@dalmia/calibrate-mcp 0.0.47 → 0.0.48

package/bin/mcp-server.js

@@ -1759,7 +1759,7 @@ var init_versions = __esm(() => {
   version2 = {
     major: 4,
     minor: 5,
-    patch: 2
+    patch: 4
   };
 });
 
@@ -3412,22 +3412,95 @@ function isRecursive(inst, stack) {
       result = true;
   };
   const def = inst._zod.def;
-  if (def.type === "lazy") {
-    check(inst._zod.innerType);
-  } else {
-    const shape = def.shape;
-    if (shape)
-      for (const key of Reflect.ownKeys(shape))
-        check(shape[key]);
-    for (const key in def) {
-      const value = def[key];
-      if (!value || typeof value !== "object")
-        continue;
-      if (value._zod)
-        check(value);
-      else if (Array.isArray(value))
-        for (const el of value)
-          check(el);
+  const kind = def.type;
+  switch (kind) {
+    case "object": {
+      for (const key of Reflect.ownKeys(def.shape))
+        check(def.shape[key]);
+      check(def.catchall);
+      break;
+    }
+    case "array":
+      check(def.element);
+      break;
+    case "tuple":
+      for (const el of def.items)
+        check(el);
+      check(def.rest);
+      break;
+    case "record":
+    case "map":
+      check(def.keyType);
+      check(def.valueType);
+      break;
+    case "set":
+      check(def.valueType);
+      break;
+    case "union":
+      for (const el of def.options)
+        check(el);
+      break;
+    case "intersection":
+      check(def.left);
+      check(def.right);
+      break;
+    case "optional":
+    case "nullable":
+    case "default":
+    case "prefault":
+    case "catch":
+    case "readonly":
+    case "nonoptional":
+    case "promise":
+    case "success":
+      check(def.innerType);
+      break;
+    case "pipe":
+      check(def.in);
+      check(def.out);
+      break;
+    case "function":
+      check(def.input);
+      check(def.output);
+      break;
+    case "lazy":
+      check(inst._zod.innerType);
+      break;
+    case "template_literal":
+    case "string":
+    case "number":
+    case "int":
+    case "boolean":
+    case "bigint":
+    case "symbol":
+    case "undefined":
+    case "null":
+    case "void":
+    case "never":
+    case "any":
+    case "unknown":
+    case "date":
+    case "nan":
+    case "enum":
+    case "literal":
+    case "file":
+    case "transform":
+    case "custom":
+      break;
+    default: {
+      for (const key in def) {
+        const desc = Object.getOwnPropertyDescriptor(def, key);
+        if (!desc || desc.get)
+          continue;
+        const value = desc.value;
+        if (!value || typeof value !== "object")
+          continue;
+        if (value._zod)
+          check(value);
+        else if (Array.isArray(value))
+          for (const el of value)
+            check(el);
+      }
     }
   }
   stack.delete(inst);
@@ -4319,6 +4392,7 @@ function initializeContext(params) {
     cycles: params?.cycles ?? "ref",
     reused: params?.reused ?? "inline",
     intersections: [],
+    deferred: [],
     external: params?.external ?? undefined
   };
 }
@@ -4663,6 +4737,8 @@ function finalize(ctx, schema) {
         compactTypeUnion(entry[1].def ?? entry[1].schema);
       }
     }
+    for (const rewrite of ctx.deferred)
+      rewrite();
     if (ctx.intersections.length) {
       const carriers = new Map;
       for (const seen of ctx.seen.values()) {
@@ -4824,6 +4900,68 @@ function inputOptin(schema) {
   }
   return schema._zod.optin;
 }
+function stringifyKeyNames(bySchema, json, visited) {
+  if (json.$ref) {
+    if (visited.has(json))
+      return json;
+    visited.add(json);
+    const def = bySchema.get(json)?.def;
+    if (!def)
+      return json;
+    const inlined = stringifyKeyNames(bySchema, def, visited);
+    return inlined === def ? json : inlined;
+  }
+  for (const keyword of ["anyOf", "oneOf"]) {
+    const branches = json[keyword];
+    if (!Array.isArray(branches))
+      continue;
+    const mapped = branches.map((branch) => stringifyKeyNames(bySchema, branch, visited));
+    if (mapped.some((branch, i) => branch !== branches[i]))
+      json = { ...json, [keyword]: mapped };
+  }
+  const types = Array.isArray(json.type) ? json.type : [json.type];
+  const numericType = !types.includes("string") && types.some((t) => t === "number" || t === "integer");
+  const values = json.enum ?? (json.const !== undefined ? [json.const] : undefined);
+  if (!numericType && !values?.some((v) => typeof v === "number"))
+    return json;
+  const { minimum, maximum: maximum2, exclusiveMinimum, exclusiveMaximum, multipleOf, format, id, ...rest } = json;
+  if (rest.enum)
+    rest.enum = rest.enum.map((v) => typeof v === "number" ? String(v) : v);
+  else if (typeof rest.const === "number")
+    rest.const = String(rest.const);
+  if (!numericType)
+    return rest;
+  rest.type = "string";
+  if (!values)
+    rest.pattern = (types.includes("number") ? number : integer).source;
+  return rest;
+}
+function rewriteKeyNames(ctx) {
+  const bySchema = new Map;
+  for (const entry of ctx.seen.values()) {
+    if (entry.def && !bySchema.has(entry.schema))
+      bySchema.set(entry.schema, entry);
+  }
+  const rewrites = new Map;
+  for (const record of pendingRecords.get(ctx) ?? []) {
+    const seen = ctx.seen.get(record);
+    const names = (seen?.def ?? seen?.schema)?.propertyNames;
+    if (!names || names === true || rewrites.has(names))
+      continue;
+    const rewritten = stringifyKeyNames(bySchema, names, new Set);
+    if (rewritten !== names)
+      rewrites.set(names, rewritten);
+  }
+  if (!rewrites.size)
+    return;
+  for (const entry of ctx.seen.values()) {
+    for (const carrier of [entry.schema, entry.def]) {
+      const rewritten = carrier && rewrites.get(carrier.propertyNames);
+      if (rewritten)
+        carrier.propertyNames = rewritten;
+    }
+  }
+}
 function serializeDefaultValue(value, schema, ctx, json, params) {
   let unrepresentable = false;
   const serialized = JSON.stringify(value, (_, val) => {
@@ -4890,12 +5028,12 @@ var formatMap, stringProcessor = (schema, ctx, _json, _params) => {
   if (contentEncoding)
     json.contentEncoding = contentEncoding;
   if (patterns && patterns.size > 0) {
-    const regexes = [...patterns];
-    if (regexes.length === 1)
-      json.pattern = regexes[0].source;
-    else if (regexes.length > 1) {
+    const patternList = [...patterns];
+    if (patternList.length === 1)
+      json.pattern = patternList[0].source;
+    else if (patternList.length > 1) {
       json.allOf = [
-        ...regexes.map((regex) => ({
+        ...patternList.map((regex) => ({
           ...ctx.target === "draft-07" || ctx.target === "draft-04" || ctx.target === "openapi-3.0" ? { type: "string" } : {},
           pattern: regex.source
         }))
@@ -5198,7 +5336,7 @@ var formatMap, stringProcessor = (schema, ctx, _json, _params) => {
     json.minItems = minimum;
   if (typeof maximum2 === "number")
     json.maxItems = maximum2;
-}, recordProcessor = (schema, ctx, _json, params) => {
+}, pendingRecords, recordProcessor = (schema, ctx, _json, params) => {
   const json = _json;
   const def = schema._zod.def;
   json.type = "object";
@@ -5220,6 +5358,13 @@ var formatMap, stringProcessor = (schema, ctx, _json, _params) => {
         ...params,
         path: [...params.path, "propertyNames"]
       });
+      let pending = pendingRecords.get(ctx);
+      if (!pending) {
+        pending = [];
+        pendingRecords.set(ctx, pending);
+        ctx.deferred.push(() => rewriteKeyNames(ctx));
+      }
+      pending.push(schema);
     }
     json.additionalProperties = process2(def.valueType, ctx, {
       ...params,
@@ -5231,7 +5376,7 @@ var formatMap, stringProcessor = (schema, ctx, _json, _params) => {
   if (keyValues && !def.partial && !omittableOnInput) {
     const validKeyValues = [...keyValues].filter((v) => typeof v === "string" || typeof v === "number");
     if (validKeyValues.length > 0) {
-      json.required = validKeyValues;
+      json.required = validKeyValues.map(String);
     }
   }
 }, nullableProcessor = (schema, ctx, json, params) => {
@@ -5310,6 +5455,7 @@ var formatMap, stringProcessor = (schema, ctx, _json, _params) => {
   seen.ref = innerType;
 }, allProcessors;
 var init_json_schema_processors = __esm(() => {
+  init_regexes();
   init_to_json_schema();
   init_util();
   formatMap = {
@@ -5319,6 +5465,7 @@ var init_json_schema_processors = __esm(() => {
     json_string: "json-string",
     regex: ""
   };
+  pendingRecords = new WeakMap;
   UNREPRESENTABLE_DEFAULT = Symbol();
   allProcessors = {
     string: stringProcessor,
@@ -26219,15 +26366,19 @@ var require_utils2 = __commonJS(function(exports, module) {
     if (!obj || typeof obj !== "object") {
       return false;
     }
-    return !!(obj.constructor && obj.constructor.isBuffer && obj.constructor.isBuffer(obj));
+    return !!(obj.constructor && typeof obj.constructor.isBuffer === "function" && obj.constructor.isBuffer(obj));
   };
   var combine = function combine2(a, b, arrayLimit, plainObjects, throwOnLimitExceeded) {
     if (isOverflow(a)) {
       if (throwOnLimitExceeded) {
         throw new RangeError("Array limit exceeded. Only " + arrayLimit + " element" + (arrayLimit === 1 ? "" : "s") + " allowed in an array.");
       }
-      var newIndex = getMaxIndex(a) + 1;
-      a[newIndex] = b;
+      var bValues = isArray(b) ? b : [b];
+      var newIndex = getMaxIndex(a);
+      for (var i = 0;i < bValues.length; ++i) {
+        newIndex += 1;
+        a[newIndex] = bValues[i];
+      }
       setMaxIndex(a, newIndex);
       return a;
     }
@@ -26300,6 +26451,7 @@ var require_stringify = __commonJS(function(exports, module) {
     charsetSentinel: false,
     commaRoundTrip: false,
     delimiter: "&",
+    depth: Infinity,
     encode: true,
     encodeDotInKeys: false,
     encoder: utils.encode,
@@ -26318,8 +26470,11 @@ var require_stringify = __commonJS(function(exports, module) {
     return typeof v === "string" || typeof v === "number" || typeof v === "boolean" || typeof v === "symbol" || typeof v === "bigint";
   };
   var sentinel = {};
-  var stringify = function stringify2(object2, prefix, generateArrayPrefix, commaRoundTrip, allowEmptyArrays, strictNullHandling, skipNulls, encodeDotInKeys, encoder, filter, sort, allowDots, serializeDate, format, formatter, encodeValuesOnly, charset, sideChannel) {
+  var stringify = function stringify2(object2, prefix, generateArrayPrefix, commaRoundTrip, allowEmptyArrays, strictNullHandling, skipNulls, encodeDotInKeys, encoder, filter, sort, allowDots, serializeDate, format, formatter, encodeValuesOnly, charset, sideChannel, depth, currentDepth) {
     var obj = object2;
+    if (currentDepth > depth) {
+      throw new RangeError("Input depth exceeded depth option of " + depth);
+    }
     var tmpSc = sideChannel;
     var step = 0;
     var findFlag = false;
@@ -26337,9 +26492,8 @@ var require_stringify = __commonJS(function(exports, module) {
         step = 0;
       }
     }
-    if (typeof filter === "function") {
-      obj = filter(prefix, obj);
-    } else if (obj instanceof Date) {
+    obj = typeof filter === "function" ? filter(prefix, obj) : obj;
+    if (obj instanceof Date) {
       obj = serializeDate(obj);
     } else if (generateArrayPrefix === "comma" && isArray(obj)) {
       obj = utils.maybeMap(obj, function(value2) {
@@ -26382,7 +26536,7 @@ var require_stringify = __commonJS(function(exports, module) {
     }
     var encodedPrefix = encodeDotInKeys ? String(prefix).replace(/\./g, "%2E") : String(prefix);
     var adjustedPrefix = commaRoundTrip && isArray(obj) && obj.length === 1 ? encodedPrefix + "[]" : encodedPrefix;
-    if (allowEmptyArrays && isArray(obj) && obj.length === 0) {
+    if (allowEmptyArrays && isArray(obj) && obj.length === 0 && Object.keys(obj).length === 0) {
       return adjustedPrefix + "[]";
     }
     for (var j = 0;j < objKeys.length; ++j) {
@@ -26396,7 +26550,7 @@ var require_stringify = __commonJS(function(exports, module) {
       sideChannel.set(object2, step);
       var valueSideChannel = getSideChannel();
       valueSideChannel.set(sentinel, sideChannel);
-      pushToArray(values, stringify2(value, keyPrefix, generateArrayPrefix, commaRoundTrip, allowEmptyArrays, strictNullHandling, skipNulls, encodeDotInKeys, generateArrayPrefix === "comma" && encodeValuesOnly && isArray(obj) ? null : encoder, filter, sort, allowDots, serializeDate, format, formatter, encodeValuesOnly, charset, valueSideChannel));
+      pushToArray(values, stringify2(value, keyPrefix, generateArrayPrefix, commaRoundTrip, allowEmptyArrays, strictNullHandling, skipNulls, encodeDotInKeys, generateArrayPrefix === "comma" && encodeValuesOnly && isArray(obj) ? null : encoder, filter, sort, allowDots, serializeDate, format, formatter, encodeValuesOnly, charset, valueSideChannel, depth, currentDepth + 1));
     }
     return values;
   };
@@ -26450,6 +26604,7 @@ var require_stringify = __commonJS(function(exports, module) {
       charsetSentinel: typeof opts.charsetSentinel === "boolean" ? opts.charsetSentinel : defaults.charsetSentinel,
       commaRoundTrip: !!opts.commaRoundTrip,
       delimiter: typeof opts.delimiter === "undefined" ? defaults.delimiter : opts.delimiter,
+      depth: typeof opts.depth === "number" ? opts.depth : defaults.depth,
       encode: typeof opts.encode === "boolean" ? opts.encode : defaults.encode,
       encodeDotInKeys: typeof opts.encodeDotInKeys === "boolean" ? opts.encodeDotInKeys : defaults.encodeDotInKeys,
       encoder: typeof opts.encoder === "function" ? opts.encoder : defaults.encoder,
@@ -26497,7 +26652,8 @@ var require_stringify = __commonJS(function(exports, module) {
       if (options.skipNulls && value === null) {
         continue;
       }
-      pushToArray(keys, stringify(value, key, generateArrayPrefix, commaRoundTrip, options.allowEmptyArrays, options.strictNullHandling, options.skipNulls, options.encodeDotInKeys, options.encode ? options.encoder : null, options.filter, options.sort, options.allowDots, options.serializeDate, options.format, options.formatter, options.encodeValuesOnly, options.charset, sideChannel));
+      var encodedKey = options.encodeDotInKeys ? String(key).replace(/\./g, "%2E") : String(key);
+      pushToArray(keys, stringify(value, encodedKey, generateArrayPrefix, commaRoundTrip, options.allowEmptyArrays, options.strictNullHandling, options.skipNulls, options.encodeDotInKeys, options.encode ? options.encoder : null, options.filter, options.sort, options.allowDots, options.serializeDate, options.format, options.formatter, options.encodeValuesOnly, options.charset, sideChannel, options.depth, 0));
     }
     var joined = keys.join(options.delimiter);
     var prefix = options.addQueryPrefix === true ? "?" : "";
@@ -26546,9 +26702,9 @@ var require_parse = __commonJS(function(exports, module) {
       return String.fromCharCode(parseInt(numberStr, 10));
     });
   };
-  var parseArrayValue = function(val, options, currentArrayLength, isFlatArrayValue) {
+  var parseArrayValue = function(val, options, currentArrayLength) {
     if (val && typeof val === "string" && options.comma && val.indexOf(",") > -1) {
-      if (isFlatArrayValue && options.throwOnLimitExceeded) {
+      if (options.throwOnLimitExceeded) {
         var commaCount = 0;
         var commaIndex = val.indexOf(",");
         while (commaIndex > -1) {
@@ -26608,7 +26764,7 @@ var require_parse = __commonJS(function(exports, module) {
       } else {
         key = options.decoder(part.slice(0, pos), defaults.decoder, charset, "key");
         if (key !== null) {
-          val = utils.maybeMap(parseArrayValue(part.slice(pos + 1), options, isArray(obj[key]) ? obj[key].length : 0, part.indexOf("[]=") === -1), function(encodedVal) {
+          val = utils.maybeMap(parseArrayValue(part.slice(pos + 1), options, isArray(obj[key]) ? obj[key].length : 0), function(encodedVal) {
             return options.decoder(encodedVal, defaults.decoder, charset, "value");
           });
         }
@@ -52864,9 +53020,9 @@ var init_config = __esm(() => {
   SDK_METADATA = {
     language: "typescript",
     openapiDocVersion: "0.1.0",
-    sdkVersion: "0.0.47",
+    sdkVersion: "0.0.48",
     genVersion: "2.915.1",
-    userAgent: "speakeasy-sdk/mcp-typescript 0.0.47 2.915.1 0.1.0 @dalmia/calibrate-mcp"
+    userAgent: "speakeasy-sdk/mcp-typescript 0.0.48 2.915.1 0.1.0 @dalmia/calibrate-mcp"
   };
 });
 
@@ -56062,7 +56218,8 @@ var init_testcaseresult = __esm(() => {
     passed: boolean2().nullable().optional().describe("Whether the case passed"),
     reasoning: string2().nullable().optional().describe("The judge's reasoning, or the tool-call diff for a tool-call test"),
     test_case: record(string2(), any()).nullable().optional().describe("The test case definition that was run"),
-    test_case_id: string2().nullable().optional().describe("ID of the test case within the run")
+    test_case_id: string2().nullable().optional().describe("ID of the test case within the run"),
+    unanswered: boolean2().default(false).describe("Whether this case produced no answer because the agent or the judge could not be reached, in which case `reasoning` carries the error and `passed` is not a verdict on the agent")
   });
 });
 
@@ -56297,10 +56454,12 @@ var init_testrunstatusresponse = __esm(() => {
     results: array(TestCaseResult$zodSchema).nullable().optional().describe("Results for each test case"),
     share_token: string2().nullable().optional().describe("Token for building the public share URL"),
     status: TaskStatus$zodSchema,
+    stopped_early: boolean2().default(false).describe("Whether the run stopped before starting every test case, after too many failed in a row"),
     task_id: string2().describe("Test run job ID"),
     test_uuids: array(string2()).nullable().optional().describe("IDs of the tests this run executed, in run order"),
     total_tests: int().nullable().optional().describe("Total number of test cases"),
-    total_tokens: record(string2(), any()).nullable().optional().describe("Aggregated token usage as `{mean, min, max, count}`")
+    total_tokens: record(string2(), any()).nullable().optional().describe("Aggregated token usage as `{mean, min, max, count}`"),
+    unanswered_tests: int().nullable().optional().describe("Number of test cases that produced no answer because the agent or the judge could not be reached, which makes the pass rate an unfair measure of the agent")
   });
 });
 
@@ -56822,7 +56981,7 @@ var init_testruncasesummary = __esm(() => {
   init_zod();
   TestRunCaseSummary$zodSchema = object({
     name: string2().nullable().optional().describe("Name of the test case"),
-    passed: boolean2().nullable().optional().describe("Whether the case passed (null if it errored or is still running)")
+    passed: boolean2().nullable().optional().describe("Whether the case passed (null while the case is still running)")
   }).describe(`Flat summary for one test case in the run-LIST endpoints. Carries only
 enough to render a run's pass/fail breakdown and a case name. The full detail
 for each case (agent output, judge verdicts, reasoning, latency, cost, the
@@ -56859,6 +57018,7 @@ var init_agenttestrunlistitem = __esm(() => {
     total_tests: int().nullable().optional().describe("Total number of test cases"),
     total_tokens: record(string2(), any()).nullable().optional().describe("Aggregated token usage as `{mean, min, max, count}`"),
     type: AgentTestRunListItemType$zodSchema.describe("What kind of run this is:\n- `llm-unit-test`: a single run of the agent's tests\n- `llm-benchmark`: a multi-model comparison"),
+    unanswered_tests: int().nullable().optional().describe("Number of test cases that produced no answer because the agent or the judge could not be reached, which makes the pass rate an unfair measure of the agent"),
     updated_at: string2().describe("When the run was last updated (ISO 8601 UTC)"),
     uuid: string2().describe("Test run job ID")
   });
@@ -61482,7 +61642,7 @@ hits its trace limit.
 function createMCPServer(deps) {
   const server = new McpServer({
     name: "CalibrateMcp",
-    version: "0.0.47"
+    version: "0.0.48"
   });
   const getClient = deps.getSDK || (() => new CalibrateMcpCore({
     security: deps.security,
@@ -62760,7 +62920,7 @@ http_headers = { "api-key-auth" = "YOUR_API_KEY_AUTH" }`;
         <h1>Instructions</h1>
         <p>One-click installation for Claude Desktop users</p>
         <div class="instruction-item">
-          <a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.47/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
+          <a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.48/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
             \uD83D\uDCE5 Download MCP Bundle
           </a>
         </div>
@@ -65647,7 +65807,7 @@ var routes = buildRouteMap({
 var app = buildApplication(routes, {
   name: "mcp",
   versionInfo: {
-    currentVersion: "0.0.47"
+    currentVersion: "0.0.48"
   }
 });
 run(app, process4.argv.slice(2), buildContext(process4));
@@ -65655,5 +65815,5 @@ export {
   app
 };
 
-//# debugId=D6CE9DB41F5C613964756E2164756E21
+//# debugId=070CB4FCEC8F307264756E2164756E21
 //# sourceMappingURL=mcp-server.js.map