@danceiny/gotry 0.0.1-rc.21 → 0.0.1-rc.24

package/dist/scripts/benchmark-environment-bridge-e2e.js

@@ -1,3 +1,10 @@
+/** Offline contract for the opt-in benchmark environment bridge.
+ *
+ * Covers default-off, explicit opt-in, and fail-closed configuration paths.
+ * A local developer run exercises the source checkout. The packaged consumer
+ * path is built from the current root @deepseek-ai/dsh 0.1.5-rc.1 closure;
+ * version/source counterexamples use isolated synthetic fixtures.
+ */
 import assert from 'node:assert/strict';
 import { createServer } from 'node:http';
 import { chmodSync, cpSync, existsSync, mkdirSync, mkdtempSync, readFileSync, rmSync, writeFileSync } from 'node:fs';
@@ -6,115 +13,55 @@ import { tmpdir } from 'node:os';
 import { dirname, join } from 'node:path';
 import { pathToFileURL } from 'node:url';
 import { spawn } from 'node:child_process';
-import { benchmarkRuntimeSupported, selectDshCwd, selectDshRuntime, supportsNodeVersion } from '../../bin/gotry-runtime-resolution.js';
+import { benchmarkRuntimeSupported, selectDshCwd, selectDshRuntime, supportsNodeVersion, } from '../../bin/gotry-runtime-resolution.js';
 const ROOT = join(import.meta.dirname, '..', '..');
 const BIN = join(ROOT, 'bin', 'gotry-inner.js');
 const TOOL = 'gotry_benchmark_environment';
 const MARKER = 'BENCHMARK_BRIDGE_LOOKUP_OK';
 const TIMEOUT_MS = 30_000;
-const LOOKUP_INPUT_SCHEMA = {
-    type: 'object',
-    properties: {
-        city: {
-            type: 'string',
-            enum: [
-                'Dubai',
-                'Singapore'
-            ]
-        }
-    },
-    required: [
-        'city'
-    ],
-    additionalProperties: false
-};
+const LOOKUP_INPUT_SCHEMA = { type: 'object', properties: { city: { type: 'string', enum: ['Dubai', 'Singapore'] } }, required: ['city'], additionalProperties: false };
 const TSX_LOADER = pathToFileURL(createRequire(import.meta.url).resolve('tsx')).href;
 function runRuntimeProbe(options) {
     const fixture = mkdtempSync(join(tmpdir(), 'gotry-runtime-probe-'));
     try {
-        writeFileSync(join(fixture, 'package.json'), JSON.stringify({
-            name: 'runtime-probe',
-            type: 'module'
-        }));
-        const writeDsh = (root, version)=>{
-            mkdirSync(join(root, 'lib'), {
-                recursive: true
-            });
-            writeFileSync(join(root, 'package.json'), JSON.stringify({
-                name: '@deepseek-ai/dsh',
-                version,
-                type: 'module'
-            }));
+        writeFileSync(join(fixture, 'package.json'), JSON.stringify({ name: 'runtime-probe', type: 'module' }));
+        const writeDsh = (root, version) => {
+            mkdirSync(join(root, 'lib'), { recursive: true });
+            writeFileSync(join(root, 'package.json'), JSON.stringify({ name: '@deepseek-ai/dsh', version, type: 'module' }));
             writeFileSync(join(root, 'lib', 'bin.js'), 'export {}\n');
         };
-        if (options.rootVersion) writeDsh(join(fixture, 'node_modules', '@deepseek-ai', 'dsh'), options.rootVersion);
-        if (options.vendorVersion) writeDsh(join(fixture, 'ts', 'dsh-runtime', 'node_modules', '@deepseek-ai', 'dsh'), options.vendorVersion);
+        if (options.rootVersion)
+            writeDsh(join(fixture, 'node_modules', '@deepseek-ai', 'dsh'), options.rootVersion);
+        if (options.vendorVersion)
+            writeDsh(join(fixture, 'ts', 'dsh-runtime', 'node_modules', '@deepseek-ai', 'dsh'), options.vendorVersion);
         const runtime = selectDshRuntime({
             repoRoot: fixture,
             rootResolver: createRequire(join(fixture, 'package.json')),
-            benchmark: options.benchmark === true
-        });
-        return runtime ? {
-            source: runtime.source,
-            version: runtime.version
-        } : null;
-    } finally{
-        rmSync(fixture, {
-            recursive: true,
-            force: true
+            benchmark: options.benchmark === true,
         });
+        return runtime ? { source: runtime.source, version: runtime.version } : null;
+    }
+    finally {
+        rmSync(fixture, { recursive: true, force: true });
     }
 }
 function assertRuntimeSelectionAndVersionGuards() {
-    const sourcePriority = runRuntimeProbe({
-        rootVersion: '0.1.2-alpha.3',
-        vendorVersion: '0.1.2-alpha.1'
-    });
-    assert.deepEqual(sourcePriority, {
-        source: 'root',
-        version: '0.1.2-alpha.3'
-    }, 'source checkout uses the root dsh package even when legacy vendor is alpha.1');
-    const legacyFallback = runRuntimeProbe({
-        vendorVersion: '0.1.2-alpha.1'
-    });
+    const sourcePriority = runRuntimeProbe({ rootVersion: '0.1.5-rc.1', vendorVersion: '0.1.2-alpha.1' });
+    assert.deepEqual(sourcePriority, { source: 'root', version: '0.1.5-rc.1' }, 'source checkout uses the root dsh package even when legacy vendor is alpha.1');
+    const legacyFallback = runRuntimeProbe({ vendorVersion: '0.1.2-alpha.1' });
     assert.deepEqual(legacyFallback, null, 'non-benchmark source checkout fail-closes instead of using the removed legacy vendored dsh fallback');
-    const wrongBenchmarkVersion = runRuntimeProbe({
-        rootVersion: '0.1.2-alpha.1',
-        vendorVersion: '0.1.2-alpha.1',
-        benchmark: true
-    });
-    assert.deepEqual(wrongBenchmarkVersion, {
-        source: 'root',
-        version: '0.1.2-alpha.1'
-    });
-    assert.equal(benchmarkRuntimeSupported(wrongBenchmarkVersion), false, 'benchmark mode rejects a non-alpha.3 dsh runtime before spawn');
-    assert.equal(runRuntimeProbe({
-        vendorVersion: '0.1.2-alpha.1',
-        benchmark: true
-    }), null, 'benchmark mode never falls back to legacy vendored dsh');
+    const wrongBenchmarkVersion = runRuntimeProbe({ rootVersion: '0.1.2-alpha.1', vendorVersion: '0.1.2-alpha.1', benchmark: true });
+    assert.deepEqual(wrongBenchmarkVersion, { source: 'root', version: '0.1.2-alpha.1' });
+    assert.equal(benchmarkRuntimeSupported(wrongBenchmarkVersion), false, 'benchmark mode rejects a non-target dsh runtime before spawn');
+    assert.equal(runRuntimeProbe({ vendorVersion: '0.1.2-alpha.1', benchmark: true }), null, 'benchmark mode never falls back to legacy vendored dsh');
     assert.equal(supportsNodeVersion('22.14.0'), false, 'Node 22.14 is rejected before dsh resolution/spawn');
     assert.equal(supportsNodeVersion('22.15.0'), true, 'Node 22.15 is the accepted minimum');
     assert.equal(supportsNodeVersion('24.0.0'), true, 'newer Node majors remain accepted');
     const invocationCwd = join(tmpdir(), 'gotry-runtime-invocation-cwd');
     const sourceStateRoot = join(ROOT, 'ts/dsh-runtime');
-    assert.equal(selectDshCwd({
-        repoRoot: ROOT,
-        invocationCwd,
-        sourceCheckoutMode: true,
-        benchmark: false
-    }), sourceStateRoot, 'source checkout normal mode keeps dsh cwd at ts/dsh-runtime for gotry-state continuity');
-    assert.equal(selectDshCwd({
-        repoRoot: ROOT,
-        invocationCwd,
-        sourceCheckoutMode: true,
-        benchmark: true
-    }), invocationCwd, 'source checkout benchmark mode uses the isolated invocation cwd');
-    assert.equal(selectDshCwd({
-        repoRoot: ROOT,
-        invocationCwd,
-        sourceCheckoutMode: false,
-        benchmark: false
-    }), invocationCwd, 'installed package normal mode uses the user invocation cwd');
+    assert.equal(selectDshCwd({ repoRoot: ROOT, invocationCwd, sourceCheckoutMode: true, benchmark: false }), sourceStateRoot, 'source checkout normal mode keeps dsh cwd at ts/dsh-runtime for gotry-state continuity');
+    assert.equal(selectDshCwd({ repoRoot: ROOT, invocationCwd, sourceCheckoutMode: true, benchmark: true }), invocationCwd, 'source checkout benchmark mode uses the isolated invocation cwd');
+    assert.equal(selectDshCwd({ repoRoot: ROOT, invocationCwd, sourceCheckoutMode: false, benchmark: false }), invocationCwd, 'installed package normal mode uses the user invocation cwd');
 }
 function writeResolutionProbe(path, resultPath, block) {
     writeFileSync(path, `
@@ -167,197 +114,93 @@ Module.createRequire = function (...args) {
   return required
 }
 Module.syncBuiltinESMExports()
-`, {
-        mode: 0o600,
-        flag: 'wx'
-    });
+`, { mode: 0o600, flag: 'wx' });
 }
 function sse(payload) {
     return `data: ${JSON.stringify(payload)}\n\n`;
 }
 function finalText(text) {
-    return sse({
-        id: 'bridge-final',
-        object: 'chat.completion.chunk',
-        choices: [
-            {
-                delta: {
-                    role: 'assistant',
-                    content: text
-                },
-                finish_reason: null
-            }
-        ]
-    }) + sse({
-        id: 'bridge-final-stop',
-        object: 'chat.completion.chunk',
-        choices: [
-            {
-                delta: {},
-                finish_reason: 'stop'
-            }
-        ]
-    }) + 'data: [DONE]\n\n';
+    return sse({ id: 'bridge-final', object: 'chat.completion.chunk', choices: [{ delta: { role: 'assistant', content: text }, finish_reason: null }] })
+        + sse({ id: 'bridge-final-stop', object: 'chat.completion.chunk', choices: [{ delta: {}, finish_reason: 'stop' }] }) + 'data: [DONE]\n\n';
 }
 function toolCall(callId = 'bridge-call-1', city = 'Dubai') {
-    return sse({
-        id: `bridge-${callId}`,
-        object: 'chat.completion.chunk',
-        choices: [
-            {
-                delta: {
-                    role: 'assistant',
-                    tool_calls: [
-                        {
-                            index: 0,
-                            id: callId,
-                            type: 'function',
-                            function: {
-                                name: TOOL,
-                                arguments: JSON.stringify({
-                                    action: 'call',
-                                    tool: 'lookup',
-                                    arguments: {
-                                        city
-                                    }
-                                })
-                            }
-                        }
-                    ]
-                },
-                finish_reason: null
-            }
-        ]
-    }) + sse({
-        id: 'bridge-call-stop',
-        object: 'chat.completion.chunk',
-        choices: [
-            {
-                delta: {},
-                finish_reason: 'tool_calls'
-            }
-        ]
-    }) + 'data: [DONE]\n\n';
+    return sse({ id: `bridge-${callId}`, object: 'chat.completion.chunk', choices: [{ delta: { role: 'assistant', tool_calls: [{ index: 0, id: callId, type: 'function', function: { name: TOOL, arguments: JSON.stringify({ action: 'call', tool: 'lookup', arguments: { city } }) } }] }, finish_reason: null }] })
+        + sse({ id: 'bridge-call-stop', object: 'chat.completion.chunk', choices: [{ delta: {}, finish_reason: 'tool_calls' }] }) + 'data: [DONE]\n\n';
 }
 function names(body) {
-    return (body.tools ?? []).map((t)=>{
-        const f = t.function;
-        return String(f?.name ?? t.name ?? '');
-    }).filter(Boolean);
+    return (body.tools ?? []).map(t => { const f = t.function; return String(f?.name ?? t.name ?? ''); }).filter(Boolean);
 }
 function toolResultPresent(body) {
-    return (body.messages ?? []).some((m)=>m.role === 'tool' && JSON.stringify(m).includes(MARKER));
+    return (body.messages ?? []).some(m => m.role === 'tool' && JSON.stringify(m).includes(MARKER));
 }
 function anyToolResultPresent(body) {
-    return (body.messages ?? []).some((m)=>m.role === 'tool');
+    return (body.messages ?? []).some(m => m.role === 'tool');
 }
-const BRIDGE_E2E_BODY_SCHEMA = {
-    type: 'object',
-    properties: {
-        status: {
-            type: 'string'
-        },
-        payload: {
-            type: 'string'
-        }
-    },
-    required: [],
-    additionalProperties: false
-};
+// Round 12(#215):bridge config 显式携带无数据值 closed terminal body schema;
+// 覆盖两个 e2e 终态形态 {"status":"succeeded"} 与 {payload:"…"}。
+const BRIDGE_E2E_BODY_SCHEMA = { type: 'object', properties: { status: { type: 'string' }, payload: { type: 'string' } }, required: [], additionalProperties: false };
 const TERMINAL_OUTLINE = 'object{?status:string,?payload:string}';
 async function runCase(mode, executableOverride, extraEnv = {}) {
     const requests = [];
     let servedToolCalls = 0;
     let spawnTarget = '';
     const domainRecoveryMode = mode === 'domain-recovery' || mode === 'domain-recovery-failed';
-    const server = createServer((req, res)=>{
+    const server = createServer((req, res) => {
         const chunks = [];
-        req.on('data', (c)=>chunks.push(Buffer.from(c)));
-        req.on('end', ()=>{
+        req.on('data', c => chunks.push(Buffer.from(c)));
+        req.on('end', () => {
             let body = {};
             try {
                 body = JSON.parse(Buffer.concat(chunks).toString());
-            } catch  {}
+            }
+            catch { /* diagnostic remains structural */ }
             requests.push(body);
-            res.writeHead(200, {
-                'content-type': 'text/event-stream'
-            });
-            if (mode === 'spawn-failed' && names(body).includes(TOOL) && !anyToolResultPresent(body) && spawnTarget) rmSync(spawnTarget, {
-                force: true
-            });
+            res.writeHead(200, { 'content-type': 'text/event-stream' });
+            if (mode === 'spawn-failed' && names(body).includes(TOOL) && !anyToolResultPresent(body) && spawnTarget)
+                rmSync(spawnTarget, { force: true });
             if (domainRecoveryMode && servedToolCalls === 1 && names(body).includes(TOOL) && anyToolResultPresent(body) && !toolResultPresent(body)) {
                 servedToolCalls += 1;
                 res.end(toolCall('bridge-call-2', 'Singapore'));
-            } else if (mode !== 'disabled' && mode !== 'invalid-path' && mode !== 'invalid-schema' && mode !== 'unsafe-config' && names(body).includes(TOOL) && !anyToolResultPresent(body)) {
+            }
+            else if (mode !== 'disabled' && mode !== 'invalid-path' && mode !== 'invalid-schema' && mode !== 'unsafe-config' && names(body).includes(TOOL) && !anyToolResultPresent(body)) {
                 servedToolCalls += 1;
                 res.end(toolCall());
-            } else {
+            }
+            else {
                 res.end(finalText('<benchmark_terminal>{"status":"succeeded"}</benchmark_terminal>'));
             }
         });
     });
-    await new Promise((resolve)=>server.listen(0, '127.0.0.1', resolve));
+    await new Promise(resolve => server.listen(0, '127.0.0.1', resolve));
     const port = server.address().port;
     const cwd = mkdtempSync(join(tmpdir(), 'gotry-bridge-cwd-'));
     const dsh = mkdtempSync(join(tmpdir(), 'gotry-bridge-dsh-'));
     const probe = join(cwd, 'benchmark-resolution-probe.cjs');
     const probeResult = join(cwd, 'benchmark-resolution-hits.json');
-    writeFileSync(probeResult, '', {
-        mode: 0o600,
-        flag: 'wx'
-    });
+    writeFileSync(probeResult, '', { mode: 0o600, flag: 'wx' });
     writeResolutionProbe(probe, probeResult, mode === 'disabled');
     const runner = join(cwd, 'synthetic-runner.js');
     const runnerTrace = join(cwd, 'synthetic-runner-trace.jsonl');
     spawnTarget = join(cwd, 'synthetic-spawn-target.js');
     const configPath = join(cwd, mode === 'invalid-path' ? 'benchmark-env-config-\n.json' : 'benchmark-env-config.json');
-    const runnerBody = mode === 'domain-recovery-failed' ? `if (args.city === 'Dubai') process.stdout.write(JSON.stringify({ schema_version: 'gotry_benchmark_tool_result_v1', status: 'miss', code: 'NOT_FOUND', recovery: 'revise_arguments' })); else { process.stderr.write('PRIVATE_RECOVERY_RUNNER_DIAGNOSTIC_DO_NOT_REFLECT'); process.exit(17) }` : mode === 'domain-recovery' ? `if (args.city === 'Dubai') process.stdout.write(JSON.stringify({ schema_version: 'gotry_benchmark_tool_result_v1', status: 'miss', code: 'NOT_FOUND', recovery: 'revise_arguments' })); else process.stdout.write(JSON.stringify({ schema_version: 'gotry_benchmark_tool_result_v1', status: 'ok', result: { marker: '${MARKER}', leaked: [] } }))` : mode === 'timeout' ? `setTimeout(() => {}, 60_000)` : mode === 'runner-failed' ? `process.stderr.write('PRIVATE_RUNNER_DIAGNOSTIC_DO_NOT_REFLECT'); process.exit(17)` : mode === 'output-truncated' ? `process.stdout.write('x'.repeat(20_000))` : mode === 'unexpected-output' ? `process.stdout.write(JSON.stringify({ schema_version: 'gotry_benchmark_tool_result_v1', status: 'ok', result: { marker: '${MARKER}', leaked: [], unexpected: 'must-not-reflect' } }))` : `const forbidden = ['GOTRY_BENCHMARK_ENV_CONFIG', 'GOTRY_BENCHMARK_BRIDGE_PARENT_SECRET', 'LLM_API_KEY', 'LLM_BASE_URL', 'LLM_MODEL', 'DEEPSEEK_BASE_URL', 'GOTRY_LLM_MODEL', 'DATABASE_URL', 'SSH_AUTH_SOCK', 'AWS_PROFILE', 'HTTPS_PROXY']; const leaked = forbidden.filter(name => process.env[name] !== undefined); process.stdout.write(JSON.stringify({ schema_version: 'gotry_benchmark_tool_result_v1', status: 'ok', result: { marker: '${MARKER}', leaked } }))`;
+    const runnerBody = mode === 'domain-recovery-failed'
+        ? `if (args.city === 'Dubai') process.stdout.write(JSON.stringify({ schema_version: 'gotry_benchmark_tool_result_v1', status: 'miss', code: 'NOT_FOUND', recovery: 'revise_arguments' })); else { process.stderr.write('PRIVATE_RECOVERY_RUNNER_DIAGNOSTIC_DO_NOT_REFLECT'); process.exit(17) }`
+        : mode === 'domain-recovery'
+            ? `if (args.city === 'Dubai') process.stdout.write(JSON.stringify({ schema_version: 'gotry_benchmark_tool_result_v1', status: 'miss', code: 'NOT_FOUND', recovery: 'revise_arguments' })); else process.stdout.write(JSON.stringify({ schema_version: 'gotry_benchmark_tool_result_v1', status: 'ok', result: { marker: '${MARKER}', leaked: [] } }))`
+            : mode === 'timeout'
+                ? `setTimeout(() => {}, 60_000)`
+                : mode === 'runner-failed'
+                    ? `process.stderr.write('PRIVATE_RUNNER_DIAGNOSTIC_DO_NOT_REFLECT'); process.exit(17)`
+                    : mode === 'output-truncated'
+                        ? `process.stdout.write('x'.repeat(20_000))`
+                        : mode === 'unexpected-output'
+                            ? `process.stdout.write(JSON.stringify({ schema_version: 'gotry_benchmark_tool_result_v1', status: 'ok', result: { marker: '${MARKER}', leaked: [], unexpected: 'must-not-reflect' } }))`
+                            : `const forbidden = ['GOTRY_BENCHMARK_ENV_CONFIG', 'GOTRY_BENCHMARK_BRIDGE_PARENT_SECRET', 'LLM_API_KEY', 'LLM_BASE_URL', 'LLM_MODEL', 'DEEPSEEK_BASE_URL', 'GOTRY_LLM_MODEL', 'DATABASE_URL', 'SSH_AUTH_SOCK', 'AWS_PROFILE', 'HTTPS_PROXY']; const leaked = forbidden.filter(name => process.env[name] !== undefined); process.stdout.write(JSON.stringify({ schema_version: 'gotry_benchmark_tool_result_v1', status: 'ok', result: { marker: '${MARKER}', leaked } }))`;
     writeFileSync(runner, `if (process.argv.length !== 5 || process.argv[2] !== 'call' || process.argv[3] !== 'lookup' || (!${JSON.stringify(domainRecoveryMode)} && JSON.parse(process.argv[4]).city !== 'Dubai') || (${JSON.stringify(domainRecoveryMode)} && !['Dubai', 'Singapore'].includes(JSON.parse(process.argv[4]).city))) process.exit(2); const fs = require('node:fs'); const args = JSON.parse(process.argv[4]); fs.appendFileSync(${JSON.stringify(runnerTrace)}, JSON.stringify({ city: args.city }) + '\\n'); ${runnerBody}`);
-    writeFileSync(spawnTarget, '#!/usr/bin/env node\nprocess.exit(0)\n', {
-        mode: 0o700
-    });
-    writeFileSync(configPath, JSON.stringify({
-        schema_version: mode === 'invalid-schema' ? 'invalid' : 'gotry_benchmark_environment_bridge_v4',
-        enabled: true,
-        executable: mode === 'spawn-failed' ? spawnTarget : process.execPath,
-        cwd,
-        argv_prefix: mode === 'spawn-failed' ? [
-            'placeholder'
-        ] : [
-            runner
-        ],
-        tools: [
-            {
-                name: 'lookup',
-                description: 'Lookup.',
-                input_schema: LOOKUP_INPUT_SCHEMA,
-                output_keys: [
-                    'marker',
-                    'leaked'
-                ],
-                domain_outcomes: [
-                    {
-                        status: 'miss',
-                        code: 'NOT_FOUND',
-                        recovery: domainRecoveryMode ? 'revise_arguments' : 'none'
-                    }
-                ]
-            }
-        ],
-        timeout_ms: mode === 'timeout' ? 50 : 10_000,
-        max_output_bytes: mode === 'output-truncated' ? 1_024 : 4_096,
-        terminal_output: {
-            tag: 'benchmark_terminal',
-            max_bytes: 4_096,
-            body_schema: BRIDGE_E2E_BODY_SCHEMA
-        },
-        isolation: {
-            mode: 'host-enforced',
-            writes: 'forbidden',
-            network: 'denied'
-        }
-    }));
-    if (mode === 'unsafe-config') chmodSync(configPath, 0o666);
+    writeFileSync(spawnTarget, '#!/usr/bin/env node\nprocess.exit(0)\n', { mode: 0o700 });
+    writeFileSync(configPath, JSON.stringify({ schema_version: mode === 'invalid-schema' ? 'invalid' : 'gotry_benchmark_environment_bridge_v4', enabled: true, executable: mode === 'spawn-failed' ? spawnTarget : process.execPath, cwd, argv_prefix: mode === 'spawn-failed' ? ['placeholder'] : [runner], tools: [{ name: 'lookup', description: 'Lookup.', input_schema: LOOKUP_INPUT_SCHEMA, output_keys: ['marker', 'leaked'], domain_outcomes: [{ status: 'miss', code: 'NOT_FOUND', recovery: domainRecoveryMode ? 'revise_arguments' : 'none' }] }], timeout_ms: mode === 'timeout' ? 50 : 10_000, max_output_bytes: mode === 'output-truncated' ? 1_024 : 4_096, terminal_output: { tag: 'benchmark_terminal', max_bytes: 4_096, body_schema: BRIDGE_E2E_BODY_SCHEMA }, isolation: { mode: 'host-enforced', writes: 'forbidden', network: 'denied' } }));
+    if (mode === 'unsafe-config')
+        chmodSync(configPath, 0o666);
     const env = {
         ...process.env,
         DSH_TOOLS_MODE: 'both',
@@ -367,108 +210,53 @@ async function runCase(mode, executableOverride, extraEnv = {}) {
         LLM_MODEL: 'synthetic-bridge-model',
         DEEPSEEK_API_KEY: 'synthetic-bridge-key',
         DEEPSEEK_BASE_URL: `http://127.0.0.1:${port}/v1`,
-        ...mode !== 'disabled' ? {
+        ...(mode !== 'disabled' ? {
             DATABASE_URL: 'postgres://sentinel',
             SSH_AUTH_SOCK: '/tmp/sentinel.sock',
-            AWS_PROFILE: 'sentinel-profile'
-        } : {},
+            AWS_PROFILE: 'sentinel-profile',
+        } : {}),
         GOTRY_BENCHMARK_ENV_CONFIG: mode === 'disabled' ? '' : configPath,
-        ...mode !== 'disabled' ? {
-            GOTRY_BENCHMARK_BRIDGE_PARENT_SECRET: 'do-not-leak'
-        } : {},
-        ...mode === 'debug-redaction' ? {
-            GOTRY_DEBUG: '1'
-        } : {},
+        ...(mode !== 'disabled' ? { GOTRY_BENCHMARK_BRIDGE_PARENT_SECRET: 'do-not-leak' } : {}),
+        ...(mode === 'debug-redaction' ? { GOTRY_DEBUG: '1' } : {}),
         ...extraEnv,
-        NODE_OPTIONS: [
-            process.env.NODE_OPTIONS,
-            ...!executableOverride ? [
-                `--import=${TSX_LOADER}`
-            ] : [],
-            `--require=${probe}`
-        ].filter(Boolean).join(' ')
+        NODE_OPTIONS: [process.env.NODE_OPTIONS, ...(!executableOverride ? [`--import=${TSX_LOADER}`] : []), `--require=${probe}`].filter(Boolean).join(' '),
     };
-    for (const key of [
-        'GOTRY_LLM_MODEL',
-        'HTTP_PROXY',
-        'HTTPS_PROXY',
-        'ALL_PROXY',
-        'http_proxy',
-        'https_proxy',
-        'all_proxy'
-    ])delete env[key];
+    for (const key of ['GOTRY_LLM_MODEL', 'HTTP_PROXY', 'HTTPS_PROXY', 'ALL_PROXY', 'http_proxy', 'https_proxy', 'all_proxy'])
+        delete env[key];
     env.NO_PROXY = '127.0.0.1,localhost';
-    if (mode !== 'disabled') env.HTTPS_PROXY = 'https://sentinel-proxy';
-    if (mode === 'disabled') delete env.GOTRY_BENCHMARK_ENV_CONFIG;
+    if (mode !== 'disabled')
+        env.HTTPS_PROXY = 'https://sentinel-proxy';
+    if (mode === 'disabled')
+        delete env.GOTRY_BENCHMARK_ENV_CONFIG;
     const executable = executableOverride || process.execPath;
-    const invocation = mode === 'web-mode' ? [
-        'web',
-        '--no-open'
-    ] : [
-        mode === 'debug-redaction' ? 'PRIVATE_QUERY_SENTINEL_DO_NOT_REFLECT' : 'bridge smoke'
-    ];
-    const argv = executableOverride ? invocation : [
-        BIN,
-        ...invocation
-    ];
+    const invocation = mode === 'web-mode'
+        ? ['web', '--no-open']
+        : [mode === 'debug-redaction' ? 'PRIVATE_QUERY_SENTINEL_DO_NOT_REFLECT' : 'bridge smoke'];
+    const argv = executableOverride ? invocation : [BIN, ...invocation];
     try {
-        const child = spawn(executable, argv, {
-            cwd,
-            env,
-            stdio: [
-                'ignore',
-                'pipe',
-                'pipe'
-            ]
-        });
+        const child = spawn(executable, argv, { cwd, env, stdio: ['ignore', 'pipe', 'pipe'] });
         let stdout = '';
         let stderr = '';
-        child.stdout.on('data', (c)=>{
-            stdout += c;
-        });
-        child.stderr.on('data', (c)=>{
-            stderr += c;
-        });
-        const exit = await new Promise((resolve)=>{
-            const timer = setTimeout(()=>{
-                child.kill('SIGKILL');
-                resolve(null);
-            }, TIMEOUT_MS);
-            child.once('close', (code)=>{
-                clearTimeout(timer);
-                resolve(code);
-            });
-        });
-        const optionalResolutionHits = readFileSync(probeResult, 'utf8').split('\n').filter(Boolean).reduce((hits, line)=>{
+        child.stdout.on('data', c => { stdout += c; });
+        child.stderr.on('data', c => { stderr += c; });
+        const exit = await new Promise(resolve => { const timer = setTimeout(() => { child.kill('SIGKILL'); resolve(null); }, TIMEOUT_MS); child.once('close', code => { clearTimeout(timer); resolve(code); }); });
+        const optionalResolutionHits = readFileSync(probeResult, 'utf8').split('\n').filter(Boolean).reduce((hits, line) => {
             const event = JSON.parse(line);
-            if (event.target?.includes('dsh-calendar')) hits.calendar += 1;
-            if (event.target?.includes('dsh-map-tools')) hits.map += 1;
+            if (event.target?.includes('dsh-calendar'))
+                hits.calendar += 1;
+            if (event.target?.includes('dsh-map-tools'))
+                hits.map += 1;
             return hits;
-        }, {
-            calendar: 0,
-            map: 0
-        });
-        const runnerArguments = existsSync(runnerTrace) ? readFileSync(runnerTrace, 'utf8').split('\n').filter(Boolean).map((line)=>JSON.parse(line)) : [];
-        return {
-            exit,
-            stdout,
-            stderr,
-            output: stdout + stderr,
-            requests,
-            optionalResolutionHits,
-            servedToolCalls,
-            runnerArguments
-        };
-    } finally{
-        await new Promise((resolve)=>server.close(()=>resolve()));
-        rmSync(dsh, {
-            recursive: true,
-            force: true
-        });
-        rmSync(cwd, {
-            recursive: true,
-            force: true
-        });
+        }, { calendar: 0, map: 0 });
+        const runnerArguments = existsSync(runnerTrace)
+            ? readFileSync(runnerTrace, 'utf8').split('\n').filter(Boolean).map(line => JSON.parse(line))
+            : [];
+        return { exit, stdout, stderr, output: stdout + stderr, requests, optionalResolutionHits, servedToolCalls, runnerArguments };
+    }
+    finally {
+        await new Promise(resolve => server.close(() => resolve()));
+        rmSync(dsh, { recursive: true, force: true });
+        rmSync(cwd, { recursive: true, force: true });
     }
 }
 async function assertRuntimeContract(executableOverride) {
@@ -479,93 +267,80 @@ async function assertRuntimeContract(executableOverride) {
     assert.equal(binSource.includes('headless-startup-hold'), false, `${target} bin must not inject a timer-based startup hold`);
     const disabled = await runCase('disabled', executableOverride);
     assert.equal(disabled.exit, 0, `${target} default-off child exit=${disabled.exit}; output tail=${disabled.output.slice(-2_000)}`);
-    assert.ok(disabled.requests.length > 0 && !disabled.requests.some((r)=>names(r).includes(TOOL)), `${target} default-off must reach the relay without exposing benchmark tool; exit=${disabled.exit}; requests=${disabled.requests.length}; output=${disabled.output.slice(-2_000)}`);
+    assert.ok(disabled.requests.length > 0 && !disabled.requests.some(r => names(r).includes(TOOL)), `${target} default-off must reach the relay without exposing benchmark tool; exit=${disabled.exit}; requests=${disabled.requests.length}; output=${disabled.output.slice(-2_000)}`);
     assert.ok(disabled.optionalResolutionHits.calendar > 0 || disabled.optionalResolutionHits.map > 0, `${target} default-off must retain optional plugin resolution as a counter-proof`);
+    const ordinaryPlannerRequests = disabled.requests.filter(request => names(request).some(name => name.startsWith('gotry_')));
+    assert.ok(ordinaryPlannerRequests.length > 0, `${target} default-off must surface at least one ordinary planner request exposing a gotry_* tool (auxiliary title requests are excluded); requests=${disabled.requests.length}; tool surfaces=${JSON.stringify(disabled.requests.map(names))}; output=${disabled.output.slice(-2_000)}`);
+    const ORDINARY_IDENTITY_SENTENCE = '你是 GoTry——从出发到下一次出发的 AI 旅行伙伴';
+    const CANONICAL_RAW_VARIABLES = ['{{current_date}}', '{{time_anchor_card}}', '{{motivation_brief}}', '{{channel_routing_card}}'];
+    const EXPANDED_DATE_PATTERN = /今天是\s*\d{4}-\d{2}-\d{2}/;
+    for (const request of ordinaryPlannerRequests) {
+        const prompt = JSON.stringify(request);
+        assert.ok(prompt.includes(ORDINARY_IDENTITY_SENTENCE), `${target} ordinary planner request must carry the canonical Chinese GoTry identity sentence; prompt head=${prompt.slice(0, 600)}`);
+        for (const variable of CANONICAL_RAW_VARIABLES) {
+            assert.equal(prompt.includes(variable), false, `${target} ordinary planner request must not retain raw ${variable}; prompt head=${prompt.slice(0, 600)}`);
+        }
+        assert.match(prompt, EXPANDED_DATE_PATTERN, `${target} ordinary planner request must show the expanded "今天是 YYYY-MM-DD" shape from the canonical {{current_date}} variable; prompt head=${prompt.slice(0, 600)}`);
+    }
     const enabled = await runCase('enabled', executableOverride);
     assert.equal(enabled.exit, 0, `${target} opt-in child exit=${enabled.exit}; requests=${enabled.requests.length}; tool surfaces=${JSON.stringify(enabled.requests.map(names))}; output tail=${enabled.output.slice(-10000)}`);
-    assert.ok(enabled.requests.some((r)=>names(r).includes(TOOL)), `${target} opt-in planner request must expose benchmark tool; schemas=${JSON.stringify(enabled.requests.map(names))}; output=${enabled.output.slice(-4_000)}`);
-    const enabledToolNames = [
-        ...new Set(enabled.requests.flatMap(names))
-    ].sort();
-    assert.deepEqual(enabledToolNames, [
-        TOOL
-    ], `${target} enabled runtime must expose exactly the benchmark tool; observed tool names=${JSON.stringify(enabledToolNames)}`);
-    const bridgeTool = enabled.requests.find((request)=>names(request).includes(TOOL))?.tools?.find((tool)=>names({
-            tools: [
-                tool
-            ]
-        }).includes(TOOL));
+    assert.ok(enabled.requests.some(r => names(r).includes(TOOL)), `${target} opt-in planner request must expose benchmark tool; schemas=${JSON.stringify(enabled.requests.map(names))}; output=${enabled.output.slice(-4_000)}`);
+    const enabledToolNames = [...new Set(enabled.requests.flatMap(names))].sort();
+    assert.deepEqual(enabledToolNames, [TOOL], `${target} enabled runtime must expose exactly the benchmark tool; observed tool names=${JSON.stringify(enabledToolNames)}`);
+    const bridgeTool = enabled.requests.find(request => names(request).includes(TOOL))?.tools?.find(tool => names({ tools: [tool] }).includes(TOOL));
     const flatSchema = bridgeTool?.function?.parameters;
     assert.equal(flatSchema?.type, 'object', `${target} bridge exposes an object-root wire schema`);
     assert.equal(flatSchema?.oneOf, undefined, `${target} bridge wire has no top-level oneOf`);
-    assert.deepEqual(flatSchema?.required, [
-        'action'
-    ]);
+    assert.deepEqual(flatSchema?.required, ['action']);
     assert.equal(flatSchema?.additionalProperties, false);
-    assert.deepEqual(flatSchema?.properties?.action?.enum, [
-        'tools',
-        'call',
-        'errors'
-    ]);
-    assert.deepEqual(flatSchema?.properties?.tool?.enum, [
-        'lookup'
-    ]);
-    assert.deepEqual(flatSchema?.properties?.arguments, {
-        type: 'object',
-        additionalProperties: true,
-        description: 'action=call 时传给工具的参数对象'
-    });
-    assert.equal(enabledToolNames.some((name)=>name.startsWith('calendar_') || name.startsWith('map_')), false, `${target} benchmark projection must not expose calendar/map tools`);
-    assert.deepEqual(enabled.optionalResolutionHits, {
-        calendar: 0,
-        map: 0
-    }, `${target} benchmark mode must not resolve optional calendar/map plugins`);
+    assert.deepEqual(flatSchema?.properties?.action?.enum, ['tools', 'call', 'errors']);
+    assert.deepEqual(flatSchema?.properties?.tool?.enum, ['lookup']);
+    assert.deepEqual(flatSchema?.properties?.arguments, { type: 'object', additionalProperties: true, description: 'action=call 时传给工具的参数对象' });
+    assert.equal(enabledToolNames.some(name => name.startsWith('calendar_') || name.startsWith('map_')), false, `${target} benchmark projection must not expose calendar/map tools`);
+    assert.deepEqual(enabled.optionalResolutionHits, { calendar: 0, map: 0 }, `${target} benchmark mode must not resolve optional calendar/map plugins`);
     assert.ok(enabled.requests.some(toolResultPresent), `${target} marker must enter model history as tool result`);
-    const leakedReport = enabled.requests.map((r)=>JSON.stringify(r).match(/\\?"leaked\\?":\[(.*?)\]/)?.[1]).filter(Boolean).join('|');
-    assert.ok(enabled.requests.some((r)=>/\\?"leaked\\?":\[\]/.test(JSON.stringify(r))), `${target} tool result must report no forbidden environment names; observed names=${leakedReport || '(none)'}`);
-    assert.equal(enabled.requests.some((r)=>JSON.stringify(r).includes('do-not-leak')), false, `${target} tool result must not expose the parent secret value`);
-    const enabledPrompt = enabled.requests.map((r)=>JSON.stringify(r)).join('\n');
-    for (const variable of [
-        '{{current_date}}',
-        '{{time_anchor_card}}',
-        '{{motivation_brief}}'
-    ]){
+    const leakedReport = enabled.requests.map(r => JSON.stringify(r).match(/\\?"leaked\\?":\[(.*?)\]/)?.[1]).filter(Boolean).join('|');
+    assert.ok(enabled.requests.some(r => /\\?"leaked\\?":\[\]/.test(JSON.stringify(r))), `${target} tool result must report no forbidden environment names; observed names=${leakedReport || '(none)'}`);
+    assert.equal(enabled.requests.some(r => JSON.stringify(r).includes('do-not-leak')), false, `${target} tool result must not expose the parent secret value`);
+    const enabledPrompt = enabled.requests.map(r => JSON.stringify(r)).join('\n');
+    for (const variable of ['{{current_date}}', '{{time_anchor_card}}', '{{motivation_brief}}']) {
         assert.equal(enabledPrompt.includes(variable), false, `${target} benchmark persona must not retain ${variable}`);
     }
     assert.equal(enabledPrompt.includes('gotry_feasibility_check'), false, `${target} benchmark persona must not retain ordinary GoTry tool instructions`);
-    assert.ok(enabled.requests.length > 0 && enabled.requests.some((request)=>{
+    // Persona invariant: every request that exposes the benchmark tool (a
+    // planner request) must carry each stable sentence exactly once. The
+    // separate session-title request does NOT expose the benchmark tool and is
+    // intentionally not required to carry the persona — dsh-system-prompt
+    // 0.1.5-rc.1 emits a distinct system prompt for it (auto-title). This
+    // is the precise observed public protocol invariant under target closure,
+    // not a relaxed `some` over the full request stream.
+    const SENTENCE_A = 'You are GoTry, a task-agnostic travel planning assistant.';
+    const SENTENCE_B = 'Use only the current conversation and tools available in this benchmark session.';
+    const plannerRequests = enabled.requests.filter(request => names(request).includes(TOOL));
+    assert.ok(plannerRequests.length > 0, `${target} must emit at least one planner request exposing the benchmark tool; requests=${enabled.requests.length}; schemas=${JSON.stringify(enabled.requests.map(names))}`);
+    for (const request of plannerRequests) {
         const prompt = JSON.stringify(request);
-        return (prompt.match(/You are GoTry, a task-agnostic travel planning assistant\./g) ?? []).length === 1 && (prompt.match(/Use only the current conversation and tools available in this benchmark session\./g) ?? []).length === 1;
-    }), `${target} benchmark persona has each stable sentence exactly once per request`);
+        const aCount = (prompt.match(new RegExp(SENTENCE_A.replace(/[.*+?^${}()|[\]\\]/g, '\\$&'), 'g')) ?? []).length;
+        const bCount = (prompt.match(new RegExp(SENTENCE_B.replace(/[.*+?^${}()|[\]\\]/g, '\\$&'), 'g')) ?? []).length;
+        assert.equal(aCount, 1, `${target} planner request must carry stable sentence A exactly once (observed=${aCount}); prompt head=${prompt.slice(0, 400)}`);
+        assert.equal(bCount, 1, `${target} planner request must carry stable sentence B exactly once (observed=${bCount}); prompt head=${prompt.slice(0, 400)}`);
+    }
     assert.match(enabled.output, /benchmark_terminal/);
     const recovered = await runCase('domain-recovery', executableOverride);
     assert.equal(recovered.exit, 0, `${target} model-driven domain miss recovery exits successfully; output=${recovered.output.slice(-2_000)}`);
     assert.equal(recovered.servedToolCalls, 2, `${target} model emits exactly two tool calls around one declared miss`);
-    assert.deepEqual(recovered.runnerArguments, [
-        {
-            city: 'Dubai'
-        },
-        {
-            city: 'Singapore'
-        }
-    ], `${target} model revises the declared city before the second adapter invocation`);
-    assert.ok(recovered.requests.some((request)=>(request.messages ?? []).some((message)=>{
-            if (message.role !== 'tool') return false;
-            const serialized = JSON.stringify(message);
-            return /status\\?":\\?"miss/.test(serialized) && /recovery\\?":\\?"revise_arguments/.test(serialized);
-        })), `${target} declared typed miss reaches model history`);
+    assert.deepEqual(recovered.runnerArguments, [{ city: 'Dubai' }, { city: 'Singapore' }], `${target} model revises the declared city before the second adapter invocation`);
+    assert.ok(recovered.requests.some(request => (request.messages ?? []).some(message => {
+        if (message.role !== 'tool')
+            return false;
+        const serialized = JSON.stringify(message);
+        return /status\\?":\\?"miss/.test(serialized) && /recovery\\?":\\?"revise_arguments/.test(serialized);
+    })), `${target} declared typed miss reaches model history`);
     assert.match(recovered.stdout, /<benchmark_terminal>/, `${target} corrected second call reaches tagged terminal output`);
     const failedRecovery = await runCase('domain-recovery-failed', executableOverride);
     assert.equal(failedRecovery.exit, 1, `${target} infrastructure failure after a declared miss cannot be masked by the earlier domain outcome`);
     assert.equal(failedRecovery.servedToolCalls, 2, `${target} failed recovery still exercises exactly two model-owned tool calls`);
-    assert.deepEqual(failedRecovery.runnerArguments, [
-        {
-            city: 'Dubai'
-        },
-        {
-            city: 'Singapore'
-        }
-    ], `${target} failed recovery reaches the revised second adapter invocation`);
+    assert.deepEqual(failedRecovery.runnerArguments, [{ city: 'Dubai' }, { city: 'Singapore' }], `${target} failed recovery reaches the revised second adapter invocation`);
     assert.equal(failedRecovery.stdout, '', `${target} failed recovery releases no terminal stdout`);
     assert.match(failedRecovery.stderr, /benchmark terminal output unavailable \(child_bridge_runner_failed\)/, `${target} failed recovery preserves the second runner failure classification`);
     assert.equal(failedRecovery.output.includes('PRIVATE_RECOVERY_RUNNER_DIAGNOSTIC_DO_NOT_REFLECT'), false, `${target} failed recovery never reflects private runner stderr`);
@@ -575,8 +350,8 @@ async function assertRuntimeContract(executableOverride) {
     const unexpected = await runCase('unexpected-output', executableOverride);
     assert.equal(unexpected.exit, 1, `${target} unexpected output must fail the successful-call conformance gate; exit=${unexpected.exit}; output tail=${unexpected.output.slice(-1000)}`);
     assert.equal(unexpected.stdout, '', `${target} rejected runner output keeps stdout empty`);
-    assert.ok(unexpected.requests.some((r)=>JSON.stringify(r).includes('forbidden_output')), `${target} positive output allowlist must reject unexpected output`);
-    assert.equal(unexpected.requests.some((r)=>JSON.stringify(r).includes('must-not-reflect')), false, `${target} positive output rejection must not reflect unexpected key/value`);
+    assert.ok(unexpected.requests.some(r => JSON.stringify(r).includes('forbidden_output')), `${target} positive output allowlist must reject unexpected output`);
+    assert.equal(unexpected.requests.some(r => JSON.stringify(r).includes('must-not-reflect')), false, `${target} positive output rejection must not reflect unexpected key/value`);
     assert.equal(unexpected.output.includes('<benchmark_terminal>'), false, `${target} failed bridge result cannot release a terminal body`);
     const invalidPath = await runCase('invalid-path', executableOverride);
     assert.equal(invalidPath.exit, 1, `${target} invalid config basename child exit=${invalidPath.exit}`);
@@ -598,23 +373,23 @@ async function assertRuntimeContract(executableOverride) {
     assert.match(webMode.output, /benchmark environment requires headless mode/);
     const truncated = await runCase('output-truncated', executableOverride);
     assert.equal(truncated.exit, 1, `${target} output truncation fails the successful-call conformance gate`);
-    assert.ok(truncated.requests.some((r)=>JSON.stringify(r).includes('output_truncated')), `${target} real runner output over the configured cap is rejected`);
+    assert.ok(truncated.requests.some(r => JSON.stringify(r).includes('output_truncated')), `${target} real runner output over the configured cap is rejected`);
     assert.equal(truncated.stdout, '', `${target} output truncation keeps stdout empty`);
     assert.match(truncated.stderr, /benchmark terminal output unavailable \(child_bridge_output_truncated\)/, `${target} output truncation emits a stable bridge reason code`);
     const timedOut = await runCase('timeout', executableOverride);
     assert.equal(timedOut.exit, 1, `${target} timeout fails the successful-call conformance gate`);
-    assert.ok(timedOut.requests.some((r)=>JSON.stringify(r).includes('timed_out')), `${target} real runner deadline is enforced`);
+    assert.ok(timedOut.requests.some(r => JSON.stringify(r).includes('timed_out')), `${target} real runner deadline is enforced`);
     assert.equal(timedOut.stdout, '', `${target} timeout keeps stdout empty`);
     assert.match(timedOut.stderr, /benchmark terminal output unavailable \(child_bridge_timed_out\)/, `${target} timeout emits a stable child bridge reason code`);
     const runnerFailed = await runCase('runner-failed', executableOverride);
     assert.equal(runnerFailed.exit, 1, `${target} runner failure fails the successful-call conformance gate`);
-    assert.ok(runnerFailed.requests.some((r)=>JSON.stringify(r).includes('runner_failed')), `${target} nonzero runner exit is surfaced structurally`);
+    assert.ok(runnerFailed.requests.some(r => JSON.stringify(r).includes('runner_failed')), `${target} nonzero runner exit is surfaced structurally`);
     assert.equal(runnerFailed.stdout, '', `${target} runner failure keeps stdout empty`);
     assert.match(runnerFailed.stderr, /benchmark terminal output unavailable \(child_bridge_runner_failed\)/, `${target} runner failure emits a stable child bridge reason code`);
     assert.equal(runnerFailed.output.includes('PRIVATE_RUNNER_DIAGNOSTIC_DO_NOT_REFLECT'), false, `${target} runner stderr is never reflected`);
     const spawnFailed = await runCase('spawn-failed', executableOverride);
     assert.equal(spawnFailed.exit, 1, `${target} runner spawn failure fails the successful-call conformance gate`);
-    assert.ok(spawnFailed.requests.some((r)=>JSON.stringify(r).includes('spawn_failed')), `${target} runner spawn failure is surfaced structurally`);
+    assert.ok(spawnFailed.requests.some(r => JSON.stringify(r).includes('spawn_failed')), `${target} runner spawn failure is surfaced structurally`);
     assert.equal(spawnFailed.stdout, '', `${target} runner spawn failure keeps stdout empty`);
     assert.match(spawnFailed.stderr, /benchmark terminal output unavailable \(child_bridge_spawn_failed\)/, `${target} runner spawn failure emits a stable child bridge reason code`);
 }
@@ -625,7 +400,8 @@ async function assertSourceRuntimeContractWhenAvailable() {
         try {
             createRequire(BIN).resolve('@deepseek-ai/dsh/lib/bin.js');
             sourceRuntimeAvailable = true;
-        } catch  {
+        }
+        catch {
             sourceRuntimeAvailable = false;
         }
     }
@@ -655,42 +431,34 @@ async function assertPackagedPatchProjection(executable) {
     const reorderedPoisonModule = join(probePackageRoot, 'future-reordered-plugin.cjs');
     const inlinePoisonProof = join(probeParent, 'future-inline-loaded.txt');
     const reorderedPoisonProof = join(probeParent, 'future-reordered-loaded.txt');
-    cpSync(sourcePackageRoot, probePackageRoot, {
-        recursive: true
-    });
+    cpSync(sourcePackageRoot, probePackageRoot, { recursive: true });
     chmodSync(probeExecutable, 0o755);
     const basePatch = readFileSync(patchPath, 'utf8');
     const stableError = /benchmark environment configuration unavailable/;
     const poisonEnv = {
         GOTRY_FUTURE_INLINE_PROOF: inlinePoisonProof,
-        GOTRY_FUTURE_REORDERED_PROOF: reorderedPoisonProof
+        GOTRY_FUTURE_REORDERED_PROOF: reorderedPoisonProof,
     };
-    const runRejectedPatch = async (label, patch, forbiddenValues = [])=>{
-        rmSync(inlinePoisonProof, {
-            force: true
-        });
-        rmSync(reorderedPoisonProof, {
-            force: true
-        });
+    const runRejectedPatch = async (label, patch, forbiddenValues = []) => {
+        rmSync(inlinePoisonProof, { force: true });
+        rmSync(reorderedPoisonProof, { force: true });
         writeFileSync(patchPath, patch);
         const result = await runCase('enabled', probeExecutable, poisonEnv);
         assert.equal(result.exit, 1, `${label} must reject the benchmark startup`);
         assert.equal(result.requests.length, 0, `${label} must fail before relay activity`);
-        assert.deepEqual(result.optionalResolutionHits, {
-            calendar: 0,
-            map: 0
-        }, `${label} must fail before optional plugin resolution`);
+        assert.deepEqual(result.optionalResolutionHits, { calendar: 0, map: 0 }, `${label} must fail before optional plugin resolution`);
         assert.match(result.output, stableError, `${label} emits a stable generic error`);
         assert.equal(result.output.includes(probePackageRoot), false, `${label} must not reflect the package path`);
         assert.equal(existsSync(inlinePoisonProof) || existsSync(reorderedPoisonProof), false, `${label} must not execute a poison plugin`);
-        for (const value of forbiddenValues)assert.equal(result.output.includes(value), false, `${label} must not reflect rejected input`);
+        for (const value of forbiddenValues)
+            assert.equal(result.output.includes(value), false, `${label} must not reflect rejected input`);
     };
     try {
         writeFileSync(inlinePoisonModule, `const fs = require('node:fs'); fs.appendFileSync(process.env.GOTRY_FUTURE_INLINE_PROOF, 'loaded\\n'); exports.name = 'round4-future-inline'; exports.apply = () => {}`);
         writeFileSync(reorderedPoisonModule, `const fs = require('node:fs'); fs.appendFileSync(process.env.GOTRY_FUTURE_REORDERED_PROOF, 'loaded\\n'); exports.name = 'round4-future-reordered'; exports.apply = () => {}`);
         const futureEntries = [
             `    - { id: dsh-future-inline, name: '${inlinePoisonModule}' }`,
-            `    - name: '${reorderedPoisonModule}'\n      id: dsh-future-reordered`
+            `    - name: '${reorderedPoisonModule}'\n      id: dsh-future-reordered`,
         ].join('\n');
         const futurePatch = basePatch.replace("    - id: dsh-map-tools", `${futureEntries}\n    - id: dsh-map-tools`);
         assert.notEqual(futurePatch, basePatch, 'future-plugin fixture must enter the insert sequence');
@@ -698,65 +466,42 @@ async function assertPackagedPatchProjection(executable) {
         const ordinary = await runCase('disabled', probeExecutable, poisonEnv);
         assert.ok(existsSync(inlinePoisonProof), `default-off must execute the inline future-plugin top level; exit=${ordinary.exit}; output=${ordinary.output.slice(-2_000)}`);
         assert.ok(existsSync(reorderedPoisonProof), `default-off must execute the reordered future-plugin top level; exit=${ordinary.exit}; output=${ordinary.output.slice(-2_000)}`);
-        rmSync(inlinePoisonProof, {
-            force: true
-        });
-        rmSync(reorderedPoisonProof, {
-            force: true
-        });
+        rmSync(inlinePoisonProof, { force: true });
+        rmSync(reorderedPoisonProof, { force: true });
         const benchmark = await runCase('enabled', probeExecutable, poisonEnv);
         assert.equal(benchmark.exit, 0, `benchmark future-plugin projection exits 0; output=${benchmark.output.slice(-2_000)}`);
-        assert.ok(benchmark.requests.some((request)=>names(request).includes(TOOL)), 'benchmark future-plugin projection reaches the bridge relay');
+        assert.ok(benchmark.requests.some(request => names(request).includes(TOOL)), 'benchmark future-plugin projection reaches the bridge relay');
         assert.equal(existsSync(inlinePoisonProof) || existsSync(reorderedPoisonProof), false, 'benchmark projection must not execute inline or reordered future plugins');
-        assert.deepEqual(benchmark.optionalResolutionHits, {
-            calendar: 0,
-            map: 0
-        }, 'benchmark future-plugin projection does not resolve optional host plugins');
+        assert.deepEqual(benchmark.optionalResolutionHits, { calendar: 0, map: 0 }, 'benchmark future-plugin projection does not resolve optional host plugins');
         await runRejectedPatch('missing gotry-tools', basePatch.replace('    - id: gotry-tools', '    - id: gotry-tools-missing'));
         await runRejectedPatch('duplicate gotry-tools', basePatch.replace('    - id: dsh-map-tools', "    - id: gotry-tools\n      name: 'duplicate/gotry-tools'\n    - id: dsh-map-tools"));
-        await runRejectedPatch('second insert block', `${basePatch}\n- insert:\n    - id: dsh-second-insert\n      name: '${inlinePoisonModule}'\n`, [
-            inlinePoisonModule
-        ]);
-        await runRejectedPatch('flow second insert block', `${basePatch}\n- insert: [{ id: dsh-flow-second-insert, name: '${inlinePoisonModule}' }]\n`, [
-            inlinePoisonModule
-        ]);
-        await runRejectedPatch('spoofed gotry-tools name', basePatch.replace("name: 'placeholder/ts/src/index.ts'", `name: '${inlinePoisonModule}'`), [
-            inlinePoisonModule
-        ]);
-        await runRejectedPatch('spoofed gotry-tools name with decoy anchor', `${basePatch.replace("name: 'placeholder/ts/src/index.ts'", `name: '${inlinePoisonModule}'`)}\n- id: benchmark-name-decoy\n  name: 'placeholder/ts/src/index.ts'\n`, [
-            inlinePoisonModule
-        ]);
-        await runRejectedPatch('spoofed gotry-tools name with nested decoy anchor', basePatch.replace("name: 'placeholder/ts/src/index.ts'", `name: '${inlinePoisonModule}'`).replace("        stateRoot: '.'", "        name: 'placeholder/ts/src/index.ts'\n        stateRoot: '.'"), [
-            inlinePoisonModule
-        ]);
+        await runRejectedPatch('second insert block', `${basePatch}\n- insert:\n    - id: dsh-second-insert\n      name: '${inlinePoisonModule}'\n`, [inlinePoisonModule]);
+        await runRejectedPatch('flow second insert block', `${basePatch}\n- insert: [{ id: dsh-flow-second-insert, name: '${inlinePoisonModule}' }]\n`, [inlinePoisonModule]);
+        await runRejectedPatch('spoofed gotry-tools name', basePatch.replace("name: 'placeholder/ts/src/index.ts'", `name: '${inlinePoisonModule}'`), [inlinePoisonModule]);
+        await runRejectedPatch('spoofed gotry-tools name with decoy anchor', `${basePatch.replace("name: 'placeholder/ts/src/index.ts'", `name: '${inlinePoisonModule}'`)}\n- id: benchmark-name-decoy\n  name: 'placeholder/ts/src/index.ts'\n`, [inlinePoisonModule]);
+        await runRejectedPatch('spoofed gotry-tools name with nested decoy anchor', basePatch
+            .replace("name: 'placeholder/ts/src/index.ts'", `name: '${inlinePoisonModule}'`)
+            .replace("        stateRoot: '.'", "        name: 'placeholder/ts/src/index.ts'\n        stateRoot: '.'"), [inlinePoisonModule]);
         await runRejectedPatch('missing benchmark config anchor', basePatch.replace(/^\s*hbcliBin:.*\n/m, ''));
         await runRejectedPatch('missing benchmark config anchor with decoy', `${basePatch.replace(/^\s*hbcliBin:.*\n/m, '')}\n- id: benchmark-anchor-decoy\n  hbcliBin: 'hbcli'\n`);
-        await runRejectedPatch('missing benchmark config anchor with nested decoy', basePatch.replace(/^\s*hbcliBin:.*\n/m, '').replace("        stateRoot: '.'", "        nestedAnchorDecoy:\n          hbcliBin: 'hbcli'\n        stateRoot: '.'"));
+        await runRejectedPatch('missing benchmark config anchor with nested decoy', basePatch
+            .replace(/^\s*hbcliBin:.*\n/m, '')
+            .replace("        stateRoot: '.'", "        nestedAnchorDecoy:\n          hbcliBin: 'hbcli'\n        stateRoot: '.'"));
         await runRejectedPatch('duplicate benchmark config anchor', basePatch.replace("        hbcliBin: 'hbcli'", "        hbcliBin: 'hbcli'\n        hbcliBin: 'hbcli'"));
-        await runRejectedPatch('pre-existing benchmark config path', basePatch.replace("        hbcliBin: 'hbcli'", "        hbcliBin: 'hbcli'\n        benchmarkEnvironmentConfigPath: '/not/used'"), [
-            '/not/used'
-        ]);
+        await runRejectedPatch('pre-existing benchmark config path', basePatch.replace("        hbcliBin: 'hbcli'", "        hbcliBin: 'hbcli'\n        benchmarkEnvironmentConfigPath: '/not/used'"), ['/not/used']);
         await runRejectedPatch('missing system-prompt anchor', basePatch.replace(/^- id: system-prompt[\s\S]*$/m, ''));
-        await runRejectedPatch('duplicate system-prompt anchor', `${basePatch}\n- id: system-prompt\n  config:\n    persona: >-\n      duplicate\n`);
-        await runRejectedPatch('quoted system-prompt duplicate', `${basePatch}\n- id: 'system-prompt'\n  config:\n    persona: >-\n      quoted duplicate\n`);
+        await runRejectedPatch('duplicate system-prompt anchor', `${basePatch}\n- id: system-prompt\n  config:\n    personaPrefix: >-\n      duplicate\n`);
+        await runRejectedPatch('quoted system-prompt duplicate', `${basePatch}\n- id: 'system-prompt'\n  config:\n    personaPrefix: >-\n      quoted duplicate\n`);
         const systemPromptMutationSentinel = 'ROUND7_SYSTEM_PROMPT_MUTATION_SENTINEL_DO_NOT_REFLECT';
-        await runRejectedPatch('quoted mapping-key system-prompt duplicate', `${basePatch}\n- "id": system-prompt\n  config:\n    persona: >-\n      ${systemPromptMutationSentinel}\n`, [
-            systemPromptMutationSentinel
-        ]);
-        await runRejectedPatch('flow quoted-key system-prompt duplicate', `${basePatch}\n- { "id": system-prompt, config: { persona: ${systemPromptMutationSentinel} } }\n`, [
-            systemPromptMutationSentinel
-        ]);
-        await runRejectedPatch('reordered system-prompt duplicate', `${basePatch}\n- name: reordered-system-prompt\n  id: system-prompt\n  config:\n    persona: >-\n      reordered duplicate\n`);
-        await runRejectedPatch('flow system-prompt duplicate', `${basePatch}\n- { id: system-prompt, config: { persona: flow duplicate } }\n`);
-        await runRejectedPatch('noncanonical insert id root item', `${basePatch}\n- id: insert\n  config:\n    persona: >-\n      ${systemPromptMutationSentinel}\n`, [
-            systemPromptMutationSentinel
-        ]);
-        await runRejectedPatch('malformed system-prompt persona', basePatch.replace(/^    persona: >-$/m, '    persona: plain'));
-    } finally{
-        rmSync(probeParent, {
-            recursive: true,
-            force: true
-        });
+        await runRejectedPatch('quoted mapping-key system-prompt duplicate', `${basePatch}\n- "id": system-prompt\n  config:\n    personaPrefix: >-\n      ${systemPromptMutationSentinel}\n`, [systemPromptMutationSentinel]);
+        await runRejectedPatch('flow quoted-key system-prompt duplicate', `${basePatch}\n- { "id": system-prompt, config: { personaPrefix: ${systemPromptMutationSentinel} } }\n`, [systemPromptMutationSentinel]);
+        await runRejectedPatch('reordered system-prompt duplicate', `${basePatch}\n- name: reordered-system-prompt\n  id: system-prompt\n  config:\n    personaPrefix: >-\n      reordered duplicate\n`);
+        await runRejectedPatch('flow system-prompt duplicate', `${basePatch}\n- { id: system-prompt, config: { personaPrefix: flow duplicate } }\n`);
+        await runRejectedPatch('noncanonical insert id root item', `${basePatch}\n- id: insert\n  config:\n    personaPrefix: >-\n      ${systemPromptMutationSentinel}\n`, [systemPromptMutationSentinel]);
+        await runRejectedPatch('malformed system-prompt persona', basePatch.replace(/^    personaPrefix: >-$/m, '    personaPrefix: plain'));
+    }
+    finally {
+        rmSync(probeParent, { recursive: true, force: true });
     }
 }
 const LARGE_TERMINAL_PAYLOAD = 'x'.repeat(80 * 1024);
@@ -764,23 +509,21 @@ function taggedTerminal(valid) {
     return `<benchmark_terminal>${valid ? '{"status":"succeeded"}' : '{"status":'}</benchmark_terminal>`;
 }
 function conformanceResponse(mode, request, plannerCount) {
-    const hasToolResult = (request.messages ?? []).some((message)=>message.role === 'tool');
-    if (mode === 'a' && plannerCount === 1 && !hasToolResult) return finalText('assistant prose without a call');
-    const call = mode === 'a' && plannerCount === 2 || [
-        'b',
-        'd',
-        'f',
-        'large',
-        'schema'
-    ].includes(mode) && plannerCount === 1;
-    if (call && !hasToolResult) return toolCall();
-    if (mode === 'f' && plannerCount === 3) return toolCall('bridge-call-retry');
+    const hasToolResult = (request.messages ?? []).some(message => message.role === 'tool');
+    if (mode === 'a' && plannerCount === 1 && !hasToolResult)
+        return finalText('assistant prose without a call');
+    const call = mode === 'a' && plannerCount === 2
+        || ['b', 'd', 'f', 'large', 'schema'].includes(mode) && plannerCount === 1;
+    if (call && !hasToolResult)
+        return toolCall();
+    if (mode === 'f' && plannerCount === 3)
+        return toolCall('bridge-call-retry');
     if (mode === 'large' && hasToolResult) {
-        return finalText(`<benchmark_terminal>${JSON.stringify({
-            payload: LARGE_TERMINAL_PAYLOAD
-        })}</benchmark_terminal>`);
+        return finalText(`<benchmark_terminal>${JSON.stringify({ payload: LARGE_TERMINAL_PAYLOAD })}</benchmark_terminal>`);
     }
     if (mode === 'schema') {
+        // Round 11 真实失败形态:结构合法信封内多出 root 键(extra budget)——
+        // 信封解析通过,exact body schema 必须拒绝,且纠正后仍不 autofix。
         return finalText('<benchmark_terminal>{"status":"succeeded","budget":{"total_cost":1}}</benchmark_terminal>');
     }
     const valid = mode === 'a' ? hasToolResult : mode === 'b' ? plannerCount >= 3 : mode === 'e' ? true : false;
@@ -791,51 +534,40 @@ async function runConformanceCase(mode, executableOverride) {
     let plannerCount = 0;
     let recoveredAttempts = 0;
     let servedToolCalls = 0;
-    const server = createServer((req, res)=>{
+    const server = createServer((req, res) => {
         const chunks = [];
-        req.on('data', (chunk)=>chunks.push(Buffer.from(chunk)));
-        req.on('end', ()=>{
+        req.on('data', chunk => chunks.push(Buffer.from(chunk)));
+        req.on('end', () => {
             let body = {};
             try {
                 body = JSON.parse(Buffer.concat(chunks).toString());
-            } catch  {}
+            }
+            catch { /* structural failure */ }
             requests.push(body);
             const requestHasToolResult = anyToolResultPresent(body);
             const plannerRequest = names(body).includes(TOOL);
-            if (mode === 'exhausted' || mode === 'unknown' || mode === 'post-failure' && requestHasToolResult) {
-                res.writeHead(mode === 'exhausted' ? 429 : mode === 'unknown' ? 418 : 500, {
-                    'content-type': 'application/json'
-                });
-                res.end(JSON.stringify({
-                    error: {
-                        message: 'PRIVATE_SENTINEL_DO_NOT_REFLECT',
-                        type: 'server_error'
-                    }
-                }));
+            if (mode === 'exhausted' || mode === 'unknown' || (mode === 'post-failure' && requestHasToolResult)) {
+                res.writeHead(mode === 'exhausted' ? 429 : mode === 'unknown' ? 418 : 500, { 'content-type': 'application/json' });
+                res.end(JSON.stringify({ error: { message: 'PRIVATE_SENTINEL_DO_NOT_REFLECT', type: 'server_error' } }));
                 return;
             }
             if (mode === 'recovered' && plannerRequest && !requestHasToolResult && recoveredAttempts++ === 0) {
-                res.writeHead(503, {
-                    'content-type': 'application/json'
-                });
-                res.end(JSON.stringify({
-                    error: {
-                        message: 'PRIVATE_SENTINEL_DO_NOT_REFLECT',
-                        type: 'server_error'
-                    }
-                }));
+                res.writeHead(503, { 'content-type': 'application/json' });
+                res.end(JSON.stringify({ error: { message: 'PRIVATE_SENTINEL_DO_NOT_REFLECT', type: 'server_error' } }));
                 return;
             }
-            if (plannerRequest) plannerCount += 1;
-            res.writeHead(200, {
-                'content-type': 'text/event-stream'
-            });
-            const response = plannerRequest ? conformanceResponse(mode === 'recovered' ? 'b' : mode === 'post-failure' ? 'f' : mode, body, plannerCount) : finalText('auxiliary request');
-            if (response.includes(`"name":"${TOOL}"`)) servedToolCalls += 1;
+            if (plannerRequest)
+                plannerCount += 1;
+            res.writeHead(200, { 'content-type': 'text/event-stream' });
+            const response = plannerRequest
+                ? conformanceResponse(mode === 'recovered' ? 'b' : mode === 'post-failure' ? 'f' : mode, body, plannerCount)
+                : finalText('auxiliary request');
+            if (response.includes(`"name":"${TOOL}"`))
+                servedToolCalls += 1;
             res.end(response);
         });
     });
-    await new Promise((resolve)=>server.listen(0, '127.0.0.1', resolve));
+    await new Promise(resolve => server.listen(0, '127.0.0.1', resolve));
     const port = server.address().port;
     const cwd = mkdtempSync(join(tmpdir(), 'gotry-conformance-cwd-'));
     const dsh = mkdtempSync(join(tmpdir(), 'gotry-conformance-dsh-'));
@@ -844,139 +576,56 @@ async function runConformanceCase(mode, executableOverride) {
     const configPath = join(cwd, 'benchmark-env-config.json');
     writeFileSync(runner, `const fs = require('node:fs'); const path = ${JSON.stringify(runnerCount)}; const count = fs.existsSync(path) ? Number(fs.readFileSync(path, 'utf8')) : 0; fs.writeFileSync(path, String(count + 1)); process.stdout.write(JSON.stringify({ schema_version: 'gotry_benchmark_tool_result_v1', status: 'ok', result: { marker: '${MARKER}' } }))`);
     writeFileSync(configPath, JSON.stringify({
-        schema_version: 'gotry_benchmark_environment_bridge_v4',
-        enabled: true,
-        executable: process.execPath,
-        cwd,
-        argv_prefix: [
-            runner
-        ],
-        tools: [
-            {
-                name: 'lookup',
-                description: 'Lookup.',
-                input_schema: LOOKUP_INPUT_SCHEMA,
-                output_keys: [
-                    'marker'
-                ],
-                domain_outcomes: [
-                    {
-                        status: 'miss',
-                        code: 'NOT_FOUND',
-                        recovery: 'none'
-                    }
-                ]
-            }
-        ],
-        timeout_ms: 2_000,
-        max_output_bytes: 4_096,
-        terminal_output: {
-            tag: 'benchmark_terminal',
-            max_bytes: mode === 'large' ? 128 * 1024 : 4_096,
-            body_schema: BRIDGE_E2E_BODY_SCHEMA
-        },
-        isolation: {
-            mode: 'host-enforced',
-            writes: 'forbidden',
-            network: 'denied'
-        }
+        schema_version: 'gotry_benchmark_environment_bridge_v4', enabled: true,
+        executable: process.execPath, cwd, argv_prefix: [runner],
+        tools: [{ name: 'lookup', description: 'Lookup.', input_schema: LOOKUP_INPUT_SCHEMA, output_keys: ['marker'], domain_outcomes: [{ status: 'miss', code: 'NOT_FOUND', recovery: 'none' }] }], timeout_ms: 2_000, max_output_bytes: 4_096,
+        terminal_output: { tag: 'benchmark_terminal', max_bytes: mode === 'large' ? 128 * 1024 : 4_096, body_schema: BRIDGE_E2E_BODY_SCHEMA },
+        isolation: { mode: 'host-enforced', writes: 'forbidden', network: 'denied' },
     }));
     const env = {
         ...process.env,
-        DSH_TOOLS_MODE: 'both',
-        DSH_HOME: dsh,
-        LLM_API_KEY: 'synthetic-conformance-key',
-        LLM_BASE_URL: `http://127.0.0.1:${port}/v1`,
-        LLM_MODEL: 'synthetic-conformance-model',
-        GOTRY_BENCHMARK_ENV_CONFIG: configPath,
-        DEEPSEEK_API_KEY: 'synthetic-conformance-key',
-        DEEPSEEK_BASE_URL: `http://127.0.0.1:${port}/v1`,
+        DSH_TOOLS_MODE: 'both', DSH_HOME: dsh,
+        LLM_API_KEY: 'synthetic-conformance-key', LLM_BASE_URL: `http://127.0.0.1:${port}/v1`,
+        LLM_MODEL: 'synthetic-conformance-model', GOTRY_BENCHMARK_ENV_CONFIG: configPath,
+        DEEPSEEK_API_KEY: 'synthetic-conformance-key', DEEPSEEK_BASE_URL: `http://127.0.0.1:${port}/v1`,
         NODE_OPTIONS: [
             process.env.NODE_OPTIONS,
-            ...!executableOverride ? [
-                `--import=${TSX_LOADER}`
-            ] : []
-        ].filter(Boolean).join(' ')
+            ...(!executableOverride ? [`--import=${TSX_LOADER}`] : []),
+        ].filter(Boolean).join(' '),
     };
-    for (const key of [
-        'GOTRY_LLM_MODEL',
-        'HTTP_PROXY',
-        'HTTPS_PROXY',
-        'ALL_PROXY',
-        'http_proxy',
-        'https_proxy',
-        'all_proxy'
-    ])delete env[key];
+    for (const key of ['GOTRY_LLM_MODEL', 'HTTP_PROXY', 'HTTPS_PROXY', 'ALL_PROXY', 'http_proxy', 'https_proxy', 'all_proxy'])
+        delete env[key];
     env.NO_PROXY = '127.0.0.1,localhost';
     let stdout = '';
     let stderr = '';
     try {
         const executable = executableOverride || process.execPath;
-        const argv = executableOverride ? [
-            'conformance smoke'
-        ] : [
-            BIN,
-            'conformance smoke'
-        ];
-        const child = spawn(executable, argv, {
-            cwd,
-            env,
-            stdio: [
-                'ignore',
-                'pipe',
-                'pipe'
-            ]
-        });
-        child.stdout.on('data', (chunk)=>{
-            stdout += chunk.toString();
-        });
-        child.stderr.on('data', (chunk)=>{
-            stderr += chunk.toString();
-        });
-        const exit = await new Promise((resolve)=>{
-            const timer = setTimeout(()=>{
-                child.kill('SIGKILL');
-                resolve(null);
-            }, TIMEOUT_MS);
-            child.once('close', (code)=>{
-                clearTimeout(timer);
-                resolve(code);
-            });
+        const argv = executableOverride ? ['conformance smoke'] : [BIN, 'conformance smoke'];
+        const child = spawn(executable, argv, { cwd, env, stdio: ['ignore', 'pipe', 'pipe'] });
+        child.stdout.on('data', chunk => { stdout += chunk.toString(); });
+        child.stderr.on('data', chunk => { stderr += chunk.toString(); });
+        const exit = await new Promise(resolve => {
+            const timer = setTimeout(() => { child.kill('SIGKILL'); resolve(null); }, TIMEOUT_MS);
+            child.once('close', code => { clearTimeout(timer); resolve(code); });
         });
         const runnerInvocations = existsSync(runnerCount) ? Number(readFileSync(runnerCount, 'utf8')) : 0;
-        return {
-            exit,
-            stdout,
-            stderr,
-            requests,
-            servedToolCalls,
-            runnerInvocations
-        };
-    } finally{
-        await new Promise((resolve)=>server.close(()=>resolve()));
-        rmSync(dsh, {
-            recursive: true,
-            force: true
-        });
-        rmSync(cwd, {
-            recursive: true,
-            force: true
-        });
+        return { exit, stdout, stderr, requests, servedToolCalls, runnerInvocations };
+    }
+    finally {
+        await new Promise(resolve => server.close(() => resolve()));
+        rmSync(dsh, { recursive: true, force: true });
+        rmSync(cwd, { recursive: true, force: true });
     }
 }
 async function assertTerminalDiagnostics(executableOverride) {
-    const terminalReasons = (stderr)=>[
-            ...stderr.matchAll(/benchmark terminal output unavailable \(([^)]+)\)/g)
-        ].map((match)=>match[1]);
+    const terminalReasons = (stderr) => [...stderr.matchAll(/benchmark terminal output unavailable \(([^)]+)\)/g)].map(match => match[1]);
     const exhausted = await runConformanceCase('exhausted', executableOverride);
     assert.notEqual(exhausted.exit, 0, 'exhausted transient model failure exits non-zero');
     assert.match(exhausted.stderr, /child_model_capacity/, 'exhausted transient model failure emits coarse capacity enum');
     assert.ok(exhausted.requests.length > 1, 'exhausted case actually exercises retry attempts');
     assert.equal(exhausted.stdout, '', 'exhausted transient model failure releases no terminal stdout');
     assert.equal(exhausted.stderr.includes('PRIVATE_SENTINEL_DO_NOT_REFLECT'), false, 'exhausted error body is never reflected');
-    assert.deepEqual(terminalReasons(exhausted.stderr), [
-        'child_model_capacity'
-    ], 'exhausted emits exactly one terminal reason');
+    assert.deepEqual(terminalReasons(exhausted.stderr), ['child_model_capacity'], 'exhausted emits exactly one terminal reason');
     const recovered = await runConformanceCase('recovered', executableOverride);
     assert.equal(recovered.exit, 0, 'transient model failure followed by valid terminal recovers');
     assert.match(recovered.stdout, /<benchmark_terminal>/, 'recovered run releases terminal stdout');
@@ -987,23 +636,17 @@ async function assertTerminalDiagnostics(executableOverride) {
     assert.equal(postFailure.runnerInvocations, 1, 'post-bridge failure follows exactly one successful bridge invocation');
     assert.match(postFailure.stderr, /child_model_server/, 'post-bridge model failure emits server enum');
     assert.equal(postFailure.stdout, '', 'post-bridge model failure releases no terminal stdout');
-    assert.deepEqual(terminalReasons(postFailure.stderr), [
-        'child_model_server'
-    ], 'post-bridge emits exactly one terminal reason');
+    assert.deepEqual(terminalReasons(postFailure.stderr), ['child_model_server'], 'post-bridge emits exactly one terminal reason');
     const unknown = await runConformanceCase('unknown', executableOverride);
     assert.notEqual(unknown.exit, 0, 'unknown model failure exits non-zero');
     assert.match(unknown.stderr, /child_runtime_error/, 'unknown model failure collapses to generic runtime enum');
     assert.equal(unknown.stdout, '', 'unknown model failure releases no terminal stdout');
     assert.equal(unknown.stderr.includes('PRIVATE_SENTINEL_DO_NOT_REFLECT'), false, 'unknown error body is never reflected');
-    assert.deepEqual(terminalReasons(unknown.stderr), [
-        'child_runtime_error'
-    ], 'unknown emits exactly one terminal reason');
+    assert.deepEqual(terminalReasons(unknown.stderr), ['child_runtime_error'], 'unknown emits exactly one terminal reason');
     const precedence = await runConformanceCase('f', executableOverride);
     assert.match(precedence.stderr, /child_conformance_failure/, 'conformance-specific failure remains higher precedence than final generic error');
     assert.equal(precedence.stderr.includes('child_runtime_error'), false, 'generic terminal classification does not double-write');
-    assert.deepEqual(terminalReasons(precedence.stderr), [
-        'child_conformance_failure'
-    ], 'precedence emits exactly one terminal reason');
+    assert.deepEqual(terminalReasons(precedence.stderr), ['child_conformance_failure'], 'precedence emits exactly one terminal reason');
 }
 async function assertOutputConformance(executableOverride) {
     const a = await runConformanceCase('a', executableOverride);
@@ -1011,24 +654,19 @@ async function assertOutputConformance(executableOverride) {
     assert.equal(a.servedToolCalls, 1, 'A exposes exactly one bridge call');
     assert.equal(a.runnerInvocations, 1, 'A executes the bridge subprocess exactly once');
     assert.ok(a.stdout.includes('<benchmark_terminal>'), 'A forwards only tagged terminal output');
-    assert.ok(a.requests.some((request)=>JSON.stringify(request).includes(`matching exactly ${TERMINAL_OUTLINE}`)), 'A system prompt projects the exact terminal schema outline');
+    // Round 12(#215):system prompt 投影 exact body schema outline(结构合同单一来源)。
+    assert.ok(a.requests.some(request => JSON.stringify(request).includes(`matching exactly ${TERMINAL_OUTLINE}`)), 'A system prompt projects the exact terminal schema outline');
     const b = await runConformanceCase('b', executableOverride);
     assert.equal(b.exit, 0, 'B malformed terminal correction then valid terminal exits 0');
-    assert.equal(b.servedToolCalls, 1, `B exposes exactly one bridge call; request shapes=${JSON.stringify(b.requests.map((request)=>({
-            tools: names(request),
-            roles: (request.messages ?? []).map((message)=>message.role)
-        })))}`);
+    assert.equal(b.servedToolCalls, 1, `B exposes exactly one bridge call; request shapes=${JSON.stringify(b.requests.map(request => ({ tools: names(request), roles: (request.messages ?? []).map(message => message.role) })))}`);
     assert.equal(b.runnerInvocations, 1, 'B format-only correction does not rerun the bridge subprocess');
-    assert.ok(b.requests.some((request)=>JSON.stringify(request).includes('BENCHMARK_CONFORMANCE_TERMINAL') && JSON.stringify(request).includes(`matching exactly ${TERMINAL_OUTLINE}`)), 'B terminal correction projects the same schema outline as the system prompt');
+    assert.ok(b.requests.some(request => JSON.stringify(request).includes('BENCHMARK_CONFORMANCE_TERMINAL') && JSON.stringify(request).includes(`matching exactly ${TERMINAL_OUTLINE}`)), 'B terminal correction projects the same schema outline as the system prompt');
     const schema = await runConformanceCase('schema', executableOverride);
     assert.notEqual(schema.exit, 0, 'schema-invalid terminal body (extra root key) is rejected after the single correction');
     assert.match(schema.stderr, /benchmark terminal output unavailable \(child_conformance_failure\)/, 'schema-invalid terminal emits the stable conformance reason code');
     assert.equal(schema.runnerInvocations, 1, 'schema-invalid terminal still executed the bridge exactly once');
     assert.equal(schema.stdout.includes('<benchmark_terminal>'), false, 'schema-invalid terminal body is never released to stdout');
-    for (const mode of [
-        'c',
-        'd'
-    ]){
+    for (const mode of ['c', 'd']) {
         const result = await runConformanceCase(mode, executableOverride);
         assert.notEqual(result.exit, 0, `${mode.toUpperCase()} repeated invalid output is non-zero`);
         assert.match(result.stderr, /benchmark terminal output unavailable \(child_conformance_failure\)/, `${mode.toUpperCase()} emits a stable conformance reason code`);
@@ -1056,8 +694,10 @@ async function assertOutputConformance(executableOverride) {
 const packaged = process.env.GOTRY_BRIDGE_E2E_BIN;
 assertRuntimeSelectionAndVersionGuards();
 const sourceRuntimeChecked = await assertSourceRuntimeContractWhenAvailable();
-if (sourceRuntimeChecked) await assertOutputConformance();
-if (sourceRuntimeChecked) await assertTerminalDiagnostics();
+if (sourceRuntimeChecked)
+    await assertOutputConformance();
+if (sourceRuntimeChecked)
+    await assertTerminalDiagnostics();
 if (packaged) {
     await assertRuntimeContract(packaged);
     await assertOutputConformance(packaged);
@@ -1069,53 +709,16 @@ if (packaged) {
     try {
         const configPath = join(missingServiceRoot, 'bridge.json');
         writeFileSync(configPath, JSON.stringify({
-            schema_version: 'gotry_benchmark_environment_bridge_v4',
-            enabled: true,
-            executable: process.execPath,
-            cwd: missingServiceRoot,
-            argv_prefix: [
-                '-e',
-                'process.exit(0)'
-            ],
-            tools: [
-                {
-                    name: 'lookup',
-                    description: 'Lookup.',
-                    input_schema: LOOKUP_INPUT_SCHEMA,
-                    output_keys: [
-                        'marker'
-                    ],
-                    domain_outcomes: [
-                        {
-                            status: 'miss',
-                            code: 'NOT_FOUND',
-                            recovery: 'none'
-                        }
-                    ]
-                }
-            ],
-            timeout_ms: 100,
-            max_output_bytes: 4_096,
-            terminal_output: {
-                tag: 'benchmark_terminal',
-                max_bytes: 4_096,
-                body_schema: BRIDGE_E2E_BODY_SCHEMA
-            },
-            isolation: {
-                mode: 'host-enforced',
-                writes: 'forbidden',
-                network: 'denied'
-            }
+            schema_version: 'gotry_benchmark_environment_bridge_v4', enabled: true,
+            executable: process.execPath, cwd: missingServiceRoot, argv_prefix: ['-e', 'process.exit(0)'],
+            tools: [{ name: 'lookup', description: 'Lookup.', input_schema: LOOKUP_INPUT_SCHEMA, output_keys: ['marker'], domain_outcomes: [{ status: 'miss', code: 'NOT_FOUND', recovery: 'none' }] }], timeout_ms: 100, max_output_bytes: 4_096,
+            terminal_output: { tag: 'benchmark_terminal', max_bytes: 4_096, body_schema: BRIDGE_E2E_BODY_SCHEMA },
+            isolation: { mode: 'host-enforced', writes: 'forbidden', network: 'denied' },
         }));
-        assert.throws(()=>packagedBridge.registerBenchmarkEnvironmentBridge(configPath, ()=>{}, undefined), /benchmark environment bridge subprocess unavailable/, 'packaged explicit opt-in fails hard when no active subprocess provider exists');
-    } finally{
-        rmSync(missingServiceRoot, {
-            recursive: true,
-            force: true
-        });
+        assert.throws(() => packagedBridge.registerBenchmarkEnvironmentBridge(configPath, () => { }, undefined), /benchmark environment bridge subprocess unavailable/, 'packaged explicit opt-in fails hard when no active subprocess provider exists');
+    }
+    finally {
+        rmSync(missingServiceRoot, { recursive: true, force: true });
     }
 }
-console.log(`benchmark environment bridge E2E: OK (${packaged ? sourceRuntimeChecked ? 'source + packaged' : 'source-static + packaged' : sourceRuntimeChecked ? 'source' : 'source-static'})`);
-
-
-//# sourceURL=ts/scripts/benchmark-environment-bridge-e2e.ts
+console.log(`benchmark environment bridge E2E: OK (${packaged ? (sourceRuntimeChecked ? 'source + packaged' : 'source-static + packaged') : (sourceRuntimeChecked ? 'source' : 'source-static')})`);