@langwatch/scenario 0.4.3 → 0.4.5

package/dist/index.js

@@ -969,6 +969,7 @@ __export(index_exports, {
   AgentAdapter: () => AgentAdapter,
   AgentRole: () => AgentRole,
   DEFAULT_MAX_TURNS: () => DEFAULT_MAX_TURNS,
+  DEFAULT_TOKEN_THRESHOLD: () => DEFAULT_TOKEN_THRESHOLD,
   DEFAULT_VERBOSE: () => DEFAULT_VERBOSE,
   JudgeAgentAdapter: () => JudgeAgentAdapter,
   JudgeSpanCollector: () => JudgeSpanCollector,
@@ -982,7 +983,10 @@ __export(index_exports, {
   allAgentRoles: () => allAgentRoles,
   default: () => index_default,
   defineConfig: () => defineConfig,
+  estimateTokens: () => estimateTokens,
+  expandTrace: () => expandTrace,
   fail: () => fail,
+  grepTrace: () => grepTrace,
   judge: () => judge,
   judgeAgent: () => judgeAgent,
   judgeSpanCollector: () => judgeSpanCollector,
@@ -1004,9 +1008,13 @@ module.exports = __toCommonJS(index_exports);
 // src/agents/index.ts
 var agents_exports = {};
 __export(agents_exports, {
+  DEFAULT_TOKEN_THRESHOLD: () => DEFAULT_TOKEN_THRESHOLD,
   JudgeSpanCollector: () => JudgeSpanCollector,
   JudgeSpanDigestFormatter: () => JudgeSpanDigestFormatter,
   RealtimeAgentAdapter: () => RealtimeAgentAdapter,
+  estimateTokens: () => estimateTokens,
+  expandTrace: () => expandTrace,
+  grepTrace: () => grepTrace,
   judgeAgent: () => judgeAgent,
   judgeSpanCollector: () => judgeSpanCollector,
   judgeSpanDigestFormatter: () => judgeSpanDigestFormatter,
@@ -1086,6 +1094,275 @@ var JudgeUtils = {
   }
 };
 
+// src/agents/judge/estimate-tokens.ts
+var DEFAULT_TOKEN_THRESHOLD = 8192;
+function estimateTokens(text) {
+  const byteLength = new TextEncoder().encode(text).byteLength;
+  return Math.ceil(byteLength / 4);
+}
+
+// src/agents/judge/span-utils.ts
+var import_observability = require("langwatch/observability");
+
+// src/agents/judge/deep-transform.ts
+function deepTransform(value, fn) {
+  const result = fn(value);
+  if (result !== value) return result;
+  if (Array.isArray(value)) {
+    return value.map((v) => deepTransform(v, fn));
+  }
+  if (value !== null && typeof value === "object") {
+    const out = {};
+    for (const [k, v] of Object.entries(value)) {
+      out[k] = deepTransform(v, fn);
+    }
+    return out;
+  }
+  return value;
+}
+
+// src/agents/judge/truncate-media.ts
+function truncateMediaUrl(str) {
+  const match = str.match(
+    /^data:((image|audio|video)\/[a-z0-9+.-]+);base64,(.+)$/i
+  );
+  if (!match) return str;
+  const [, mimeType, category, data] = match;
+  return `[${category.toUpperCase()}: ${mimeType}, ~${data.length} bytes]`;
+}
+function truncateMediaPart(v) {
+  var _a;
+  if (v === null || typeof v !== "object" || Array.isArray(v)) return null;
+  const obj = v;
+  if (obj.type === "file" && typeof obj.mediaType === "string" && typeof obj.data === "string") {
+    const category = ((_a = obj.mediaType.split("/")[0]) == null ? void 0 : _a.toUpperCase()) ?? "FILE";
+    return {
+      ...obj,
+      data: `[${category}: ${obj.mediaType}, ~${obj.data.length} bytes]`
+    };
+  }
+  if (obj.type === "image" && typeof obj.image === "string") {
+    const imageData = obj.image;
+    const dataUrlMatch = imageData.match(
+      /^data:((image)\/[a-z0-9+.-]+);base64,(.+)$/i
+    );
+    if (dataUrlMatch) {
+      return {
+        ...obj,
+        image: `[IMAGE: ${dataUrlMatch[1]}, ~${dataUrlMatch[3].length} bytes]`
+      };
+    }
+    if (imageData.length > 1e3 && /^[A-Za-z0-9+/=]+$/.test(imageData)) {
+      return {
+        ...obj,
+        image: `[IMAGE: unknown, ~${imageData.length} bytes]`
+      };
+    }
+  }
+  return null;
+}
+
+// src/agents/judge/span-utils.ts
+function hrTimeToMs(hrTime) {
+  return hrTime[0] * 1e3 + hrTime[1] / 1e6;
+}
+function formatDuration(ms) {
+  if (ms < 1e3) return `${Math.round(ms)}ms`;
+  return `${(ms / 1e3).toFixed(2)}s`;
+}
+function calculateSpanDuration(span) {
+  return hrTimeToMs(span.endTime) - hrTimeToMs(span.startTime);
+}
+function getStatusIndicator(span) {
+  if (span.status.code === 2) {
+    return ` \u26A0\uFE0F ERROR: ${span.status.message ?? "unknown"}`;
+  }
+  return "";
+}
+function getTokenUsage(span) {
+  const input = span.attributes["gen_ai.usage.input_tokens"];
+  const output = span.attributes["gen_ai.usage.output_tokens"];
+  if (input == null && output == null) return "";
+  const total = (Number(input) || 0) + (Number(output) || 0);
+  return `, ${total} tokens`;
+}
+function cleanAttributes(attrs) {
+  const cleaned = {};
+  const seen = /* @__PURE__ */ new Set();
+  const excludedKeys = [
+    import_observability.attributes.ATTR_LANGWATCH_THREAD_ID,
+    "langwatch.scenario.id",
+    "langwatch.scenario.name"
+  ];
+  for (const [key, value] of Object.entries(attrs)) {
+    if (excludedKeys.includes(key)) {
+      continue;
+    }
+    const cleanKey = key.replace(/^(langwatch)\./, "");
+    if (!seen.has(cleanKey)) {
+      seen.add(cleanKey);
+      cleaned[cleanKey] = value;
+    }
+  }
+  return cleaned;
+}
+function formatValue(value) {
+  const processed = transformValue(value);
+  return typeof processed === "string" ? processed : JSON.stringify(processed);
+}
+function transformValue(value) {
+  return deepTransform(value, (v) => {
+    const mediaPart = truncateMediaPart(v);
+    if (mediaPart) return mediaPart;
+    if (typeof v !== "string") return v;
+    const truncated = truncateMediaUrl(v);
+    if (truncated !== v) return truncated;
+    if (looksLikeJson(v)) {
+      try {
+        const parsed = transformValue(JSON.parse(v));
+        return JSON.stringify(parsed);
+      } catch {
+      }
+    }
+    return v;
+  });
+}
+function looksLikeJson(str) {
+  const t = str.trim();
+  return t.startsWith("{") && t.endsWith("}") || t.startsWith("[") && t.endsWith("]");
+}
+function indexSpans(spans) {
+  const sorted = [...spans].sort((a, b) => {
+    return hrTimeToMs(a.startTime) - hrTimeToMs(b.startTime);
+  });
+  return sorted.map((span) => ({
+    span,
+    children: [],
+    shortId: span.spanContext().spanId.slice(0, 8)
+  }));
+}
+
+// src/agents/judge/trace-tools.ts
+var TOOL_RESULT_TOKEN_BUDGET = 4096;
+var TOOL_RESULT_CHAR_BUDGET = TOOL_RESULT_TOKEN_BUDGET * 4;
+var MAX_GREP_MATCHES = 20;
+function renderFullSpanNode(node) {
+  const span = node.span;
+  const duration = calculateSpanDuration(span);
+  const timestamp = new Date(hrTimeToMs(span.startTime)).toISOString();
+  const status = getStatusIndicator(span);
+  const lines = [];
+  lines.push(
+    `[${node.shortId}] ${timestamp} ${span.name} (${formatDuration(duration)})${status}`
+  );
+  const attrs = cleanAttributes(span.attributes);
+  if (Object.keys(attrs).length > 0) {
+    for (const [key, value] of Object.entries(attrs)) {
+      lines.push(`    ${key}: ${formatValue(value)}`);
+    }
+  }
+  if (span.events.length > 0) {
+    for (const event of span.events) {
+      lines.push(`    [event] ${event.name}`);
+      if (event.attributes) {
+        const eventAttrs = cleanAttributes(event.attributes);
+        for (const [key, value] of Object.entries(eventAttrs)) {
+          lines.push(`      ${key}: ${formatValue(value)}`);
+        }
+      }
+    }
+  }
+  return lines;
+}
+function truncateToCharBudget(text) {
+  if (text.length <= TOOL_RESULT_CHAR_BUDGET) return text;
+  const truncated = text.slice(0, TOOL_RESULT_CHAR_BUDGET);
+  return truncated + "\n\n[TRUNCATED] Output exceeded ~4000 token budget. Use grep_trace(pattern) to search for specific content, or expand_trace with fewer span IDs.";
+}
+function spanToSearchableText(span) {
+  const parts = [span.name];
+  const attrs = cleanAttributes(span.attributes);
+  for (const [key, value] of Object.entries(attrs)) {
+    parts.push(`${key}: ${formatValue(value)}`);
+  }
+  for (const event of span.events) {
+    parts.push(event.name);
+    if (event.attributes) {
+      const eventAttrs = cleanAttributes(event.attributes);
+      for (const [key, value] of Object.entries(eventAttrs)) {
+        parts.push(`${key}: ${formatValue(value)}`);
+      }
+    }
+  }
+  return parts.join("\n");
+}
+function expandTrace(spans, spanIds) {
+  const nodes = indexSpans(spans);
+  if (nodes.length === 0) {
+    return "No spans recorded.";
+  }
+  if (spanIds.length === 0) {
+    return "Error: provide at least one span ID.";
+  }
+  const selected = nodes.filter((n) => {
+    const fullId = n.span.spanContext().spanId;
+    return spanIds.some((prefix) => fullId.startsWith(prefix));
+  });
+  if (selected.length === 0) {
+    const available = nodes.map((n) => n.shortId).join(", ");
+    return `Error: no spans matched the given ID(s). Available span IDs: ${available}`;
+  }
+  const lines = [];
+  for (const node of selected) {
+    const spanLines = renderFullSpanNode(node);
+    lines.push(...spanLines);
+    lines.push("");
+  }
+  return truncateToCharBudget(lines.join("\n").trimEnd());
+}
+function grepTrace(spans, pattern) {
+  const nodes = indexSpans(spans);
+  if (nodes.length === 0) {
+    return "No spans recorded.";
+  }
+  const regex = new RegExp(escapeRegex(pattern), "i");
+  const matches = [];
+  for (const node of nodes) {
+    const searchText = spanToSearchableText(node.span);
+    const lines2 = searchText.split("\n");
+    const matchingLines = lines2.filter((line) => regex.test(line));
+    if (matchingLines.length > 0) {
+      matches.push({ node, matchingLines });
+    }
+  }
+  if (matches.length === 0) {
+    const spanNames = Array.from(new Set(nodes.map((n) => n.span.name)));
+    return `No matches found for "${pattern}". Available span names: ${spanNames.join(", ")}`;
+  }
+  const totalMatches = matches.length;
+  const limited = matches.slice(0, MAX_GREP_MATCHES);
+  const lines = [];
+  for (const { node, matchingLines } of limited) {
+    const duration = calculateSpanDuration(node.span);
+    lines.push(
+      `--- [${node.shortId}] ${node.span.name} (${formatDuration(duration)}) ---`
+    );
+    for (const line of matchingLines) {
+      lines.push(`  ${line}`);
+    }
+    lines.push("");
+  }
+  if (totalMatches > MAX_GREP_MATCHES) {
+    lines.push(
+      `[${totalMatches - MAX_GREP_MATCHES} more matches omitted. Refine your search pattern for more specific results.]`
+    );
+  }
+  return truncateToCharBudget(lines.join("\n").trimEnd());
+}
+function escapeRegex(str) {
+  return str.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+}
+
 // src/config/env.ts
 var import_v4 = require("zod/v4");
 
@@ -1441,7 +1718,7 @@ var criterionToParamName = (criterion) => {
 };
 
 // src/agents/judge/judge-span-collector.ts
-var import_observability = require("langwatch/observability");
+var import_observability2 = require("langwatch/observability");
 var JudgeSpanCollector = class {
   spans = [];
   onStart() {
@@ -1484,7 +1761,7 @@ var JudgeSpanCollector = class {
       const spanId = span.spanContext().spanId;
       if (visited.has(spanId)) return false;
       visited.add(spanId);
-      if (span.attributes[import_observability.attributes.ATTR_LANGWATCH_THREAD_ID] === threadId) {
+      if (span.attributes[import_observability2.attributes.ATTR_LANGWATCH_THREAD_ID] === threadId) {
         return true;
       }
       const parentId = getParentSpanId(span);
@@ -1503,26 +1780,6 @@ function getParentSpanId(span) {
 }
 var judgeSpanCollector = new JudgeSpanCollector();
 
-// src/agents/judge/judge-span-digest-formatter.ts
-var import_observability2 = require("langwatch/observability");
-
-// src/agents/judge/deep-transform.ts
-function deepTransform(value, fn) {
-  const result = fn(value);
-  if (result !== value) return result;
-  if (Array.isArray(value)) {
-    return value.map((v) => deepTransform(v, fn));
-  }
-  if (value !== null && typeof value === "object") {
-    const out = {};
-    for (const [k, v] of Object.entries(value)) {
-      out[k] = deepTransform(v, fn);
-    }
-    return out;
-  }
-  return value;
-}
-
 // src/agents/judge/string-deduplicator.ts
 var StringDeduplicator = class {
   seen = /* @__PURE__ */ new Map();
@@ -1556,51 +1813,49 @@ var StringDeduplicator = class {
   }
 };
 
-// src/agents/judge/truncate-media.ts
-function truncateMediaUrl(str) {
-  const match = str.match(
-    /^data:((image|audio|video)\/[a-z0-9+.-]+);base64,(.+)$/i
-  );
-  if (!match) return str;
-  const [, mimeType, category, data] = match;
-  return `[${category.toUpperCase()}: ${mimeType}, ~${data.length} bytes]`;
-}
-function truncateMediaPart(v) {
-  var _a;
-  if (v === null || typeof v !== "object" || Array.isArray(v)) return null;
-  const obj = v;
-  if (obj.type === "file" && typeof obj.mediaType === "string" && typeof obj.data === "string") {
-    const category = ((_a = obj.mediaType.split("/")[0]) == null ? void 0 : _a.toUpperCase()) ?? "FILE";
-    return {
-      ...obj,
-      data: `[${category}: ${obj.mediaType}, ~${obj.data.length} bytes]`
-    };
-  }
-  if (obj.type === "image" && typeof obj.image === "string") {
-    const imageData = obj.image;
-    const dataUrlMatch = imageData.match(
-      /^data:((image)\/[a-z0-9+.-]+);base64,(.+)$/i
-    );
-    if (dataUrlMatch) {
-      return {
-        ...obj,
-        image: `[IMAGE: ${dataUrlMatch[1]}, ~${dataUrlMatch[3].length} bytes]`
-      };
-    }
-    if (imageData.length > 1e3 && /^[A-Za-z0-9+/=]+$/.test(imageData)) {
-      return {
-        ...obj,
-        image: `[IMAGE: unknown, ~${imageData.length} bytes]`
-      };
-    }
-  }
-  return null;
-}
-
 // src/agents/judge/judge-span-digest-formatter.ts
 var JudgeSpanDigestFormatter = class {
   logger = new Logger("JudgeSpanDigestFormatter");
   deduplicator = new StringDeduplicator({ threshold: 50 });
+  /**
+   * Formats spans into a structure-only digest showing span tree hierarchy
+   * without attributes, events, or content. Used for large traces that
+   * exceed the token threshold, paired with expand_trace/grep_trace tools.
+   *
+   * @param spans - All spans for a thread
+   * @returns Plain text digest with only structural information
+   */
+  formatStructureOnly(spans) {
+    this.logger.debug("formatStructureOnly() called", {
+      spanCount: spans.length
+    });
+    if (spans.length === 0) {
+      return "No spans recorded.";
+    }
+    const sortedSpans = this.sortByStartTime(spans);
+    const tree = this.buildHierarchy(sortedSpans);
+    const totalDuration = this.calculateTotalDuration(sortedSpans);
+    const lines = [
+      `Spans: ${spans.length} | Total Duration: ${formatDuration(totalDuration)}`,
+      ""
+    ];
+    const rootCount = tree.length;
+    tree.forEach((node, idx) => {
+      this.renderStructureNode(
+        node,
+        lines,
+        0,
+        idx === rootCount - 1
+      );
+    });
+    const errors = this.collectErrors(spans);
+    if (errors.length > 0) {
+      lines.push("");
+      lines.push("=== ERRORS ===");
+      errors.forEach((e) => lines.push(e));
+    }
+    return lines.join("\n");
+  }
   /**
    * Formats spans into a complete digest with full content and nesting.
    * @param spans - All spans for a thread
@@ -1624,19 +1879,17 @@ var JudgeSpanDigestFormatter = class {
       totalDuration
     });
     const lines = [
-      `Spans: ${spans.length} | Total Duration: ${this.formatDuration(
+      `Spans: ${spans.length} | Total Duration: ${formatDuration(
         totalDuration
       )}`,
       ""
     ];
-    let sequence = 1;
     const rootCount = tree.length;
     tree.forEach((node, idx) => {
-      sequence = this.renderNode(
+      this.renderNode(
         node,
         lines,
         0,
-        sequence,
         idx === rootCount - 1
       );
     });
@@ -1650,9 +1903,7 @@ var JudgeSpanDigestFormatter = class {
   }
   sortByStartTime(spans) {
     return [...spans].sort((a, b) => {
-      const aTime = this.hrTimeToMs(a.startTime);
-      const bTime = this.hrTimeToMs(b.startTime);
-      return aTime - bTime;
+      return hrTimeToMs(a.startTime) - hrTimeToMs(b.startTime);
     });
   }
   buildHierarchy(spans) {
@@ -1672,46 +1923,66 @@ var JudgeSpanDigestFormatter = class {
     }
     return roots;
   }
-  renderNode(node, lines, depth, sequence, isLast = true) {
+  renderStructureNode(node, lines, depth, isLast = true) {
     const span = node.span;
-    const duration = this.calculateSpanDuration(span);
-    const timestamp = this.formatTimestamp(span.startTime);
-    const status = this.getStatusIndicator(span);
+    const shortId = span.spanContext().spanId.slice(0, 8);
+    const duration = calculateSpanDuration(span);
+    const timestamp = new Date(hrTimeToMs(span.startTime)).toISOString();
+    const status = getStatusIndicator(span);
+    const tokens = getTokenUsage(span);
     const prefix = this.getTreePrefix(depth, isLast);
     lines.push(
-      `${prefix}[${sequence}] ${new Date(timestamp).toISOString()} ${span.name} (${this.formatDuration(duration)})${status}`
+      `${prefix}[${shortId}] ${timestamp} ${span.name} (${formatDuration(duration)}${tokens})${status}`
+    );
+    lines.push("");
+    const childCount = node.children.length;
+    node.children.forEach((child, idx) => {
+      this.renderStructureNode(
+        child,
+        lines,
+        depth + 1,
+        idx === childCount - 1
+      );
+    });
+  }
+  renderNode(node, lines, depth, isLast = true) {
+    const span = node.span;
+    const shortId = span.spanContext().spanId.slice(0, 8);
+    const duration = calculateSpanDuration(span);
+    const timestamp = new Date(hrTimeToMs(span.startTime)).toISOString();
+    const status = getStatusIndicator(span);
+    const prefix = this.getTreePrefix(depth, isLast);
+    lines.push(
+      `${prefix}[${shortId}] ${timestamp} ${span.name} (${formatDuration(duration)})${status}`
     );
     const attrIndent = this.getAttrIndent(depth, isLast);
-    const attrs = this.cleanAttributes(span.attributes);
+    const attrs = cleanAttributes(span.attributes);
     if (Object.keys(attrs).length > 0) {
       for (const [key, value] of Object.entries(attrs)) {
-        lines.push(`${attrIndent}${key}: ${this.formatValue(value)}`);
+        lines.push(`${attrIndent}${key}: ${this.formatValueWithDedup(value)}`);
       }
     }
     if (span.events.length > 0) {
       for (const event of span.events) {
         lines.push(`${attrIndent}[event] ${event.name}`);
         if (event.attributes) {
-          const eventAttrs = this.cleanAttributes(event.attributes);
+          const eventAttrs = cleanAttributes(event.attributes);
           for (const [key, value] of Object.entries(eventAttrs)) {
-            lines.push(`${attrIndent}  ${key}: ${this.formatValue(value)}`);
+            lines.push(`${attrIndent}  ${key}: ${this.formatValueWithDedup(value)}`);
           }
         }
       }
     }
     lines.push("");
-    let nextSeq = sequence + 1;
     const childCount = node.children.length;
     node.children.forEach((child, idx) => {
-      nextSeq = this.renderNode(
+      this.renderNode(
         child,
         lines,
         depth + 1,
-        nextSeq,
         idx === childCount - 1
       );
     });
-    return nextSeq;
   }
   getTreePrefix(depth, isLast) {
     if (depth === 0) return "";
@@ -1723,42 +1994,26 @@ var JudgeSpanDigestFormatter = class {
     const continuation = isLast ? "    " : "\u2502   ";
     return "\u2502   ".repeat(depth - 1) + continuation + "    ";
   }
-  cleanAttributes(attrs) {
-    const cleaned = {};
-    const seen = /* @__PURE__ */ new Set();
-    const excludedKeys = [
-      import_observability2.attributes.ATTR_LANGWATCH_THREAD_ID,
-      "langwatch.scenario.id",
-      "langwatch.scenario.name"
-    ];
-    for (const [key, value] of Object.entries(attrs)) {
-      if (excludedKeys.includes(key)) {
-        continue;
-      }
-      const cleanKey = key.replace(/^(langwatch)\./, "");
-      if (!seen.has(cleanKey)) {
-        seen.add(cleanKey);
-        cleaned[cleanKey] = value;
-      }
-    }
-    return cleaned;
-  }
-  formatValue(value) {
-    const processed = this.transformValue(value);
+  /**
+   * Formats a value with deduplication applied. Used by the `format()` method
+   * to reduce token usage by replacing repeated strings with markers.
+   */
+  formatValueWithDedup(value) {
+    const processed = this.transformValueWithDedup(value);
     return typeof processed === "string" ? processed : JSON.stringify(processed);
   }
-  transformValue(value) {
+  transformValueWithDedup(value) {
     return deepTransform(value, (v) => {
       const mediaPart = truncateMediaPart(v);
       if (mediaPart) return mediaPart;
       if (typeof v !== "string") return v;
-      return this.transformString(v);
+      return this.transformStringWithDedup(v);
     });
   }
-  transformString(str) {
-    if (this.looksLikeJson(str)) {
+  transformStringWithDedup(str) {
+    if (looksLikeJson(str)) {
       try {
-        const processed = this.transformValue(JSON.parse(str));
+        const processed = this.transformValueWithDedup(JSON.parse(str));
         return JSON.stringify(processed);
       } catch {
       }
@@ -1767,36 +2022,12 @@ var JudgeSpanDigestFormatter = class {
     if (truncated !== str) return truncated;
     return this.deduplicator.process(str);
   }
-  looksLikeJson(str) {
-    const t = str.trim();
-    return t.startsWith("{") && t.endsWith("}") || t.startsWith("[") && t.endsWith("]");
-  }
-  hrTimeToMs(hrTime) {
-    return hrTime[0] * 1e3 + hrTime[1] / 1e6;
-  }
-  calculateSpanDuration(span) {
-    return this.hrTimeToMs(span.endTime) - this.hrTimeToMs(span.startTime);
-  }
   calculateTotalDuration(spans) {
     if (spans.length === 0) return 0;
-    const first = this.hrTimeToMs(spans[0].startTime);
-    const last = Math.max(...spans.map((s) => this.hrTimeToMs(s.endTime)));
+    const first = hrTimeToMs(spans[0].startTime);
+    const last = Math.max(...spans.map((s) => hrTimeToMs(s.endTime)));
     return last - first;
   }
-  formatDuration(ms) {
-    if (ms < 1e3) return `${Math.round(ms)}ms`;
-    return `${(ms / 1e3).toFixed(2)}s`;
-  }
-  formatTimestamp(hrTime) {
-    const ms = this.hrTimeToMs(hrTime);
-    return new Date(ms).toISOString();
-  }
-  getStatusIndicator(span) {
-    if (span.status.code === 2) {
-      return ` \u26A0\uFE0F ERROR: ${span.status.message ?? "unknown"}`;
-    }
-    return "";
-  }
   collectErrors(spans) {
     return spans.filter((s) => s.status.code === 2).map((s) => `- ${s.name}: ${s.status.message ?? "unknown error"}`);
   }
@@ -1859,15 +2090,41 @@ function buildFinishTestTool(criteria) {
     })
   });
 }
+function buildProgressiveDiscoveryTools(spans) {
+  return {
+    expand_trace: (0, import_ai2.tool)({
+      description: "Expand one or more spans to see their full details (attributes, events, content). Use the span ID shown in brackets in the trace skeleton.",
+      inputSchema: import_v44.z.object({
+        span_ids: import_v44.z.array(import_v44.z.string()).describe("Span IDs (or 8-char prefixes) to expand")
+      }),
+      execute: async ({ span_ids }) => {
+        return expandTrace(spans, span_ids);
+      }
+    }),
+    grep_trace: (0, import_ai2.tool)({
+      description: "Search across all span attributes, events, and content for a pattern (case-insensitive). Returns matching spans with context.",
+      inputSchema: import_v44.z.object({
+        pattern: import_v44.z.string().describe("Search pattern (case-insensitive)")
+      }),
+      execute: async ({ pattern }) => {
+        return grepTrace(spans, pattern);
+      }
+    })
+  };
+}
 var JudgeAgent = class extends JudgeAgentAdapter {
   constructor(cfg) {
     super();
     this.cfg = cfg;
     this.criteria = cfg.criteria ?? [];
     this.spanCollector = cfg.spanCollector ?? judgeSpanCollector;
+    this.tokenThreshold = cfg.tokenThreshold ?? DEFAULT_TOKEN_THRESHOLD;
+    this.maxDiscoverySteps = cfg.maxDiscoverySteps ?? 10;
   }
   logger = new Logger("JudgeAgent");
   spanCollector;
+  tokenThreshold;
+  maxDiscoverySteps;
   role = "Judge" /* JUDGE */;
   criteria;
   /**
@@ -1875,7 +2132,7 @@ var JudgeAgent = class extends JudgeAgentAdapter {
    */
   invokeLLM = createLLMInvoker(this.logger);
   async call(input) {
-    var _a, _b, _c, _d;
+    var _a;
     const criteria = ((_a = input.judgmentRequest) == null ? void 0 : _a.criteria) ?? this.criteria;
     this.logger.debug("call() invoked", {
       threadId: input.threadId,
@@ -1883,8 +2140,8 @@ var JudgeAgent = class extends JudgeAgentAdapter {
       maxTurns: input.scenarioConfig.maxTurns,
       judgmentRequest: input.judgmentRequest
     });
-    const digest = this.getOpenTelemetryTracesDigest(input.threadId);
-    this.logger.debug("OpenTelemetry traces built", { digest });
+    const spans = this.spanCollector.getSpansForThread(input.threadId);
+    const { digest, isLargeTrace } = this.buildTraceDigest(spans);
     const transcript = JudgeUtils.buildTranscriptFromMessages(input.messages);
     const contentForJudge = `
     <transcript>
@@ -1907,6 +2164,7 @@ var JudgeAgent = class extends JudgeAgentAdapter {
       ...cfg
     });
     const tools = {
+      ...isLargeTrace ? buildProgressiveDiscoveryTools(spans) : {},
       continue_test: buildContinueTestTool(),
       finish_test: buildFinishTestTool(criteria)
     };
@@ -1925,26 +2183,75 @@ var JudgeAgent = class extends JudgeAgentAdapter {
       model: mergedConfig.model,
       toolChoice,
       isLastMessage,
-      enforceJudgement
+      enforceJudgement,
+      isLargeTrace
     });
-    const completion = await this.invokeLLM({
+    const completion = await this.invokeLLMWithDiscovery({
       model: mergedConfig.model,
       messages,
       temperature: mergedConfig.temperature ?? 0,
       maxOutputTokens: mergedConfig.maxTokens,
       tools,
-      toolChoice
+      toolChoice,
+      isLargeTrace
+    });
+    return this.parseToolCalls(completion, criteria);
+  }
+  /**
+   * Builds the trace digest, choosing between full inline rendering
+   * and structure-only mode based on estimated token count.
+   */
+  buildTraceDigest(spans) {
+    const fullDigest = judgeSpanDigestFormatter.format(spans);
+    const isLargeTrace = spans.length > 0 && estimateTokens(fullDigest) > this.tokenThreshold;
+    const digest = isLargeTrace ? judgeSpanDigestFormatter.formatStructureOnly(spans) + "\n\nUse expand_trace(span_id) to see span details or grep_trace(pattern) to search across spans. Reference spans by the ID shown in brackets." : fullDigest;
+    this.logger.debug("Trace digest built", {
+      isLargeTrace,
+      estimatedTokens: estimateTokens(fullDigest)
     });
+    return { digest, isLargeTrace };
+  }
+  /**
+   * Invokes the LLM, enabling multi-step tool execution for large traces.
+   * In multi-step mode, the AI SDK loops automatically: the judge can call
+   * expand_trace/grep_trace tools multiple times before reaching a terminal
+   * tool (finish_test/continue_test) or hitting the step limit.
+   *
+   * When the trace is large, toolChoice is relaxed to "required" so the
+   * judge can freely pick discovery tools (expand_trace/grep_trace) before
+   * being forced to a terminal decision.
+   */
+  async invokeLLMWithDiscovery({
+    isLargeTrace,
+    ...params
+  }) {
+    var _a, _b;
+    if (isLargeTrace) {
+      params.toolChoice = "required";
+      params.stopWhen = [
+        (0, import_ai2.stepCountIs)(this.maxDiscoverySteps),
+        (0, import_ai2.hasToolCall)("finish_test"),
+        (0, import_ai2.hasToolCall)("continue_test")
+      ];
+    }
+    const completion = await this.invokeLLM(params);
     this.logger.debug("LLM response received", {
-      toolCallCount: ((_b = completion.toolCalls) == null ? void 0 : _b.length) ?? 0,
-      toolCalls: (_c = completion.toolCalls) == null ? void 0 : _c.map((tc) => ({
+      toolCallCount: ((_a = completion.toolCalls) == null ? void 0 : _a.length) ?? 0,
+      toolCalls: (_b = completion.toolCalls) == null ? void 0 : _b.map((tc) => ({
         toolName: tc.toolName,
         args: tc.input
       }))
     });
+    return completion;
+  }
+  parseToolCalls(completion, criteria) {
+    var _a;
     let args;
-    if ((_d = completion.toolCalls) == null ? void 0 : _d.length) {
-      const toolCall = completion.toolCalls[0];
+    if ((_a = completion.toolCalls) == null ? void 0 : _a.length) {
+      const terminalCall = completion.toolCalls.find(
+        (tc) => tc.toolName === "finish_test" || tc.toolName === "continue_test"
+      );
+      const toolCall = terminalCall ?? completion.toolCalls[0];
       switch (toolCall.toolName) {
         case "finish_test": {
           args = toolCall.input;
@@ -1986,11 +2293,6 @@ var JudgeAgent = class extends JudgeAgentAdapter {
       unmetCriteria: criteria
     };
   }
-  getOpenTelemetryTracesDigest(threadId) {
-    const spans = this.spanCollector.getSpansForThread(threadId);
-    const digest = judgeSpanDigestFormatter.format(spans);
-    return digest;
-  }
 };
 var judgeAgent = (cfg) => {
   return new JudgeAgent(cfg ?? {});
@@ -4731,6 +5033,7 @@ var index_default = scenario;
   AgentAdapter,
   AgentRole,
   DEFAULT_MAX_TURNS,
+  DEFAULT_TOKEN_THRESHOLD,
   DEFAULT_VERBOSE,
   JudgeAgentAdapter,
   JudgeSpanCollector,
@@ -4743,7 +5046,10 @@ var index_default = scenario;
   agent,
   allAgentRoles,
   defineConfig,
+  estimateTokens,
+  expandTrace,
   fail,
+  grepTrace,
   judge,
   judgeAgent,
   judgeSpanCollector,