@danceiny/gotry 0.0.1-rc.19 → 0.0.1-rc.20

package/dist/scripts/anything-tests.js

@@ -11,11 +11,17 @@ try {
             await writeFile(p, `#!/bin/sh
 cat <<'JSON'
 {"candidates":[
-  {"type":"hotel","name":"Park Hyatt","hotel":{"id":"h1","name":"Park Hyatt","latitude":22.31,"longitude":114.16}},
-  {"type":"city","name":"大理市","region":{"id":"d1","name":"Dali","latitude":25.58,"longitude":100.21}},
-  {"type":"place","name":"洱海","region":{"id":"r1","name":"Erhai Lake","latitude":25.74,"longitude":100.25}}
+  {"type":"hotel","matchScore":180,"hotel":{"id":"h1","name":{"en":"Park Hyatt","zh":"柏悦酒店"},"star":5,"destinationId":"d1","latlngCoordinator":{"google":{"lat":22.31,"lng":114.16}}}},
+  {"type":"city","matchScore":250,"region":{"id":"d1","name":{"en":"Dali","zh":"大理市"},"countryCode":"CN","coordinates":{"centerLat":25.58,"centerLng":100.21}}},
+  {"type":"place","matchScore":90,"region":{"id":"r1","name":{"zh":"洱海"}},"place":{"latlngCoordinator":{"google":{"lat":25.74,"lng":100.25}}}}
 ]}
 JSON
+`, {
+                mode: 0o755
+            });
+        } else if (behaviour === 'echo-args') {
+            await writeFile(p, `#!/bin/sh
+printf '{"candidates":[{"type":"city","region":{"id":"%s"}}]}' "$*"
 `, {
                 mode: 0o755
             });
@@ -31,6 +37,13 @@ JSON
             await writeFile(p, `#!/bin/sh
 echo 'hotelbe down' >&2
 exit 1
+`, {
+                mode: 0o755
+            });
+        } else if (behaviour === 'unknown-command') {
+            await writeFile(p, `#!/bin/sh
+echo "error: unknown command 'anything'" >&2
+exit 1
 `, {
                 mode: 0o755
             });
@@ -46,44 +59,80 @@ while :; do sleep 5; done
     }
     const ok = await fakeBin('hbcli-ok', 'ok');
     const r1 = await anythingSearch({
-        keyword: 'Park Hyatt',
+        keyword: '大理',
         hbcliBin: ok
     });
     assert.equal(r1.verdict, 'hit');
     assert.equal(r1.hits.length, 3);
     assert.ok(r1.evidence.includes('hbcli-anything@'));
-    console.log(`1. Park Hyatt → hit (${r1.hits.length} candidates) OK`);
-    const empty = await fakeBin('hbcli-empty', 'empty');
+    const [hotelHit, cityHit, placeHit] = r1.hits;
+    assert.equal(hotelHit.type, 'hotel');
+    assert.equal(hotelHit.name, '柏悦酒店', 'hotel i18n name 应 zh 优先');
+    assert.equal(hotelHit.hotelId, 'h1');
+    assert.equal(hotelHit.destinationId, 'd1', 'hotel 候选 destinationId 取所属目的地');
+    assert.equal(hotelHit.star, 5);
+    assert.equal(hotelHit.latitude, 22.31);
+    assert.equal(hotelHit.longitude, 114.16);
+    assert.equal(hotelHit.score, 180);
+    assert.equal(cityHit.type, 'city');
+    assert.equal(cityHit.name, '大理市', 'region i18n name 应 zh 优先');
+    assert.equal(cityHit.destinationId, 'd1');
+    assert.equal(cityHit.latitude, 25.58, 'city 坐标取 region.coordinates.centerLat');
+    assert.equal(cityHit.longitude, 100.21);
+    assert.equal(placeHit.type, 'place');
+    assert.equal(placeHit.name, '洱海');
+    assert.equal(placeHit.latitude, 25.74, 'place 坐标回退 place.latlngCoordinator');
+    console.log('1. 真实候选 wire 形状(hotel/city/place 逐字段)OK');
+    const echo = await fakeBin('hbcli-echo-args', 'echo-args');
     const r2 = await anythingSearch({
+        keyword: '大理',
+        contentType: 'city',
+        parentDestinationId: 'd1',
+        hbcliBin: echo
+    });
+    assert.equal(r2.verdict, 'hit');
+    assert.equal(r2.hits[0].destinationId, '--json search anything 大理 --content-type city --destination-id d1', 'spawn 旗标形态(--json 前置/--destination-id 映射)应稳定,实际 ' + r2.hits[0].destinationId);
+    console.log('2. 旗标回归(--json 前置 + --destination-id 映射)OK');
+    const empty = await fakeBin('hbcli-empty', 'empty');
+    const r3 = await anythingSearch({
         keyword: 'nothing',
         hbcliBin: empty
     });
-    assert.equal(r2.verdict, 'miss');
-    assert.equal(r2.hits.length, 0);
-    console.log('2. nothing → miss OK');
+    assert.equal(r3.verdict, 'miss');
+    assert.equal(r3.hits.length, 0);
+    console.log('3. nothing → miss OK');
     const fail = await fakeBin('hbcli-fail', 'fail');
-    const r3 = await anythingSearch({
+    const r4 = await anythingSearch({
         keyword: 'x',
         hbcliBin: fail
     });
-    assert.equal(r3.verdict, 'error');
-    assert.equal(r3.ok, false);
-    assert.match(r3.evidence, /error/);
-    console.log('3. fail → error (降级) OK');
+    assert.equal(r4.verdict, 'error');
+    assert.equal(r4.ok, false);
+    assert.match(r4.evidence, /error/);
+    console.log('4. fail → error (降级) OK');
+    const old = await fakeBin('hbcli-old', 'unknown-command');
+    const r5 = await anythingSearch({
+        keyword: 'x',
+        hbcliBin: old
+    });
+    assert.equal(r5.verdict, 'error');
+    assert.match(r5.error ?? '', /hbcli update|gotry setup/, 'error 应带升级指引,实际 ' + r5.error);
+    assert.match(r5.evidence, /unknown command/, 'evidence 保留上游原话(溯源)');
+    console.log(`5. 旧 CLI unknown command → 升级指引 OK(${r5.error})`);
     const hang = await fakeBin('hbcli-hang', 'hang');
-    const r4 = await anythingSearch({
+    const r6 = await anythingSearch({
         keyword: 'x',
         hbcliBin: hang,
         timeoutMs: 800
     });
-    assert.equal(r4.verdict, 'error', 'timeout 应判 error');
-    console.log(`4. hang/timeout → error OK (latency=${r4.latencyMs}ms)`);
-    const r5 = await anythingSearch({
+    assert.equal(r6.verdict, 'error', 'timeout 应判 error');
+    console.log(`6. hang/timeout → error OK (latency=${r6.latencyMs}ms)`);
+    const r7 = await anythingSearch({
         keyword: '   '
     });
-    assert.equal(r5.verdict, 'error');
-    console.log('5. empty keyword → error OK');
-    console.log('\nANYTHING TESTS: 5/5 OK(hbcli fake + graceful degrade)');
+    assert.equal(r7.verdict, 'error');
+    console.log('7. empty keyword → error OK');
+    console.log('\nANYTHING TESTS: 7/7 OK(hbcli fake + wire 对齐 + 降级诚实)');
 } finally{
     await rm(tmp, {
         recursive: true,

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

@@ -12,6 +12,22 @@ 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 TSX_LOADER = pathToFileURL(createRequire(import.meta.url).resolve('tsx')).href;
 function runRuntimeProbe(options) {
     const fixture = mkdtempSync(join(tmpdir(), 'gotry-runtime-probe-'));
@@ -61,7 +77,7 @@ function assertRuntimeSelectionAndVersionGuards() {
     const legacyFallback = runRuntimeProbe({
         vendorVersion: '0.1.2-alpha.1'
     });
-    assert.equal(legacyFallback, null, 'D-27 removal (#120): legacy vendored fallback no longer resolves even outside benchmark — fail-closed');
+    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',
@@ -183,7 +199,7 @@ function finalText(text) {
         ]
     }) + 'data: [DONE]\n\n';
 }
-function toolCall(callId = 'bridge-call-1') {
+function toolCall(callId = 'bridge-call-1', city = 'Dubai') {
     return sse({
         id: `bridge-${callId}`,
         object: 'chat.completion.chunk',
@@ -202,7 +218,7 @@ function toolCall(callId = 'bridge-call-1') {
                                     action: 'call',
                                     tool: 'lookup',
                                     arguments: {
-                                        city: 'Dubai'
+                                        city
                                     }
                                 })
                             }
@@ -237,7 +253,9 @@ function anyToolResultPresent(body) {
 }
 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 chunks = [];
         req.on('data', (c)=>chunks.push(Buffer.from(c)));
@@ -253,8 +271,15 @@ async function runCase(mode, executableOverride, extraEnv = {}) {
             if (mode === 'spawn-failed' && names(body).includes(TOOL) && !anyToolResultPresent(body) && spawnTarget) rmSync(spawnTarget, {
                 force: true
             });
-            if (mode !== 'disabled' && mode !== 'invalid-path' && mode !== 'invalid-schema' && mode !== 'unsafe-config' && names(body).includes(TOOL) && !anyToolResultPresent(body)) res.end(toolCall());
-            else res.end(finalText(mode === 'enabled' ? '<benchmark_terminal>{"status":"succeeded"}</benchmark_terminal>' : '<benchmark_terminal>{"status":"succeeded"}</benchmark_terminal>'));
+            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)) {
+                servedToolCalls += 1;
+                res.end(toolCall());
+            } else {
+                res.end(finalText('<benchmark_terminal>{"status":"succeeded"}</benchmark_terminal>'));
+            }
         });
     });
     await new Promise((resolve)=>server.listen(0, '127.0.0.1', resolve));
@@ -269,15 +294,16 @@ async function runCase(mode, executableOverride, extraEnv = {}) {
     });
     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 === '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({ 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({ result: { marker: '${MARKER}', leaked } }))`;
-    writeFileSync(runner, `if (process.argv.length !== 5 || process.argv[2] !== 'call' || process.argv[3] !== 'lookup' || JSON.parse(process.argv[4]).city !== 'Dubai') process.exit(2); ${runnerBody}`);
+    const runnerBody = mode === 'domain-recovery-failed' ? `if (args.city === 'Dubai') process.stdout.write(JSON.stringify({ schema_version: 'gotry_benchmark_tool_result_v1', status: 'miss', code: 'NOT_FOUND', recovery: 'revise_arguments' })); else { process.stderr.write('PRIVATE_RECOVERY_RUNNER_DIAGNOSTIC_DO_NOT_REFLECT'); process.exit(17) }` : mode === 'domain-recovery' ? `if (args.city === 'Dubai') process.stdout.write(JSON.stringify({ schema_version: 'gotry_benchmark_tool_result_v1', status: 'miss', code: 'NOT_FOUND', recovery: 'revise_arguments' })); else process.stdout.write(JSON.stringify({ schema_version: 'gotry_benchmark_tool_result_v1', status: 'ok', result: { marker: '${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_v2',
+        schema_version: mode === 'invalid-schema' ? 'invalid' : 'gotry_benchmark_environment_bridge_v3',
         enabled: true,
         executable: mode === 'spawn-failed' ? spawnTarget : process.execPath,
         cwd,
@@ -286,15 +312,24 @@ async function runCase(mode, executableOverride, extraEnv = {}) {
         ] : [
             runner
         ],
-        allowed_tools: [
-            'lookup'
+        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'
+                    }
+                ]
+            }
         ],
-        allowed_output_keys: {
-            lookup: [
-                'marker',
-                'leaked'
-            ]
-        },
         timeout_ms: mode === 'timeout' ? 50 : 10_000,
         max_output_bytes: mode === 'output-truncated' ? 1_024 : 4_096,
         terminal_output: {
@@ -398,13 +433,16 @@ async function runCase(mode, executableOverride, extraEnv = {}) {
             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
+            optionalResolutionHits,
+            servedToolCalls,
+            runnerArguments
         };
     } finally{
         await new Promise((resolve)=>server.close(()=>resolve()));
@@ -437,6 +475,21 @@ async function assertRuntimeContract(executableOverride) {
     assert.deepEqual(enabledToolNames, [
         TOOL
     ], `${target} enabled runtime must expose exactly the benchmark tool; observed tool names=${JSON.stringify(enabledToolNames)}`);
+    const bridgeTool = enabled.requests.find((request)=>names(request).includes(TOOL))?.tools?.find((tool)=>names({
+            tools: [
+                tool
+            ]
+        }).includes(TOOL));
+    const flatSchema = bridgeTool?.function?.parameters;
+    assert.ok(Array.isArray(flatSchema?.oneOf), `${target} bridge exposes flat oneOf schema`);
+    const lookupBranch = flatSchema.oneOf.find((branch)=>branch?.properties?.tool?.const === 'lookup');
+    assert.deepEqual(lookupBranch?.required, [
+        'action',
+        'tool',
+        'arguments'
+    ]);
+    assert.equal(lookupBranch?.additionalProperties, false);
+    assert.deepEqual(lookupBranch?.properties?.arguments, LOOKUP_INPUT_SCHEMA);
     assert.equal(enabledToolNames.some((name)=>name.startsWith('calendar_') || name.startsWith('map_')), false, `${target} benchmark projection must not expose calendar/map tools`);
     assert.deepEqual(enabled.optionalResolutionHits, {
         calendar: 0,
@@ -460,6 +513,37 @@ async function assertRuntimeContract(executableOverride) {
         return (prompt.match(/You are GoTry, a task-agnostic travel planning assistant\./g) ?? []).length === 1 && (prompt.match(/Use only the current conversation and tools available in this benchmark session\./g) ?? []).length === 1;
     }), `${target} benchmark persona has each stable sentence exactly once per request`);
     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);
+        })), `${target} declared typed miss reaches model history`);
+    assert.match(recovered.stdout, /<benchmark_terminal>/, `${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'
+        }
+    ], `${target} failed recovery reaches the revised second adapter invocation`);
+    assert.equal(failedRecovery.stdout, '', `${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`);
     const debugRedaction = await runCase('debug-redaction', executableOverride);
     assert.equal(debugRedaction.exit, 0, `${target} benchmark debug mode preserves successful execution`);
     assert.equal(debugRedaction.output.includes('PRIVATE_QUERY_SENTINEL_DO_NOT_REFLECT'), false, `${target} benchmark debug output never reflects the private task`);
@@ -729,23 +813,32 @@ async function runConformanceCase(mode, executableOverride) {
     const runner = join(cwd, 'synthetic-runner.js');
     const runnerCount = join(cwd, 'runner-count.txt');
     const configPath = join(cwd, 'benchmark-env-config.json');
-    writeFileSync(runner, `const fs = require('node:fs'); const path = ${JSON.stringify(runnerCount)}; const count = fs.existsSync(path) ? Number(fs.readFileSync(path, 'utf8')) : 0; fs.writeFileSync(path, String(count + 1)); process.stdout.write(JSON.stringify({ result: { marker: '${MARKER}' } }))`);
+    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_v2',
+        schema_version: 'gotry_benchmark_environment_bridge_v3',
         enabled: true,
         executable: process.execPath,
         cwd,
         argv_prefix: [
             runner
         ],
-        allowed_tools: [
-            'lookup'
+        tools: [
+            {
+                name: 'lookup',
+                description: 'Lookup.',
+                input_schema: LOOKUP_INPUT_SCHEMA,
+                output_keys: [
+                    'marker'
+                ],
+                domain_outcomes: [
+                    {
+                        status: 'miss',
+                        code: 'NOT_FOUND',
+                        recovery: 'none'
+                    }
+                ]
+            }
         ],
-        allowed_output_keys: {
-            lookup: [
-                'marker'
-            ]
-        },
         timeout_ms: 2_000,
         max_output_bytes: 4_096,
         terminal_output: {
@@ -939,7 +1032,7 @@ if (packaged) {
     try {
         const configPath = join(missingServiceRoot, 'bridge.json');
         writeFileSync(configPath, JSON.stringify({
-            schema_version: 'gotry_benchmark_environment_bridge_v2',
+            schema_version: 'gotry_benchmark_environment_bridge_v3',
             enabled: true,
             executable: process.execPath,
             cwd: missingServiceRoot,
@@ -947,8 +1040,22 @@ if (packaged) {
                 '-e',
                 'process.exit(0)'
             ],
-            allowed_tools: [
-                'lookup'
+            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,