗尽后只生成部分 Markdown/pytest 资产但 parent 仍可能显示成功的问题；budget-blocked child 现在使 backend map 明确 ERROR，并保留 expected/finished/blocked/token 事实
+
 ## [0.33.5] - 2026-08-11
 
 ### 重点更新

package/dist/workflows/dag/backend-test-case-coverage-analysis.js

@@ -909,6 +909,10 @@ function symbolCaseId(symbol) {
     const match = symbol.match(/^test_(BE(?:_[A-Z0-9]+)+?_\d{2,3})(?:_|$)/i);
     return match ? canonicalCaseId(match[1].replaceAll("_", "-")) : undefined;
 }
+function metadataCaseIds(region) {
+    return orderedUnique([...region.matchAll(/^\s*Case-ID\s*:\s*(BE-[A-Z0-9]+(?:-[A-Z0-9]+)*)\s*$/gmi)]
+        .map((match) => canonicalCaseId(match[1])));
+}
 function metadataTestPoints(region, label) {
     const escaped = label.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
     const match = new RegExp(`^\\s*${escaped}\\s*:\\s*(.*)$`, "mi").exec(region);
@@ -928,6 +932,27 @@ function pytestFunctionRegionStart(source, functionIndex) {
     }
     return regionStart;
 }
+function pytestFunctionBodyEnd(source, signatureEnd, functionIndent) {
+    let cursor = source.indexOf("\n", signatureEnd);
+    if (cursor < 0)
+        return source.length;
+    cursor += 1;
+    const baseIndent = functionIndent.replaceAll("\t", "    ").length;
+    while (cursor < source.length) {
+        const nextLine = source.indexOf("\n", cursor);
+        const lineEnd = nextLine < 0 ? source.length : nextLine;
+        const line = source.slice(cursor, lineEnd).replace(/\r$/, "");
+        if (line.trim()) {
+            const indentation = line.match(/^[ \t]*/)?.[0]?.replaceAll("\t", "    ").length ?? 0;
+            if (indentation <= baseIndent)
+                return cursor;
+        }
+        if (nextLine < 0)
+            return source.length;
+        cursor = nextLine + 1;
+    }
+    return source.length;
+}
 function pytestParameterCollections(source) {
     const collections = new Map();
     for (const match of source.matchAll(/^([A-Z][A-Z0-9_]*)\s*=\s*\[([\s\S]*?)^\]/gm)) {
@@ -941,10 +966,10 @@ function pytestSymbols(script, source) {
     const parameterCollections = pytestParameterCollections(source);
     return matches.map((match, index) => {
         const regionStart = regionStarts[index];
-        const regionEnd = regionStarts[index + 1] ?? source.length;
+        const regionEnd = pytestFunctionBodyEnd(source, match.index + match[0].length, match[1]);
         const region = source.slice(match.index, regionEnd);
         const decorators = source.slice(regionStart, match.index);
-        const ids = caseIds(region);
+        const ids = metadataCaseIds(region);
         const fromSymbol = symbolCaseId(match[2]);
         if (fromSymbol)
             ids.unshift(fromSymbol);

package/dist/workflows/dag/backend-test-module-stem.js

@@ -1,5 +1,7 @@
 import path from "node:path";
 const PRIORITY_ONLY_MODULE_STEMS = new Set(["p0", "p1", "p2"]);
+const OPAQUE_HASH_MODULE_STEM = /^[a-f][a-f0-9]{6,63}$/;
+export const BACKEND_TEST_MAX_MODULE_COUNT = 8;
 export function normalizeBackendTestModuleStemCandidate(raw) {
     return path.basename(raw)
         .replace(/\.(?:md|py)$/i, "")
@@ -11,6 +13,9 @@ export function normalizeBackendTestModuleStemCandidate(raw) {
 export function isPriorityOnlyBackendTestModuleStem(raw) {
     return PRIORITY_ONLY_MODULE_STEMS.has(normalizeBackendTestModuleStemCandidate(raw));
 }
+export function isOpaqueHashBackendTestModuleStem(raw) {
+    return OPAQUE_HASH_MODULE_STEM.test(normalizeBackendTestModuleStemCandidate(raw));
+}
 export function isPriorityOnlyBackendPytestScript(script) {
     const basename = path.posix.basename(script.replaceAll("\\", "/"));
     const match = /^test_(.+)\.py$/i.exec(basename);

package/dist/workflows/dag/backend-test-writer-completeness.js

@@ -2,7 +2,7 @@ import { createHash } from "node:crypto";
 import { access, mkdir, readFile, readdir, writeFile } from "node:fs/promises";
 import path from "node:path";
 import { z } from "zod";
-import { isPriorityOnlyBackendTestModuleStem } from "./backend-test-module-stem.js";
+import { BACKEND_TEST_MAX_MODULE_COUNT, isOpaqueHashBackendTestModuleStem, isPriorityOnlyBackendTestModuleStem, } from "./backend-test-module-stem.js";
 /**
  * Maps a backend-test generation writer task id to its writer-progress role.
  * Returns undefined for nodes that are not completeness-gated generators.
@@ -103,6 +103,8 @@ function moduleStemInvalidReason(raw) {
     const stem = normalizeBackendTestModuleStem(raw);
     if (isPriorityOnlyBackendTestModuleStem(stem))
         return "priority-only-module-stem";
+    if (isOpaqueHashBackendTestModuleStem(stem))
+        return "opaque-hash-module-stem";
     if (stem.toLowerCase() === "readme")
         return "reserved-module-stem";
     if (!/^[a-z][a-z0-9_]*$/.test(stem))
@@ -114,28 +116,40 @@ function moduleStemInvalidReason(raw) {
 function looksLikeValidModuleStem(raw) {
     return moduleStemInvalidReason(raw) === undefined;
 }
-/** Inspect only authoritative README module-index candidates. */
-export function inspectModuleStemsFromReadme(readme) {
-    const candidates = [];
-    const tableRowLines = readme
+function extractCanonicalModuleIndexSection(readme) {
+    const lines = readme
         .replaceAll("\r\n", "\n")
         .replaceAll("\r", "\n")
-        .split("\n")
-        .filter((line) => line.includes("|"));
+        .split("\n");
+    const headings = lines.flatMap((line, index) => line.trim() === "## Module Index" ? [index] : []);
+    if (headings.length === 0) {
+        return { section: "", structureIssues: ["missing-module-index"] };
+    }
+    if (headings.length > 1) {
+        return { section: "", structureIssues: ["duplicate-module-index"] };
+    }
+    const start = headings[0] + 1;
+    const relativeEnd = lines.slice(start).findIndex((line) => /^##\s+\S/.test(line.trim()));
+    const end = relativeEnd < 0 ? lines.length : start + relativeEnd;
+    return { section: lines.slice(start, end).join("\n"), structureIssues: [] };
+}
+/** Inspect only candidates inside the unique exact `## Module Index` section. */
+export function inspectModuleStemsFromReadme(readme) {
+    const extracted = extractCanonicalModuleIndexSection(readme);
+    const candidates = [];
+    const tableRowLines = extracted.section.split("\n").filter((line) => line.includes("|"));
     for (const line of tableRowLines) {
         for (const match of line.matchAll(/`?testcase\/md\/([A-Za-z0-9_.-]+)\.md`?/g)) {
             candidates.push(match[1]);
         }
         for (const match of line.matchAll(/\|\s*`?([A-Za-z0-9_.-]+)`?\s*\|\s*`?testcase\/test_/g)) {
-            // This compatibility shape can also capture a numeric count column.
-            // Keep valid stems and the priority-only defect we diagnose, while
-            // ignoring non-stem incidental cells such as `1`.
-            if (looksLikeValidModuleStem(match[1]) || isPriorityOnlyBackendTestModuleStem(match[1])) {
-                candidates.push(match[1]);
+            const raw = match[1];
+            if (looksLikeValidModuleStem(raw) || moduleStemInvalidReason(raw) !== "invalid-syntax") {
+                candidates.push(raw);
             }
         }
     }
-    for (const match of readme.matchAll(/\[[^\]]+\]\(\.\/([A-Za-z0-9_.-]+)\.md\)/g)) {
+    for (const match of extracted.section.matchAll(/\[[^\]]+\]\(\.\/([A-Za-z0-9_.-]+)\.md\)/g)) {
         candidates.push(match[1]);
     }
     const uniqueCandidates = orderedUnique(candidates);
@@ -145,12 +159,18 @@ export function inspectModuleStemsFromReadme(readme) {
             ? [{ raw, normalized: normalizeBackendTestModuleStem(raw), reasonCode }]
             : [];
     });
+    const validStems = orderedUnique(uniqueCandidates
+        .filter(looksLikeValidModuleStem)
+        .map((stem) => normalizeBackendTestModuleStem(stem)));
+    const structureIssues = [...extracted.structureIssues];
+    if (validStems.length > BACKEND_TEST_MAX_MODULE_COUNT) {
+        structureIssues.push(`excessive-module-count: ${validStems.length} > ${BACKEND_TEST_MAX_MODULE_COUNT}`);
+    }
     return {
         candidates: uniqueCandidates,
-        validStems: orderedUnique(uniqueCandidates
-            .filter(looksLikeValidModuleStem)
-            .map((stem) => normalizeBackendTestModuleStem(stem))),
+        validStems,
         invalid,
+        structureIssues,
     };
 }
 /**
@@ -917,6 +937,17 @@ export async function assessBackendTestMdPlanCompleteness(workspaceRoot) {
         // are produced by map_agent children and are NOT required here.
         const stemInspection = inspectModuleStemsFromReadme(readme);
         const indexedStems = stemInspection.validStems;
+        for (const structureIssue of stemInspection.structureIssues) {
+            if (!brokenPaths.includes("testcase/md/README.md")) {
+                brokenPaths.push("testcase/md/README.md");
+            }
+            issues.push({
+                code: "T5",
+                path: "testcase/md/README.md",
+                detail: structureIssue,
+                recoverable: true,
+            });
+        }
         for (const invalid of stemInspection.invalid) {
             if (!brokenPaths.includes("testcase/md/README.md")) {
                 brokenPaths.push("testcase/md/README.md");

package/dist/workflows/dag/dynamic-runtime/map.js

@@ -367,19 +367,26 @@ export async function executeDynamicMapExpansion(input) {
         }
     }
     const budgetBlockedIds = new Set(blockedChildren.map((entry) => entry.nodeId));
+    const failOnTokenBudgetExhaustion = input.expansion.failOnTokenBudgetExhaustion === true;
     const failedChildren = tolerateChildFailures
         ? []
         : childNodeIds.filter((nodeId) => {
             const record = input.state.nodes[nodeId];
             if (!record || record.status === "FINISHED")
                 return false;
-            if (budgetBlockedIds.has(nodeId) || isBudgetBlockedRecord(record))
-                return false;
+            if (budgetBlockedIds.has(nodeId) || isBudgetBlockedRecord(record)) {
+                return failOnTokenBudgetExhaustion;
+            }
             return true;
         });
+    const finishedItemCount = childNodeIds.filter((nodeId) => input.state.nodes[nodeId]?.status === "FINISHED").length;
+    const blockedItemCount = budgetBlockedIds.size;
     const aggregate = {
         workflowNodeId: input.expansion.workflowNodeId,
         itemCount: items.length,
+        expectedItemCount: items.length,
+        finishedItemCount,
+        blockedItemCount,
         children: childNodeIds.map((nodeId, index) => ({
             nodeId,
             item: items[index],
@@ -399,21 +406,30 @@ export async function executeDynamicMapExpansion(input) {
                 ?.reason,
         })),
         tokensUsed: totalTokensUsed,
+        maxTotalTokens: tokenBudget?.maxTotalTokens,
         caseOutcomeNotes: caseOutcomeNotes.length > 0 ? caseOutcomeNotes : undefined,
     };
+    const tokenBudgetFailure = failOnTokenBudgetExhaustion && blockedItemCount > 0;
     return {
-        // With tolerateChildFailures: barrier always succeeds; product outcomes live in
-        // case evidence + result materialization. Without it: real child failures fail-close.
-        ok: failedChildren.length === 0,
+        // With tolerateChildFailures: barrier succeeds and product outcomes live in
+        // case evidence. Backend-test opts into token-budget fail-closed because a
+        // partial generated module set is never a valid downstream input.
+        ok: failedChildren.length === 0 && !tokenBudgetFailure,
         stdout: JSON.stringify(aggregate),
         stderr: caseOutcomeNotes.length > 0
             ? `map children recorded as case outcomes: ${caseOutcomeNotes
                 .map((entry) => `${entry.nodeId}=${entry.status}:${entry.reason}`)
                 .join(", ")}`
+            : tokenBudgetFailure
+                ? `dynamic map token budget exhausted: expected=${items.length} finished=${finishedItemCount} blocked=${blockedItemCount} tokensUsed=${totalTokensUsed} maxTotalTokens=${tokenBudget?.maxTotalTokens ?? "unbounded"}`
+                : failedChildren.length > 0
+                    ? `dynamic map children failed: ${failedChildren.join(", ")}`
+                    : "",
+        failureCategory: tokenBudgetFailure
+            ? "dynamic-expansion-token-budget-exhausted"
             : failedChildren.length > 0
-                ? `dynamic map children failed: ${failedChildren.join(", ")}`
-                : "",
-        failureCategory: failedChildren.length > 0 ? "dynamic-expansion-child-failed" : "success",
+                ? "dynamic-expansion-child-failed"
+                : "success",
         durationMs: Date.now() - started,
     };
 }

package/dist/workflows/dag/init-hybrid.js

@@ -3508,23 +3508,29 @@ const readme=fs.existsSync('testcase/md/README.md')?fs.readFileSync('testcase/md
 const norm=s=>String(s).toLowerCase().replace(/[^a-z0-9]+/g,'_').replace(/^_+|_+$/g,'').replace(/_+/g,'_');
 const bt=String.fromCharCode(96);
 const stripBackticks=s=>s.split(bt).join('');
-const invalidReason=raw=>{const st=norm(raw);if(/^p[0-2]$/.test(st))return 'priority-only-module-stem';if(st==='readme')return 'reserved-module-stem';if(!/^[a-z][a-z0-9_]*$/.test(st))return 'invalid-syntax';if(/^(?:be|tp|ac|req|br)[_-]/i.test(st))return 'case-like-module-stem';return null;};
+const invalidReason=raw=>{const st=norm(raw);if(/^p[0-2]$/.test(st))return 'priority-only-module-stem';if(/^[a-f][a-f0-9]{6,63}$/.test(st))return 'opaque-hash-module-stem';if(st==='readme')return 'reserved-module-stem';if(!/^[a-z][a-z0-9_]*$/.test(st))return 'invalid-syntax';if(/^(?:be|tp|ac|req|br)[_-]/i.test(st))return 'case-like-module-stem';return null;};
 const valid=raw=>invalidReason(raw)===null;
 const rxMdPath=/testcase\\/md\\/([A-Za-z0-9_.-]+)\\.md/g;
 const rxTableRow=/\\|\\s*([A-Za-z0-9_.-]+)\\s*\\|\\s*testcase\\/test_/g;
 const rxRelLink=/\\[[^\\]]+\\]\\(\\.\\/([A-Za-z0-9_.-]+)\\.md\\)/g;
+const allLines=readme.replace(/\\r\\n/g,'\\n').replace(/\\r/g,'\\n').split('\\n');
+const headings=[];for(let i=0;i<allLines.length;i++){if(allLines[i].trim()==='## Module Index')headings.push(i);}
+if(headings.length!==1){process.stderr.write((headings.length===0?'missing-module-index':'duplicate-module-index')+'; require exactly one exact ## Module Index section\\n');process.exit(2);}
+const start=headings[0]+1;let end=allLines.length;for(let i=start;i<allLines.length;i++){if(/^##\\s+\\S/.test(allLines[i].trim())){end=i;break;}}
+const section=allLines.slice(start,end).join('\\n');
 const raw=[];
-const lines=readme.replace(/\\r\\n/g,'\\n').replace(/\\r/g,'\\n').split('\\n').filter(l=>l.includes('|'));
+const lines=section.split('\\n').filter(l=>l.includes('|'));
 for(const line of lines){
   const bare=stripBackticks(line);
   for(const m of bare.matchAll(rxMdPath)){raw.push(m[1]);}
-  for(const m of bare.matchAll(rxTableRow)){if(valid(m[1])||invalidReason(m[1])==='priority-only-module-stem')raw.push(m[1]);}
+  for(const m of bare.matchAll(rxTableRow)){if(valid(m[1])||invalidReason(m[1])!=='invalid-syntax')raw.push(m[1]);}
 }
-for(const m of readme.matchAll(rxRelLink)){raw.push(m[1]);}
+for(const m of section.matchAll(rxRelLink)){raw.push(m[1]);}
 const invalid=[];for(const r of raw){const reason=invalidReason(r);if(reason)invalid.push({stem:norm(r),reason});}
 if(invalid.length){for(const item of invalid)process.stderr.write(item.reason+': '+item.stem+'; use a stable business resource/domain stem\\n');process.exit(2);}
 const seen=new Set();const modules=[];
 for(const r of raw){const st=norm(r);if(valid(r)&&!seen.has(st)){seen.add(st);modules.push({stem:st});}}
+if(modules.length>8){process.stderr.write('excessive-module-count: '+modules.length+' > 8; merge by the smallest stable business resource/domain set\\n');process.exit(2);}
 process.stdout.write(JSON.stringify({modules}));
 `;
     const encoded = Buffer.from(script, "utf8").toString("base64");
@@ -3602,7 +3608,7 @@ async function buildBackendTestHybridDag(sources) {
             "Each Rule Key must appear in exactly one Matrix row. Preserve each AC/REQ/BR Rule Key as one row; if one product rule spans multiple dimensions, use a concise composite Dimension in that single row instead of duplicating the key. Derive OpenAPI Rule Keys exactly as the deterministic analyzer does: operation token is `<HTTP-METHOD>-<PATH>` with braces removed and every non-alphanumeric run replaced by a hyphen, uppercase (for example POST `/api/resource-notes` → `POST-API-RESOURCE-NOTES`); response statuses use `API-<OPERATION>-RESPONSE-STATUS`; body/parameter fields use `API-<OPERATION>-<FIELD>-REQUIRED|ENUM|MIN-LENGTH|MAX-LENGTH|MINIMUM|MAXIMUM|PATTERN|FORMAT`. Do not invent aliases such as API-CREATE-FIELDS when a deterministic key applies.",
             "Coverage priority is strict inside the declared scope: P0 product requirements/task hard constraints always remain in scope; P1 exhaustively supplements documented operations, fields, business rules, statuses and errors only for Affected Operations; P2 adds bounded protocol robustness only when it is relevant to the change and does not invent product behavior. Coverage percentages describe the declared affected scope, never whole-API completeness unless every operation is explicitly listed. Conflicts or undefined expectations must stay visible as GAP/CONFLICT with precise source pointers, never guessed.",
             "For uniqueness/lifecycle rules cover absent, active-existing, deleted-existing, create-delete-recreate, restore-then-recreate and documented scope/case-normalization states. For every enum cover every valid value plus bounded invalid equivalence classes (unknown, case variant, whitespace, empty, null/missing and wrong types as applicable). For every length/number rule cover min-1, min, nominal, max and max+1. For format rules cover each allowed class separately plus a valid mixed value, and representative forbidden classes including uppercase, internal/leading/trailing whitespace, tab/newline, unsupported punctuation, slash, emoji or control characters when the source contract supports that expectation.",
-            "Mandatory module index: include a `## Module Index` table in README that lists every planned module as a canonical relative link of the exact form `[label](./<stem>.md)` plus a `testcase/md/<stem>.md` path cell, so a downstream deterministic manifest can parse the module list. Group by stable business resource/domain, not by CRUD operation: one resource's list/detail/create/update/delete cases belong in one module such as `resource_notes`; split only when a single module would exceed the per-child 16K output budget, and keep the total module count at the smallest safe value. Name each module file with a stable lowercase business stem such as `health` or `resource_notes`. Do not use priority-only stems `p0`, `p1` or `p2`; Priority belongs only in the Coverage Matrix and never defines module files. Do not use Case-ID-like module filenames such as `BE-HEALTH.md` or `BE-NOTES.md`. The relative link target MUST equal the on-disk filename stem the sharded writer will create. For every automatable case, `自动化映射` must name exactly `testcase/test_<module>.py`, where <module> is that Markdown filename without `.md`, lowercased, with non-alphanumeric characters replaced by underscores. Example: `testcase/md/health.md` → `testcase/test_health.py`; `testcase/md/resource_notes.md` → `testcase/test_resource_notes.py`. Never invent a different pytest path in Markdown than the module stem implies.",
+            "Mandatory module index: include a `## Module Index` table in README that lists every planned module as a canonical relative link of the exact form `[label](./<stem>.md)` plus a `testcase/md/<stem>.md` path cell, so a downstream deterministic manifest can parse the module list. Group by stable business resource/domain, not by CRUD operation: one resource's list/detail/create/update/delete cases belong in one module such as `resource_notes`; split only when a single module would exceed the per-child 16K output protocol, keep the total module count at the smallest safe value, and never exceed 8 modules. Name each module file with a stable lowercase business stem such as `health` or `resource_notes`. Pure hexadecimal/hash-like opaque stems such as `a401606` or `deadbeef` are forbidden. Do not use priority-only stems `p0`, `p1` or `p2`; Priority belongs only in the Coverage Matrix and never defines module files. Do not use Case-ID-like module filenames such as `BE-HEALTH.md` or `BE-NOTES.md`. The relative link target MUST equal the on-disk filename stem the sharded writer will create. For every automatable case, `自动化映射` must name exactly `testcase/test_<module>.py`, where <module> is that Markdown filename without `.md`, lowercased, with non-alphanumeric characters replaced by underscores. Example: `testcase/md/health.md` → `testcase/test_health.py`; `testcase/md/resource_notes.md` → `testcase/test_resource_notes.py`. Never invent a different pytest path in Markdown than the module stem implies.",
             "Before finalizing README, calculate the predicted collected-item count as `sum(max(1, number of variant Test Points in each Case))`. If the task declares an item budget, the prediction must not exceed it. Reduce excess only by removing duplicate execution and converting same-request checkpoints to assertions; never drop required rules, boundaries, enums, operation-specific inputs, or business states. Record the prediction in README. Use only environment-supported fixtures/targets/isolation, record evidence gaps in Chinese, and do not emit JSON, pytest, or execute commands.",
             intake.boundedSourceContext,
             "## Authoritative reference index",
@@ -3647,10 +3653,11 @@ async function buildBackendTestHybridDag(sources) {
             workflowNodeId: "generate-backend-md-cases-map",
             itemsFrom: "$.nodes['materialize-backend-md-module-manifest-shell'].output.modules",
             itemName: "item",
-            maxItems: 64,
-            maxExpandedNodes: 64,
+            maxItems: 8,
+            maxExpandedNodes: 8,
             childIdPrefix: "generate-backend-md-case",
-            tokenBudget: { maxTokensPerCase: 16384, maxTotalTokens: 600000 },
+            tokenBudget: { maxTotalTokens: 600000 },
+            failOnTokenBudgetExhaustion: true,
             childTask: {
                 executor: "pi",
                 role: "implementer",
@@ -3673,7 +3680,7 @@ async function buildBackendTestHybridDag(sources) {
                     "The first non-empty response line must be exactly IMPLEMENTATION_OUTCOME: changed after the module file has been written, or IMPLEMENTATION_OUTCOME: blocked when precise missing evidence prevents safe generation. already-satisfied is not valid for this node.",
                     "Write human-readable content in Simplified Chinese by default. Keep English only for machine-readable IDs and technical literals such as Case/AC/REQ/BR IDs, HTTP methods, paths, field names, enum values, commands, filenames, code symbols and exact source citations.",
                     'Write the module {{item.stem}} as readable case cards covering every in-scope rule/Test Point the README Coverage Matrix assigns to this module. Every case starts with `## BE-<MODULE>-<NNN>｜<中文用例名称>`. `<NNN>` is exactly three zero-padded digits (`001`, `002`, ...), never two digits (`01`), a bare number, or an alphabetic suffix such as `011A`. Every case must include `### 覆盖规则`, `### 测试点`, `### 场景类型`, `### 前置条件`, `### 操作步骤`, `### 预期结果`, and `### 自动化映射` Do not group cases under "## 测试类 ..." (or any h2 grouping) headings that force Cases down to h3; each Case must be a direct h2 (`##`), and its seven sections must be h3 (`###`) children of that Case. If you need to convey a pytest class, state it inside the Case\'s `### 自动化映射` instead. Forbidden: `## 测试类 X` then `### BE-PD-001` and `### 覆盖规则` at the same h3 level. Required: `## BE-PD-001` then `### 覆盖规则`.; `覆盖规则` and `测试点` must reference exact Matrix Rule Keys/Test Points. Add `测试目的`, `验收标准`, `需求依据`, and `测试数据` for readable evidence. The `验收标准` section must list the exact applicable `AC-...` IDs, and every explicit task AC must appear in at least one Case. Every automatable case explicitly names its target pytest script and exactly one primary symbol so traceability scans only that script/symbol.',
-                    "Name this module file with the stable lowercase business stem `{{item.stem}}` (filename `testcase/md/{{item.stem}}.md`). Priority-only stems `p0`, `p1` and `p2` are forbidden and must never produce `p0.md` or `test_p0.py`. Do not use Case-ID-like module filenames. For every automatable case, `自动化映射` must name exactly `testcase/test_{{item.stem}}.py`, where the module stem is this Markdown filename without `.md`, lowercased, with non-alphanumeric characters replaced by underscores. Example: `health` → `testcase/test_health.py`; `resource_notes` → `testcase/test_resource_notes.py`. Never invent a different pytest path in Markdown than the module stem implies.",
+                    "Name this module file with the stable lowercase business stem `{{item.stem}}` (filename `testcase/md/{{item.stem}}.md`). Priority-only stems `p0`, `p1` and `p2` are forbidden and must never produce `p0.md` or `test_p0.py`. Pure hexadecimal/hash-like opaque stems such as `a401606` and `deadbeef` are also forbidden. Do not use Case-ID-like module filenames. For every automatable case, `自动化映射` must name exactly `testcase/test_{{item.stem}}.py`, where the module stem is this Markdown filename without `.md`, lowercased, with non-alphanumeric characters replaced by underscores. Example: `health` → `testcase/test_health.py`; `resource_notes` → `testcase/test_resource_notes.py`. Never invent a different pytest path in Markdown than the module stem implies.",
                     "Every Case must keep at least one numbered executable line under `### 操作步骤`; a compact variant/result table may follow but must not replace the numbered action anchor. Keep numbered/bulleted independently assertable results under `### 预期结果`. The exact `### 操作步骤` and `### 预期结果` headings must remain present for every Case, including compact/table-based Cases; never compress later Cases by dropping required headings. Every result must name the observable HTTP status, response field/value, state transition or membership condition, never vague wording such as ‘符合预期’.",
                     "In every `自动化映射`, use exactly these machine-readable list labels: `脚本`, `primary symbol`, `变体测试点`, `场景断言测试点`, `横切证据测试点`. Each Test Point from `### 测试点` must appear in exactly one binding list, and every Test Point named in any binding list must also be declared in that Case's `### 测试点`; write `无` for an empty list. A variant Test Point is atomic: one exact endpoint/input/precondition/outcome row equals one exact pytest item and one exact TP ID. If a parameter table has five rows, declare five distinct variant TP IDs in Markdown; never declare one family TP and append row suffixes only in pytest. Classify as `variant` only when endpoint, request input, precondition business state, or expected outcome genuinely changes and therefore needs an independent pytest parameter item. Classify CRUD checkpoints, status/body/header/schema assertions and multiple checks over the same response/journey as `assertion`; classify shared HTTP logging/redaction/truncation evidence as `cross-cutting`. Never create a Test Point merely to parameterize a checkpoint. Every non-cross-cutting TP ID is owned by exactly one Case; when the same response/schema/error assertion is needed in different Cases, use distinct Case-specific TP IDs instead of reusing one assertion TP across Cases. Keep the script path identical to the module one-to-one path and declare exactly one primary symbol named with the canonical Case prefix, for example `BE-RN-003` → `test_BE_RN_003_<description>`; non-Case-prefixed primary symbols are forbidden because parameterized item association must remain deterministic. For redaction scenarios, list sensitive header/field key names only. Never write any header-name-and-value pair, credential placeholder, fake token, anti-example, or other secret-shaped literal in Markdown; state only that a test-only value is supplied at runtime and omitted. Put implementation-only restrictions in a concise `<details>` block rather than dominating the main case flow. Use only environment-supported fixtures/targets/isolation, record evidence gaps in Chinese, and do not emit JSON, pytest, or execute commands.",
                     intake.boundedSourceContext,
@@ -3777,10 +3784,11 @@ async function buildBackendTestHybridDag(sources) {
             workflowNodeId: "generate-backend-pytest-cases-map",
             itemsFrom: "$.nodes['materialize-backend-pytest-module-manifest-shell'].output.modules",
             itemName: "item",
-            maxItems: 64,
-            maxExpandedNodes: 64,
+            maxItems: 8,
+            maxExpandedNodes: 8,
             childIdPrefix: "generate-backend-pytest-case",
-            tokenBudget: { maxTokensPerCase: 16384, maxTotalTokens: 600000 },
+            tokenBudget: { maxTotalTokens: 600000 },
+            failOnTokenBudgetExhaustion: true,
             childTask: {
                 executor: "pi",
                 role: "implementer",

package/dist/workflows/dag/types.js

@@ -555,6 +555,13 @@ export const dagDynamicExpansionSchema = z.object({
         maxTotalTokens: z.number().int().positive().optional(),
     })
         .optional(),
+    /**
+     * When true, any child blocked by the aggregate token budget fails the map
+     * barrier instead of being retained as a case-level blocked outcome.
+     * Backend-test enables this because partial Markdown/pytest module sets are
+     * invalid generation assets; frontend-test keeps the default false.
+     */
+    failOnTokenBudgetExhaustion: z.boolean().optional(),
     /**
      * When true, map child ERROR/auth/timeout is recorded as case-level
      * failed/blocked evidence and the map barrier still succeeds (frontend-test).

package/docs/templates/backend-test-dag.json

@@ -140,7 +140,7 @@
 				]
 			},
 			"outputContract": "Write a Chinese, human-readable testcase/md/README.md as the single Markdown-first entry page with Coverage Scope, Coverage Matrix and a machine-parseable module index. Do not write module case cards here; do not execute pytest or modify production code/config.",
-			"subtask_prompt": "This is a required file-generation node. After reading the bounded inputs, immediately use write tools to create testcase/md/README.md. Do not end after analysis or planning, and do not return before a non-empty bounded diff exists. Write ONLY testcase/md/README.md in this node; module case cards are written by downstream sharded nodes.\n\nOutput budget protocol (hard, max output <=16K per turn): Never paste full Matrix, case bodies, or source text into assistant chat. README holds only Scope+Matrix+module index; never inline full case bodies. If a Completeness Gate / OUTPUT_LIMIT_RECOVERY retry is injected, continue only listed target paths.\n\nThe first non-empty response line must be exactly IMPLEMENTATION_OUTCOME: changed after the README has been written, or IMPLEMENTATION_OUTCOME: blocked when precise missing evidence prevents safe generation. already-satisfied is not valid for this node.\n\nRead the upstream environment report. Generate the Markdown-first backend test README under testcase/md/README.md.\n\nWrite human-readable content in Simplified Chinese by default. Keep English only for machine-readable IDs and technical literals such as Case/AC/REQ/BR IDs, HTTP methods, paths, field names, enum values, commands, filenames, code symbols and exact source citations.\n\nCreate testcase/md/README.md as the concise entry page: test objective, target/environment, isolation/cleanup, module summary and a linked case index table with Case ID, Chinese case name, scenario type, endpoint and expected status/result. Avoid repeating every case body in README.\n\nBefore the Coverage Matrix, write a mandatory machine-readable `## Coverage Scope` section in README using exactly `| Field | Value |`, immediately followed by the separator row `|---|---|`, and these six unique rows: `Change Classification`, `Coverage Policy`, `Affected Operations`, `Affected Rule Keys`, `Regression Floor`, `Scope Evidence`. Always set `Change Classification` to `new-operation` and `Coverage Policy` to `full-contract`; do NOT reason about whether operations are new or existing. Cover all in-scope rules from the requirement document at full depth; treat the product requirement as the coverage baseline and use API contract evidence (fields/status/enum/boundary/format) to supplement scenario dimensions. Scope is limited to operations/rules the requirement document (or its referenced API contract) explicitly describes; do not expand to unrelated operations that the requirement does not mention. List affected operations exactly as `METHOD /path`, stable rule keys separated by semicolons, and precise source pointers as Scope Evidence.\n\nCoverage depth is full over the in-scope rules: fully cover every documented status, request/response field rule, requiredness, enum, boundary, format, auth and business state of each affected operation the requirement describes, but do not re-test unrelated operations the requirement does not mention. Inspect shared validator/helper/DTO/query builder evidence and expand Affected Operations when the same affected path can affect them; unresolved impact stays visible as GAP/CONFLICT.\n\nBefore writing cases, build the mandatory machine-readable Coverage Matrix inside `testcase/md/README.md` itself. Its section heading line must be exactly `## Coverage Matrix` with no numeric prefix/suffix; never place the canonical Matrix only in a module file. Use this exact header: `| Rule Key | Priority | Source | Endpoint/Field | Dimension | Rule | Required Test Points | Case IDs | Status |`. Every data row must contain exactly 9 pipe-delimited cells and must never omit `Dimension`; use concise dimensions such as requirement, operation, response-status, requiredness, enum, boundary, format, business-state or error. Use only P0/P1/P2 and COVERED/PARTIAL/GAP/CONFLICT. Use stable `TP-<UPPERCASE-HYPHENATED-ID>` test points separated by semicolons.\n\nEach Rule Key must appear in exactly one Matrix row. Preserve each AC/REQ/BR Rule Key as one row; if one product rule spans multiple dimensions, use a concise composite Dimension in that single row instead of duplicating the key. Derive OpenAPI Rule Keys exactly as the deterministic analyzer does: operation token is `<HTTP-METHOD>-<PATH>` with braces removed and every non-alphanumeric run replaced by a hyphen, uppercase (for example POST `/api/resource-notes` → `POST-API-RESOURCE-NOTES`); response statuses use `API-<OPERATION>-RESPONSE-STATUS`; body/parameter fields use `API-<OPERATION>-<FIELD>-REQUIRED|ENUM|MIN-LENGTH|MAX-LENGTH|MINIMUM|MAXIMUM|PATTERN|FORMAT`. Do not invent aliases such as API-CREATE-FIELDS when a deterministic key applies.\n\nCoverage priority is strict inside the declared scope: P0 product requirements/task hard constraints always remain in scope; P1 exhaustively supplements documented operations, fields, business rules, statuses and errors only for Affected Operations; P2 adds bounded protocol robustness only when it is relevant to the change and does not invent product behavior. Coverage percentages describe the declared affected scope, never whole-API completeness unless every operation is explicitly listed. Conflicts or undefined expectations must stay visible as GAP/CONFLICT with precise source pointers, never guessed.\n\nFor uniqueness/lifecycle rules cover absent, active-existing, deleted-existing, create-delete-recreate, restore-then-recreate and documented scope/case-normalization states. For every enum cover every valid value plus bounded invalid equivalence classes (unknown, case variant, whitespace, empty, null/missing and wrong types as applicable). For every length/number rule cover min-1, min, nominal, max and max+1. For format rules cover each allowed class separately plus a valid mixed value, and representative forbidden classes including uppercase, internal/leading/trailing whitespace, tab/newline, unsupported punctuation, slash, emoji or control characters when the source contract supports that expectation.\n\nMandatory module index: include a `## Module Index` table in README that lists every planned module as a canonical relative link of the exact form `[label](./<stem>.md)` plus a `testcase/md/<stem>.md` path cell, so a downstream deterministic manifest can parse the module list. Group by stable business resource/domain, not by CRUD operation: one resource's list/detail/create/update/delete cases belong in one module such as `resource_notes`; split only when a single module would exceed the per-child 16K output budget, and keep the total module count at the smallest safe value. Name each module file with a stable lowercase business stem such as `health` or `resource_notes`. Do not use priority-only stems `p0`, `p1` or `p2`; Priority belongs only in the Coverage Matrix and never defines module files. Do not use Case-ID-like module filenames such as `BE-HEALTH.md` or `BE-NOTES.md`. The relative link target MUST equal the on-disk filename stem the sharded writer will create. For every automatable case, `自动化映射` must name exactly `testcase/test_<module>.py`, where <module> is that Markdown filename without `.md`, lowercased, with non-alphanumeric characters replaced by underscores. Example: `testcase/md/health.md` → `testcase/test_health.py`; `testcase/md/resource_notes.md` → `testcase/test_resource_notes.py`. Never invent a different pytest path in Markdown than the module stem implies.\n\nBefore finalizing README, calculate the predicted collected-item count as `sum(max(1, number of variant Test Points in each Case))`. If the task declares an item budget, the prediction must not exceed it. Reduce excess only by removing duplicate execution and converting same-request checkpoints to assertions; never drop required rules, boundaries, enums, operation-specific inputs, or business states. Record the prediction in README. Use only environment-supported fixtures/targets/isolation, record evidence gaps in Chinese, and do not emit JSON, pytest, or execute commands.\n\n## Derived task contract: 需求.md\n\n# Backend test\n- AC-001 proof\n\n## Authoritative reference index\n\n[]\n\nFor each index entry, use `readPath` for Pi read-tool calls and copy `path` exactly into Markdown Source References. Bound files under .harness/tasks/<taskId>/source/** are read-only inputs: reading them is allowed even though writing .harness/** is forbidden. Never resolve `path` relative to the repository root, search for substitutes, or fall back to docs/** when a bound read fails.\n\nRead only precise indexed references needed for AC/API/field/rule evidence; references remain authoritative over derived text."
+			"subtask_prompt": "This is a required file-generation node. After reading the bounded inputs, immediately use write tools to create testcase/md/README.md. Do not end after analysis or planning, and do not return before a non-empty bounded diff exists. Write ONLY testcase/md/README.md in this node; module case cards are written by downstream sharded nodes.\n\nOutput budget protocol (hard, max output <=16K per turn): Never paste full Matrix, case bodies, or source text into assistant chat. README holds only Scope+Matrix+module index; never inline full case bodies. If a Completeness Gate / OUTPUT_LIMIT_RECOVERY retry is injected, continue only listed target paths.\n\nThe first non-empty response line must be exactly IMPLEMENTATION_OUTCOME: changed after the README has been written, or IMPLEMENTATION_OUTCOME: blocked when precise missing evidence prevents safe generation. already-satisfied is not valid for this node.\n\nRead the upstream environment report. Generate the Markdown-first backend test README under testcase/md/README.md.\n\nWrite human-readable content in Simplified Chinese by default. Keep English only for machine-readable IDs and technical literals such as Case/AC/REQ/BR IDs, HTTP methods, paths, field names, enum values, commands, filenames, code symbols and exact source citations.\n\nCreate testcase/md/README.md as the concise entry page: test objective, target/environment, isolation/cleanup, module summary and a linked case index table with Case ID, Chinese case name, scenario type, endpoint and expected status/result. Avoid repeating every case body in README.\n\nBefore the Coverage Matrix, write a mandatory machine-readable `## Coverage Scope` section in README using exactly `| Field | Value |`, immediately followed by the separator row `|---|---|`, and these six unique rows: `Change Classification`, `Coverage Policy`, `Affected Operations`, `Affected Rule Keys`, `Regression Floor`, `Scope Evidence`. Always set `Change Classification` to `new-operation` and `Coverage Policy` to `full-contract`; do NOT reason about whether operations are new or existing. Cover all in-scope rules from the requirement document at full depth; treat the product requirement as the coverage baseline and use API contract evidence (fields/status/enum/boundary/format) to supplement scenario dimensions. Scope is limited to operations/rules the requirement document (or its referenced API contract) explicitly describes; do not expand to unrelated operations that the requirement does not mention. List affected operations exactly as `METHOD /path`, stable rule keys separated by semicolons, and precise source pointers as Scope Evidence.\n\nCoverage depth is full over the in-scope rules: fully cover every documented status, request/response field rule, requiredness, enum, boundary, format, auth and business state of each affected operation the requirement describes, but do not re-test unrelated operations the requirement does not mention. Inspect shared validator/helper/DTO/query builder evidence and expand Affected Operations when the same affected path can affect them; unresolved impact stays visible as GAP/CONFLICT.\n\nBefore writing cases, build the mandatory machine-readable Coverage Matrix inside `testcase/md/README.md` itself. Its section heading line must be exactly `## Coverage Matrix` with no numeric prefix/suffix; never place the canonical Matrix only in a module file. Use this exact header: `| Rule Key | Priority | Source | Endpoint/Field | Dimension | Rule | Required Test Points | Case IDs | Status |`. Every data row must contain exactly 9 pipe-delimited cells and must never omit `Dimension`; use concise dimensions such as requirement, operation, response-status, requiredness, enum, boundary, format, business-state or error. Use only P0/P1/P2 and COVERED/PARTIAL/GAP/CONFLICT. Use stable `TP-<UPPERCASE-HYPHENATED-ID>` test points separated by semicolons.\n\nEach Rule Key must appear in exactly one Matrix row. Preserve each AC/REQ/BR Rule Key as one row; if one product rule spans multiple dimensions, use a concise composite Dimension in that single row instead of duplicating the key. Derive OpenAPI Rule Keys exactly as the deterministic analyzer does: operation token is `<HTTP-METHOD>-<PATH>` with braces removed and every non-alphanumeric run replaced by a hyphen, uppercase (for example POST `/api/resource-notes` → `POST-API-RESOURCE-NOTES`); response statuses use `API-<OPERATION>-RESPONSE-STATUS`; body/parameter fields use `API-<OPERATION>-<FIELD>-REQUIRED|ENUM|MIN-LENGTH|MAX-LENGTH|MINIMUM|MAXIMUM|PATTERN|FORMAT`. Do not invent aliases such as API-CREATE-FIELDS when a deterministic key applies.\n\nCoverage priority is strict inside the declared scope: P0 product requirements/task hard constraints always remain in scope; P1 exhaustively supplements documented operations, fields, business rules, statuses and errors only for Affected Operations; P2 adds bounded protocol robustness only when it is relevant to the change and does not invent product behavior. Coverage percentages describe the declared affected scope, never whole-API completeness unless every operation is explicitly listed. Conflicts or undefined expectations must stay visible as GAP/CONFLICT with precise source pointers, never guessed.\n\nFor uniqueness/lifecycle rules cover absent, active-existing, deleted-existing, create-delete-recreate, restore-then-recreate and documented scope/case-normalization states. For every enum cover every valid value plus bounded invalid equivalence classes (unknown, case variant, whitespace, empty, null/missing and wrong types as applicable). For every length/number rule cover min-1, min, nominal, max and max+1. For format rules cover each allowed class separately plus a valid mixed value, and representative forbidden classes including uppercase, internal/leading/trailing whitespace, tab/newline, unsupported punctuation, slash, emoji or control characters when the source contract supports that expectation.\n\nMandatory module index: include a `## Module Index` table in README that lists every planned module as a canonical relative link of the exact form `[label](./<stem>.md)` plus a `testcase/md/<stem>.md` path cell, so a downstream deterministic manifest can parse the module list. Group by stable business resource/domain, not by CRUD operation: one resource's list/detail/create/update/delete cases belong in one module such as `resource_notes`; split only when a single module would exceed the per-child 16K output protocol, keep the total module count at the smallest safe value, and never exceed 8 modules. Name each module file with a stable lowercase business stem such as `health` or `resource_notes`. Pure hexadecimal/hash-like opaque stems such as `a401606` or `deadbeef` are forbidden. Do not use priority-only stems `p0`, `p1` or `p2`; Priority belongs only in the Coverage Matrix and never defines module files. Do not use Case-ID-like module filenames such as `BE-HEALTH.md` or `BE-NOTES.md`. The relative link target MUST equal the on-disk filename stem the sharded writer will create. For every automatable case, `自动化映射` must name exactly `testcase/test_<module>.py`, where <module> is that Markdown filename without `.md`, lowercased, with non-alphanumeric characters replaced by underscores. Example: `testcase/md/health.md` → `testcase/test_health.py`; `testcase/md/resource_notes.md` → `testcase/test_resource_notes.py`. Never invent a different pytest path in Markdown than the module stem implies.\n\nBefore finalizing README, calculate the predicted collected-item count as `sum(max(1, number of variant Test Points in each Case))`. If the task declares an item budget, the prediction must not exceed it. Reduce excess only by removing duplicate execution and converting same-request checkpoints to assertions; never drop required rules, boundaries, enums, operation-specific inputs, or business states. Record the prediction in README. Use only environment-supported fixtures/targets/isolation, record evidence gaps in Chinese, and do not emit JSON, pytest, or execute commands.\n\n## Derived task contract: 需求.md\n\n# Backend test\n- AC-001 proof\n\n## Authoritative reference index\n\n[]\n\nFor each index entry, use `readPath` for Pi read-tool calls and copy `path` exactly into Markdown Source References. Bound files under .harness/tasks/<taskId>/source/** are read-only inputs: reading them is allowed even though writing .harness/** is forbidden. Never resolve `path` relative to the repository root, search for substitutes, or fall back to docs/** when a bound read fails.\n\nRead only precise indexed references needed for AC/API/field/rule evidence; references remain authoritative over derived text."
 		},
 		{
 			"id": "materialize-backend-md-module-manifest-shell",
@@ -164,7 +164,7 @@
 			"subtask_prompt": "Parse testcase/md/README.md and emit exactly one trailing JSON line {modules:[{stem}]} listing every trusted module stem (table-row testcase/md/<stem>.md mentions and canonical [label](./<stem>.md) relative links only). No file writes.",
 			"shell": {
 				"commands": [
-					"node -e \"eval(Buffer.from('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','base64').toString('utf8'))\""
+					"node -e \"eval(Buffer.from('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','base64').toString('utf8'))\""
 				],
 				"cwd": ".",
 				"timeoutMs": 60000
@@ -195,13 +195,13 @@
 				"workflowNodeId": "generate-backend-md-cases-map",
 				"itemsFrom": "$.nodes['materialize-backend-md-module-manifest-shell'].output.modules",
 				"itemName": "item",
-				"maxItems": 64,
-				"maxExpandedNodes": 64,
+				"maxItems": 8,
+				"maxExpandedNodes": 8,
 				"childIdPrefix": "generate-backend-md-case",
 				"tokenBudget": {
-					"maxTokensPerCase": 16384,
 					"maxTotalTokens": 600000
 				},
+				"failOnTokenBudgetExhaustion": true,
 				"childTask": {
 					"executor": "pi",
 					"role": "implementer",
@@ -238,7 +238,7 @@
 						]
 					},
 					"outputContract": "Write exactly one Chinese module Markdown case-card file testcase/md/<stem>.md with BE-<MODULE>-<NNN> cases and the seven required h3 sections; keep machine IDs/literals exact and do not execute pytest or modify production code/config or the README.",
-					"subtaskPromptTemplate": "This is a required file-generation node for exactly one Markdown module. After reading testcase/md/README.md (Coverage Scope + Coverage Matrix + module index) and the bounded references, immediately use write tools to create the single file testcase/md/{{item.stem}}.md. Do not end after analysis or planning, and do not return before a non-empty bounded diff exists. Do not modify README.md or any other module file.\n\nOutput budget protocol (hard, max output <=16K per turn): Never paste full Matrix, other modules' case bodies, or source text into assistant chat. Each write/edit tool call touches at most one file (this module). Compact tables/lists are required; omitting required sections or in-scope variants is forbidden. If a Completeness Gate / OUTPUT_LIMIT_RECOVERY retry is injected, continue only listed target paths.\n\nThe first non-empty response line must be exactly IMPLEMENTATION_OUTCOME: changed after the module file has been written, or IMPLEMENTATION_OUTCOME: blocked when precise missing evidence prevents safe generation. already-satisfied is not valid for this node.\n\nWrite human-readable content in Simplified Chinese by default. Keep English only for machine-readable IDs and technical literals such as Case/AC/REQ/BR IDs, HTTP methods, paths, field names, enum values, commands, filenames, code symbols and exact source citations.\n\nWrite the module {{item.stem}} as readable case cards covering every in-scope rule/Test Point the README Coverage Matrix assigns to this module. Every case starts with `## BE-<MODULE>-<NNN>｜<中文用例名称>`. `<NNN>` is exactly three zero-padded digits (`001`, `002`, ...), never two digits (`01`), a bare number, or an alphabetic suffix such as `011A`. Every case must include `### 覆盖规则`, `### 测试点`, `### 场景类型`, `### 前置条件`, `### 操作步骤`, `### 预期结果`, and `### 自动化映射` Do not group cases under \"## 测试类 ...\" (or any h2 grouping) headings that force Cases down to h3; each Case must be a direct h2 (`##`), and its seven sections must be h3 (`###`) children of that Case. If you need to convey a pytest class, state it inside the Case's `### 自动化映射` instead. Forbidden: `## 测试类 X` then `### BE-PD-001` and `### 覆盖规则` at the same h3 level. Required: `## BE-PD-001` then `### 覆盖规则`.; `覆盖规则` and `测试点` must reference exact Matrix Rule Keys/Test Points. Add `测试目的`, `验收标准`, `需求依据`, and `测试数据` for readable evidence. The `验收标准` section must list the exact applicable `AC-...` IDs, and every explicit task AC must appear in at least one Case. Every automatable case explicitly names its target pytest script and exactly one primary symbol so traceability scans only that script/symbol.\n\nName this module file with the stable lowercase business stem `{{item.stem}}` (filename `testcase/md/{{item.stem}}.md`). Priority-only stems `p0`, `p1` and `p2` are forbidden and must never produce `p0.md` or `test_p0.py`. Do not use Case-ID-like module filenames. For every automatable case, `自动化映射` must name exactly `testcase/test_{{item.stem}}.py`, where the module stem is this Markdown filename without `.md`, lowercased, with non-alphanumeric characters replaced by underscores. Example: `health` → `testcase/test_health.py`; `resource_notes` → `testcase/test_resource_notes.py`. Never invent a different pytest path in Markdown than the module stem implies.\n\nEvery Case must keep at least one numbered executable line under `### 操作步骤`; a compact variant/result table may follow but must not replace the numbered action anchor. Keep numbered/bulleted independently assertable results under `### 预期结果`. The exact `### 操作步骤` and `### 预期结果` headings must remain present for every Case, including compact/table-based Cases; never compress later Cases by dropping required headings. Every result must name the observable HTTP status, response field/value, state transition or membership condition, never vague wording such as ‘符合预期’.\n\nIn every `自动化映射`, use exactly these machine-readable list labels: `脚本`, `primary symbol`, `变体测试点`, `场景断言测试点`, `横切证据测试点`. Each Test Point from `### 测试点` must appear in exactly one binding list, and every Test Point named in any binding list must also be declared in that Case's `### 测试点`; write `无` for an empty list. A variant Test Point is atomic: one exact endpoint/input/precondition/outcome row equals one exact pytest item and one exact TP ID. If a parameter table has five rows, declare five distinct variant TP IDs in Markdown; never declare one family TP and append row suffixes only in pytest. Classify as `variant` only when endpoint, request input, precondition business state, or expected outcome genuinely changes and therefore needs an independent pytest parameter item. Classify CRUD checkpoints, status/body/header/schema assertions and multiple checks over the same response/journey as `assertion`; classify shared HTTP logging/redaction/truncation evidence as `cross-cutting`. Never create a Test Point merely to parameterize a checkpoint. Every non-cross-cutting TP ID is owned by exactly one Case; when the same response/schema/error assertion is needed in different Cases, use distinct Case-specific TP IDs instead of reusing one assertion TP across Cases. Keep the script path identical to the module one-to-one path and declare exactly one primary symbol named with the canonical Case prefix, for example `BE-RN-003` → `test_BE_RN_003_<description>`; non-Case-prefixed primary symbols are forbidden because parameterized item association must remain deterministic. For redaction scenarios, list sensitive header/field key names only. Never write any header-name-and-value pair, credential placeholder, fake token, anti-example, or other secret-shaped literal in Markdown; state only that a test-only value is supplied at runtime and omitted. Put implementation-only restrictions in a concise `<details>` block rather than dominating the main case flow. Use only environment-supported fixtures/targets/isolation, record evidence gaps in Chinese, and do not emit JSON, pytest, or execute commands.\n\n## Derived task contract: 需求.md\n\n# Backend test\n- AC-001 proof\n\n## Authoritative reference index\n\n[]\n\nFor each index entry, use `readPath` for Pi read-tool calls and copy `path` exactly into Markdown Source References. Bound files under .harness/tasks/<taskId>/source/** are read-only inputs: reading them is allowed even though writing .harness/** is forbidden. Never resolve `path` relative to the repository root, search for substitutes, or fall back to docs/** when a bound read fails.\n\nRead only precise indexed references needed for AC/API/field/rule evidence; references remain authoritative over derived text."
+					"subtaskPromptTemplate": "This is a required file-generation node for exactly one Markdown module. After reading testcase/md/README.md (Coverage Scope + Coverage Matrix + module index) and the bounded references, immediately use write tools to create the single file testcase/md/{{item.stem}}.md. Do not end after analysis or planning, and do not return before a non-empty bounded diff exists. Do not modify README.md or any other module file.\n\nOutput budget protocol (hard, max output <=16K per turn): Never paste full Matrix, other modules' case bodies, or source text into assistant chat. Each write/edit tool call touches at most one file (this module). Compact tables/lists are required; omitting required sections or in-scope variants is forbidden. If a Completeness Gate / OUTPUT_LIMIT_RECOVERY retry is injected, continue only listed target paths.\n\nThe first non-empty response line must be exactly IMPLEMENTATION_OUTCOME: changed after the module file has been written, or IMPLEMENTATION_OUTCOME: blocked when precise missing evidence prevents safe generation. already-satisfied is not valid for this node.\n\nWrite human-readable content in Simplified Chinese by default. Keep English only for machine-readable IDs and technical literals such as Case/AC/REQ/BR IDs, HTTP methods, paths, field names, enum values, commands, filenames, code symbols and exact source citations.\n\nWrite the module {{item.stem}} as readable case cards covering every in-scope rule/Test Point the README Coverage Matrix assigns to this module. Every case starts with `## BE-<MODULE>-<NNN>｜<中文用例名称>`. `<NNN>` is exactly three zero-padded digits (`001`, `002`, ...), never two digits (`01`), a bare number, or an alphabetic suffix such as `011A`. Every case must include `### 覆盖规则`, `### 测试点`, `### 场景类型`, `### 前置条件`, `### 操作步骤`, `### 预期结果`, and `### 自动化映射` Do not group cases under \"## 测试类 ...\" (or any h2 grouping) headings that force Cases down to h3; each Case must be a direct h2 (`##`), and its seven sections must be h3 (`###`) children of that Case. If you need to convey a pytest class, state it inside the Case's `### 自动化映射` instead. Forbidden: `## 测试类 X` then `### BE-PD-001` and `### 覆盖规则` at the same h3 level. Required: `## BE-PD-001` then `### 覆盖规则`.; `覆盖规则` and `测试点` must reference exact Matrix Rule Keys/Test Points. Add `测试目的`, `验收标准`, `需求依据`, and `测试数据` for readable evidence. The `验收标准` section must list the exact applicable `AC-...` IDs, and every explicit task AC must appear in at least one Case. Every automatable case explicitly names its target pytest script and exactly one primary symbol so traceability scans only that script/symbol.\n\nName this module file with the stable lowercase business stem `{{item.stem}}` (filename `testcase/md/{{item.stem}}.md`). Priority-only stems `p0`, `p1` and `p2` are forbidden and must never produce `p0.md` or `test_p0.py`. Pure hexadecimal/hash-like opaque stems such as `a401606` and `deadbeef` are also forbidden. Do not use Case-ID-like module filenames. For every automatable case, `自动化映射` must name exactly `testcase/test_{{item.stem}}.py`, where the module stem is this Markdown filename without `.md`, lowercased, with non-alphanumeric characters replaced by underscores. Example: `health` → `testcase/test_health.py`; `resource_notes` → `testcase/test_resource_notes.py`. Never invent a different pytest path in Markdown than the module stem implies.\n\nEvery Case must keep at least one numbered executable line under `### 操作步骤`; a compact variant/result table may follow but must not replace the numbered action anchor. Keep numbered/bulleted independently assertable results under `### 预期结果`. The exact `### 操作步骤` and `### 预期结果` headings must remain present for every Case, including compact/table-based Cases; never compress later Cases by dropping required headings. Every result must name the observable HTTP status, response field/value, state transition or membership condition, never vague wording such as ‘符合预期’.\n\nIn every `自动化映射`, use exactly these machine-readable list labels: `脚本`, `primary symbol`, `变体测试点`, `场景断言测试点`, `横切证据测试点`. Each Test Point from `### 测试点` must appear in exactly one binding list, and every Test Point named in any binding list must also be declared in that Case's `### 测试点`; write `无` for an empty list. A variant Test Point is atomic: one exact endpoint/input/precondition/outcome row equals one exact pytest item and one exact TP ID. If a parameter table has five rows, declare five distinct variant TP IDs in Markdown; never declare one family TP and append row suffixes only in pytest. Classify as `variant` only when endpoint, request input, precondition business state, or expected outcome genuinely changes and therefore needs an independent pytest parameter item. Classify CRUD checkpoints, status/body/header/schema assertions and multiple checks over the same response/journey as `assertion`; classify shared HTTP logging/redaction/truncation evidence as `cross-cutting`. Never create a Test Point merely to parameterize a checkpoint. Every non-cross-cutting TP ID is owned by exactly one Case; when the same response/schema/error assertion is needed in different Cases, use distinct Case-specific TP IDs instead of reusing one assertion TP across Cases. Keep the script path identical to the module one-to-one path and declare exactly one primary symbol named with the canonical Case prefix, for example `BE-RN-003` → `test_BE_RN_003_<description>`; non-Case-prefixed primary symbols are forbidden because parameterized item association must remain deterministic. For redaction scenarios, list sensitive header/field key names only. Never write any header-name-and-value pair, credential placeholder, fake token, anti-example, or other secret-shaped literal in Markdown; state only that a test-only value is supplied at runtime and omitted. Put implementation-only restrictions in a concise `<details>` block rather than dominating the main case flow. Use only environment-supported fixtures/targets/isolation, record evidence gaps in Chinese, and do not emit JSON, pytest, or execute commands.\n\n## Derived task contract: 需求.md\n\n# Backend test\n- AC-001 proof\n\n## Authoritative reference index\n\n[]\n\nFor each index entry, use `readPath` for Pi read-tool calls and copy `path` exactly into Markdown Source References. Bound files under .harness/tasks/<taskId>/source/** are read-only inputs: reading them is allowed even though writing .harness/** is forbidden. Never resolve `path` relative to the repository root, search for substitutes, or fall back to docs/** when a bound read fails.\n\nRead only precise indexed references needed for AC/API/field/rule evidence; references remain authoritative over derived text."
 				}
 			}
 		},
@@ -338,7 +338,7 @@
 			"subtask_prompt": "Parse testcase/md/README.md and emit exactly one trailing JSON line {modules:[{stem}]} listing every trusted module stem (table-row testcase/md/<stem>.md mentions and canonical [label](./<stem>.md) relative links only). No file writes.",
 			"shell": {
 				"commands": [
-					"node -e \"eval(Buffer.from('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','base64').toString('utf8'))\""
+					"node -e \"eval(Buffer.from('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','base64').toString('utf8'))\""
 				],
 				"cwd": ".",
 				"timeoutMs": 60000
@@ -369,13 +369,13 @@
 				"workflowNodeId": "generate-backend-pytest-cases-map",
 				"itemsFrom": "$.nodes['materialize-backend-pytest-module-manifest-shell'].output.modules",
 				"itemName": "item",
-				"maxItems": 64,
-				"maxExpandedNodes": 64,
+				"maxItems": 8,
+				"maxExpandedNodes": 8,
 				"childIdPrefix": "generate-backend-pytest-case",
 				"tokenBudget": {
-					"maxTokensPerCase": 16384,
 					"maxTotalTokens": 600000
 				},
+				"failOnTokenBudgetExhaustion": true,
 				"childTask": {
 					"executor": "pi",
 					"role": "implementer",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tea-agent/loop-agent",
-  "version": "0.33.5",
+  "version": "0.33.6-beta.0",
   "type": "module",
   "bin": {
     "loop-agent": "bin/loop-agent.js",