@agentmark-ai/shared-utils 0.3.2 → 0.4.0

package/dist/index.d.mts

@@ -558,4 +558,43 @@ declare function normalizeSpan(resource: OtelResource, scope: OtelScope, span: O
  */
 declare function normalizeOtlpSpans(resourceSpans: OtlpResourceSpans[]): NormalizedSpan[];
 
-export { AGENTMARK_SCOPE_NAME, AgentMarkTransformer, type AgentmarkConfig, type AgentmarkModelConfig, type AgentmarkModelSchema, type AgentmarkModelSettingsConfig, type AgentmarkModelSettingsSchema, AiSdkTransformer, type AiSdkVersion, type AttributeExtractor, AgentMarkTransformer as ClaudeAgentTransformer, type GenerateTypesLanguage, MastraTransformer, type McpServerConfig, type McpServers, type McpStdioServerConfig, type McpUrlServerConfig, type Message, type ModelSettingsTypeAspectRatio, type ModelSettingsTypeImageSize, type ModelSettingsTypeSelect, type ModelSettingsTypeSlider, type NormalizedSpan, type OtelEvent, OtelGenAiTransformer, type OtelLink, type OtelResource, type OtelScope, type OtelSpan, type OtlpAttribute, type OtlpAttributeValue, type OtlpEvent, type OtlpLink, type OtlpResource, type OtlpResourceSpans, type OtlpScope, type OtlpScopeSpans, type OtlpSpan, SEMANTIC_KINDS, type ScopeTransformer, type SemanticKind, SpanType, type StandardMessageContent, type StandardTextContent, type StandardToolCallContent, type StandardToolResultContent, type ToolCall, TransformerRegistry, TypeClassifier, convertOtlpAttributes, createSignature, detectVersion, extractCustomMetadata, extractReasoningFromProviderMetadata, extractResourceScopeSpan, fetchPromptsFrontmatter, findPromptFiles, generateTypeDefinitions, generateUnique8CharString, normalizeOtlpSpans, normalizeSpan, parseAgentMarkAttributes, parseMetadata, parseTokens, registry, resolveSemanticKind, toFrontMatter, typeClassifier, verifySignature };
+/**
+ * Compute a stable identifier for a dataset row, derived from its input.
+ *
+ * The identifier survives row reordering, additions, and deletions in the
+ * dataset — anything that doesn't change the row's *content* keeps the same
+ * id. This is the property regression-vs-baseline comparisons need: the
+ * same logical test case in two different runs must produce the same name
+ * so the baseline lookup can match them.
+ *
+ * Format: first 12 hex characters of the MD5 digest of the row's input,
+ * canonicalised with sorted keys at every level. This matches the Python
+ * runner's implementation byte-for-byte so cross-runtime comparisons
+ * (TS-emitted baseline vs. Python-emitted PR run, or vice versa) produce
+ * matching identifiers.
+ *
+ * Falls back to the positional index as a string when the row has no
+ * input — in that case there is nothing stable to hash, so positional is
+ * the best we can do.
+ *
+ * Note: we use MD5 here for compatibility with the existing Python
+ * implementation. MD5 is not used as a security primitive — it's used as a
+ * non-cryptographic content fingerprint, which is its appropriate role.
+ */
+declare function computeDatasetItemName(input: unknown, fallbackIndex: number): string;
+/**
+ * Stringify a value as canonical JSON with object keys sorted recursively.
+ *
+ * Differs from `JSON.stringify`:
+ * - Object keys are sorted lexicographically at every level of nesting, so
+ *   `{a: 1, b: 2}` and `{b: 2, a: 1}` produce the same output.
+ * - Non-serializable values (functions, symbols, bigints, undefined) are
+ *   coerced to strings via `String(value)`, matching Python's `default=str`
+ *   behavior in `json.dumps`.
+ *
+ * Exported for testing and for callers that need the same canonicalisation
+ * for purposes other than item-name hashing.
+ */
+declare function canonicalJsonStringify(value: unknown): string;
+
+export { AGENTMARK_SCOPE_NAME, AgentMarkTransformer, type AgentmarkConfig, type AgentmarkModelConfig, type AgentmarkModelSchema, type AgentmarkModelSettingsConfig, type AgentmarkModelSettingsSchema, AiSdkTransformer, type AiSdkVersion, type AttributeExtractor, AgentMarkTransformer as ClaudeAgentTransformer, type GenerateTypesLanguage, MastraTransformer, type McpServerConfig, type McpServers, type McpStdioServerConfig, type McpUrlServerConfig, type Message, type ModelSettingsTypeAspectRatio, type ModelSettingsTypeImageSize, type ModelSettingsTypeSelect, type ModelSettingsTypeSlider, type NormalizedSpan, type OtelEvent, OtelGenAiTransformer, type OtelLink, type OtelResource, type OtelScope, type OtelSpan, type OtlpAttribute, type OtlpAttributeValue, type OtlpEvent, type OtlpLink, type OtlpResource, type OtlpResourceSpans, type OtlpScope, type OtlpScopeSpans, type OtlpSpan, SEMANTIC_KINDS, type ScopeTransformer, type SemanticKind, SpanType, type StandardMessageContent, type StandardTextContent, type StandardToolCallContent, type StandardToolResultContent, type ToolCall, TransformerRegistry, TypeClassifier, canonicalJsonStringify, computeDatasetItemName, convertOtlpAttributes, createSignature, detectVersion, extractCustomMetadata, extractReasoningFromProviderMetadata, extractResourceScopeSpan, fetchPromptsFrontmatter, findPromptFiles, generateTypeDefinitions, generateUnique8CharString, normalizeOtlpSpans, normalizeSpan, parseAgentMarkAttributes, parseMetadata, parseTokens, registry, resolveSemanticKind, toFrontMatter, typeClassifier, verifySignature };

package/dist/index.d.ts

@@ -558,4 +558,43 @@ declare function normalizeSpan(resource: OtelResource, scope: OtelScope, span: O
  */
 declare function normalizeOtlpSpans(resourceSpans: OtlpResourceSpans[]): NormalizedSpan[];
 
-export { AGENTMARK_SCOPE_NAME, AgentMarkTransformer, type AgentmarkConfig, type AgentmarkModelConfig, type AgentmarkModelSchema, type AgentmarkModelSettingsConfig, type AgentmarkModelSettingsSchema, AiSdkTransformer, type AiSdkVersion, type AttributeExtractor, AgentMarkTransformer as ClaudeAgentTransformer, type GenerateTypesLanguage, MastraTransformer, type McpServerConfig, type McpServers, type McpStdioServerConfig, type McpUrlServerConfig, type Message, type ModelSettingsTypeAspectRatio, type ModelSettingsTypeImageSize, type ModelSettingsTypeSelect, type ModelSettingsTypeSlider, type NormalizedSpan, type OtelEvent, OtelGenAiTransformer, type OtelLink, type OtelResource, type OtelScope, type OtelSpan, type OtlpAttribute, type OtlpAttributeValue, type OtlpEvent, type OtlpLink, type OtlpResource, type OtlpResourceSpans, type OtlpScope, type OtlpScopeSpans, type OtlpSpan, SEMANTIC_KINDS, type ScopeTransformer, type SemanticKind, SpanType, type StandardMessageContent, type StandardTextContent, type StandardToolCallContent, type StandardToolResultContent, type ToolCall, TransformerRegistry, TypeClassifier, convertOtlpAttributes, createSignature, detectVersion, extractCustomMetadata, extractReasoningFromProviderMetadata, extractResourceScopeSpan, fetchPromptsFrontmatter, findPromptFiles, generateTypeDefinitions, generateUnique8CharString, normalizeOtlpSpans, normalizeSpan, parseAgentMarkAttributes, parseMetadata, parseTokens, registry, resolveSemanticKind, toFrontMatter, typeClassifier, verifySignature };
+/**
+ * Compute a stable identifier for a dataset row, derived from its input.
+ *
+ * The identifier survives row reordering, additions, and deletions in the
+ * dataset — anything that doesn't change the row's *content* keeps the same
+ * id. This is the property regression-vs-baseline comparisons need: the
+ * same logical test case in two different runs must produce the same name
+ * so the baseline lookup can match them.
+ *
+ * Format: first 12 hex characters of the MD5 digest of the row's input,
+ * canonicalised with sorted keys at every level. This matches the Python
+ * runner's implementation byte-for-byte so cross-runtime comparisons
+ * (TS-emitted baseline vs. Python-emitted PR run, or vice versa) produce
+ * matching identifiers.
+ *
+ * Falls back to the positional index as a string when the row has no
+ * input — in that case there is nothing stable to hash, so positional is
+ * the best we can do.
+ *
+ * Note: we use MD5 here for compatibility with the existing Python
+ * implementation. MD5 is not used as a security primitive — it's used as a
+ * non-cryptographic content fingerprint, which is its appropriate role.
+ */
+declare function computeDatasetItemName(input: unknown, fallbackIndex: number): string;
+/**
+ * Stringify a value as canonical JSON with object keys sorted recursively.
+ *
+ * Differs from `JSON.stringify`:
+ * - Object keys are sorted lexicographically at every level of nesting, so
+ *   `{a: 1, b: 2}` and `{b: 2, a: 1}` produce the same output.
+ * - Non-serializable values (functions, symbols, bigints, undefined) are
+ *   coerced to strings via `String(value)`, matching Python's `default=str`
+ *   behavior in `json.dumps`.
+ *
+ * Exported for testing and for callers that need the same canonicalisation
+ * for purposes other than item-name hashing.
+ */
+declare function canonicalJsonStringify(value: unknown): string;
+
+export { AGENTMARK_SCOPE_NAME, AgentMarkTransformer, type AgentmarkConfig, type AgentmarkModelConfig, type AgentmarkModelSchema, type AgentmarkModelSettingsConfig, type AgentmarkModelSettingsSchema, AiSdkTransformer, type AiSdkVersion, type AttributeExtractor, AgentMarkTransformer as ClaudeAgentTransformer, type GenerateTypesLanguage, MastraTransformer, type McpServerConfig, type McpServers, type McpStdioServerConfig, type McpUrlServerConfig, type Message, type ModelSettingsTypeAspectRatio, type ModelSettingsTypeImageSize, type ModelSettingsTypeSelect, type ModelSettingsTypeSlider, type NormalizedSpan, type OtelEvent, OtelGenAiTransformer, type OtelLink, type OtelResource, type OtelScope, type OtelSpan, type OtlpAttribute, type OtlpAttributeValue, type OtlpEvent, type OtlpLink, type OtlpResource, type OtlpResourceSpans, type OtlpScope, type OtlpScopeSpans, type OtlpSpan, SEMANTIC_KINDS, type ScopeTransformer, type SemanticKind, SpanType, type StandardMessageContent, type StandardTextContent, type StandardToolCallContent, type StandardToolResultContent, type ToolCall, TransformerRegistry, TypeClassifier, canonicalJsonStringify, computeDatasetItemName, convertOtlpAttributes, createSignature, detectVersion, extractCustomMetadata, extractReasoningFromProviderMetadata, extractResourceScopeSpan, fetchPromptsFrontmatter, findPromptFiles, generateTypeDefinitions, generateUnique8CharString, normalizeOtlpSpans, normalizeSpan, parseAgentMarkAttributes, parseMetadata, parseTokens, registry, resolveSemanticKind, toFrontMatter, typeClassifier, verifySignature };

package/dist/index.js

@@ -40,6 +40,8 @@ __export(index_exports, {
   SpanType: () => SpanType,
   TransformerRegistry: () => TransformerRegistry,
   TypeClassifier: () => TypeClassifier,
+  canonicalJsonStringify: () => canonicalJsonStringify,
+  computeDatasetItemName: () => computeDatasetItemName,
   convertOtlpAttributes: () => convertOtlpAttributes,
   createSignature: () => createSignature,
   detectVersion: () => detectVersion,
@@ -228,27 +230,26 @@ async function generateTypeDefinitionsV1_0(prompts) {
   for (const prompt of prompts) {
     const { path: promptPath, input_schema } = prompt;
     const name = getInterfaceName(promptPath);
-    let tools = {};
     try {
       let kind = "text";
       let output_schema = null;
+      let hasTools = false;
       if ("text_config" in prompt) {
         kind = "text";
-        tools = prompt.text_config.tools || {};
+        hasTools = Array.isArray(prompt.text_config.tools) ? prompt.text_config.tools.length > 0 : !!prompt.text_config.tools;
       } else if ("object_config" in prompt) {
         kind = "object";
-        tools = prompt.object_config.tools || {};
+        hasTools = Array.isArray(prompt.object_config.tools) ? prompt.object_config.tools.length > 0 : !!prompt.object_config.tools;
         output_schema = prompt.object_config.schema;
       } else if ("image_config" in prompt) {
         kind = "image";
-        tools = prompt.image_config.tools || {};
+        hasTools = Array.isArray(prompt.image_config.tools) ? prompt.image_config.tools.length > 0 : !!prompt.image_config.tools;
       }
       const compile = await getCompile();
       const inputInterface = input_schema ? await compile(input_schema, `${name}In`, {
         bannerComment: "",
         additionalProperties: false
       }) : `interface ${name}In { [key: string]: any }`;
-      const toolTypes = await generateToolTypes(tools);
       const outputInterface = output_schema ? await compile(output_schema, `${name}Out`, {
         bannerComment: "",
         additionalProperties: false
@@ -257,12 +258,11 @@ async function generateTypeDefinitionsV1_0(prompts) {
         inputInterface.replace("export interface", "interface"),
         outputInterface.replace("export type", "type").replace("export interface", "interface")
       );
-      output += toolTypes || "";
       output += `type ${name} = {
   kind: '${kind}';
   input:  ${name}In;
-  output: ${name}Out;${toolTypes ? `
-  tools?: Array<keyof Tools>;` : ""}
+  output: ${name}Out;${hasTools ? `
+  tools?: string[];` : ""}
 };
 
 `;
@@ -689,6 +689,7 @@ async function generateTypeDefinitions(prompts, language = "typescript") {
   return generateTypeDefinitionsV0(prompts);
 }
 async function fetchPromptsFrontmatter(options) {
+  var _a, _b, _c;
   if (options.local) {
     const baseUrl = `http://localhost:${options.local}`;
     try {
@@ -698,7 +699,8 @@ async function fetchPromptsFrontmatter(options) {
           `Failed to fetch prompt paths: ${pathsResponse.statusText}`
         );
       }
-      const { paths } = await pathsResponse.json();
+      const body = await pathsResponse.json();
+      const paths = (_c = (_b = (_a = body == null ? void 0 : body.data) == null ? void 0 : _a.paths) != null ? _b : body == null ? void 0 : body.paths) != null ? _c : [];
       return Promise.all(
         paths.map(async (promptPath) => {
           const templateResponse = await fetch(
@@ -764,34 +766,6 @@ async function fetchPromptsFrontmatter(options) {
   }
   throw new Error("Either --local or --root-dir must be specified");
 }
-async function generateToolTypes(tools) {
-  const toolArgTypes = [];
-  for (const [toolName, schema] of Object.entries(tools)) {
-    const typeName = `${getToolInterfaceName(toolName)}Args`;
-    try {
-      const compile = await getCompile();
-      const argInterface = schema.parameters ? await compile(schema.parameters, typeName, {
-        bannerComment: "",
-        additionalProperties: false
-      }) : `type ${typeName} = { ${Object.entries(schema.parameters || {}).map(([key]) => `${key}: any`).join("; ")} };`;
-      toolArgTypes.push(
-        argInterface.replace("export type", "type").replace("export interface", "interface")
-      );
-    } catch (error) {
-      console.error(`Error processing tool ${toolName}:`, error);
-      toolArgTypes.push(`type ${typeName} = { [key: string]: any };`);
-    }
-  }
-  if (Object.keys(tools).length > 0) {
-    const toolsInterface = `export interface Tools {
-${Object.keys(tools).map(
-      (toolName) => `  ${toolName}: { args: ${getToolInterfaceName(toolName)}Args };`
-    ).join("\n")}
-}`;
-    return toolArgTypes.join("\n\n") + "\n\n" + toolsInterface + "\n\n";
-  }
-  return null;
-}
 
 // src/normalizer/types.ts
 var SpanType = /* @__PURE__ */ ((SpanType2) => {
@@ -1778,6 +1752,7 @@ var AgentMarkTransformer = class {
    * Transform the span and extract normalized fields from GenAI attributes.
    */
   transform(_span, attributes) {
+    var _a;
     const result = {};
     const responseModel = attributes[GenAIAttributes.RESPONSE_MODEL];
     const requestModel = attributes[GenAIAttributes.REQUEST_MODEL];
@@ -1851,7 +1826,7 @@ var AgentMarkTransformer = class {
       } catch {
       }
     }
-    const amInput = attributes["agentmark.input"];
+    const amInput = (_a = attributes["agentmark.input"]) != null ? _a : attributes["agentmark.props"];
     if (amInput && typeof amInput === "string" && !result.input) {
       result.input = [{ role: "user", content: amInput }];
     }
@@ -1917,9 +1892,16 @@ var Attrs = {
   RESPONSE_FINISH_REASONS: "gen_ai.response.finish_reasons",
   USAGE_INPUT_TOKENS: "gen_ai.usage.input_tokens",
   USAGE_OUTPUT_TOKENS: "gen_ai.usage.output_tokens",
-  // v1.37.0+ content attributes
+  // v1.37.0+ content attributes (canonical OTel GenAI semantic conventions).
+  // SDKs that emit the AgentMark-scoped equivalents (`gen_ai.request.input`
+  // / `gen_ai.response.output` — used by `claude-agent-sdk-v0-adapter`) are
+  // also accepted here as fallbacks so an SDK picking either key set
+  // doesn't silently lose IO data on ingest. The canonical pair always wins
+  // when both are present.
   INPUT_MESSAGES: "gen_ai.input.messages",
   OUTPUT_MESSAGES: "gen_ai.output.messages",
+  REQUEST_INPUT_FALLBACK: "gen_ai.request.input",
+  RESPONSE_OUTPUT_FALLBACK: "gen_ai.response.output",
   SYSTEM_INSTRUCTIONS: "gen_ai.system_instructions",
   TOOL_DEFINITIONS: "gen_ai.tool.definitions",
   // Tool call attributes (v1.37.0+)
@@ -1984,6 +1966,7 @@ var OtelGenAiTransformer = class {
     return "SPAN" /* SPAN */;
   }
   transform(span, attributes) {
+    var _a, _b;
     const result = {};
     const model = attributes[Attrs.RESPONSE_MODEL] || attributes[Attrs.REQUEST_MODEL];
     if (model && typeof model === "string") {
@@ -2004,12 +1987,12 @@ var OtelGenAiTransformer = class {
     if (typeof temperature === "number") {
       result.settings = { ...result.settings, temperature };
     }
-    const inputMessages = attributes[Attrs.INPUT_MESSAGES];
+    const inputMessages = (_a = attributes[Attrs.INPUT_MESSAGES]) != null ? _a : attributes[Attrs.REQUEST_INPUT_FALLBACK];
     if (inputMessages && typeof inputMessages === "string") {
       const messages = normalizeMessages(inputMessages);
       if (messages) result.input = messages;
     }
-    const outputMessages = attributes[Attrs.OUTPUT_MESSAGES];
+    const outputMessages = (_b = attributes[Attrs.OUTPUT_MESSAGES]) != null ? _b : attributes[Attrs.RESPONSE_OUTPUT_FALLBACK];
     if (outputMessages && typeof outputMessages === "string") {
       const extracted = extractStructuredOutput(outputMessages);
       if (extracted) {
@@ -2339,6 +2322,36 @@ function normalizeOtlpSpans(resourceSpans) {
   }
   return normalizedSpans;
 }
+
+// src/dataset-item-name.ts
+var import_node_crypto = require("crypto");
+function computeDatasetItemName(input, fallbackIndex) {
+  if (input === void 0 || input === null) {
+    return String(fallbackIndex);
+  }
+  const canonical = canonicalJsonStringify(input);
+  return (0, import_node_crypto.createHash)("md5").update(canonical).digest("hex").slice(0, 12);
+}
+function canonicalJsonStringify(value) {
+  if (value === null) return "null";
+  if (value === void 0) return JSON.stringify(String(void 0));
+  const t = typeof value;
+  if (t === "string" || t === "number" || t === "boolean") {
+    return JSON.stringify(value);
+  }
+  if (t === "bigint" || t === "function" || t === "symbol") {
+    return JSON.stringify(String(value));
+  }
+  if (Array.isArray(value)) {
+    return "[" + value.map(canonicalJsonStringify).join(",") + "]";
+  }
+  const obj = value;
+  const keys = Object.keys(obj).sort();
+  const parts = keys.map((k) => {
+    return JSON.stringify(k) + ":" + canonicalJsonStringify(obj[k]);
+  });
+  return "{" + parts.join(",") + "}";
+}
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
   AGENTMARK_SCOPE_NAME,
@@ -2351,6 +2364,8 @@ function normalizeOtlpSpans(resourceSpans) {
   SpanType,
   TransformerRegistry,
   TypeClassifier,
+  canonicalJsonStringify,
+  computeDatasetItemName,
   convertOtlpAttributes,
   createSignature,
   detectVersion,