@axiom-lattice/agent-eval 2.1.9 → 2.1.10

package/.turbo/turbo-build.log

@@ -1,5 +1,5 @@
 
-> @axiom-lattice/agent-eval@2.1.9 build /home/runner/work/agentic/agentic/packages/agent-eval
+> @axiom-lattice/agent-eval@2.1.10 build /home/runner/work/agentic/agentic/packages/agent-eval
 > tsup src/index.ts --format cjs,esm --dts --sourcemap
 
 CLI Building entry: src/index.ts
@@ -8,13 +8,13 @@
 CLI Target: es2020
 CJS Build start
 ESM Build start
-CJS dist/index.js     39.11 KB
-CJS dist/index.js.map 67.54 KB
-CJS ⚡️ Build success in 186ms
-ESM dist/index.mjs     36.92 KB
-ESM dist/index.mjs.map 67.28 KB
-ESM ⚡️ Build success in 191ms
+CJS dist/index.js     41.54 KB
+CJS dist/index.js.map 72.12 KB
+CJS ⚡️ Build success in 155ms
+ESM dist/index.mjs     39.35 KB
+ESM dist/index.mjs.map 71.85 KB
+ESM ⚡️ Build success in 159ms
 DTS Build start
-DTS ⚡️ Build success in 8789ms
-DTS dist/index.d.ts  10.68 KB
-DTS dist/index.d.mts 10.68 KB
+DTS ⚡️ Build success in 9956ms
+DTS dist/index.d.ts  10.99 KB
+DTS dist/index.d.mts 10.99 KB

package/CHANGELOG.md

@@ -1,5 +1,14 @@
 # @axiom-lattice/agent-eval
 
+## 2.1.10
+
+### Patch Changes
+
+- 773c03f: add skills
+- Updated dependencies [773c03f]
+  - @axiom-lattice/protocols@2.1.10
+  - @axiom-lattice/core@2.1.16
+
 ## 2.1.9
 
 ### Patch Changes

package/dist/index.d.mts

@@ -10,7 +10,6 @@ type OutputFileContent = {
 };
 type OutputMessageContent = {
     type: "message_content";
-    message: string;
 };
 type OutputType = OutputFileContent | OutputMessageContent;
 interface LatticeEvalProjectType {
@@ -170,6 +169,11 @@ interface LatticeEvalCaseRunResult {
     judge_thread_id?: string;
     test_prompt?: string;
     final_output?: string;
+    messages?: Array<{
+        role: string;
+        content: string;
+        id?: string;
+    }>;
     logs: LatticeEvalLogEvent[];
 }
 /**
@@ -185,12 +189,18 @@ declare class LatticeEval {
     private lastTestPrompt?;
     private lastFinalOutput?;
     private lastDurationMs;
+    private lastMessages;
     getLastRunMeta(): {
         duration_ms: number;
         thread_id: string | undefined;
         judge_thread_id: string | undefined;
         test_prompt: string | undefined;
         final_output: string | undefined;
+        messages: {
+            role: string;
+            content: string;
+            id?: string;
+        }[];
     };
     /**
      * Create a new LatticeEval instance
@@ -255,6 +265,11 @@ interface CaseRunResult {
     judge_thread_id?: string;
     test_prompt?: string;
     final_output?: string;
+    messages?: Array<{
+        role: string;
+        content: string;
+        id?: string;
+    }>;
     error_stack?: string;
 }
 /**

package/dist/index.d.ts

@@ -10,7 +10,6 @@ type OutputFileContent = {
 };
 type OutputMessageContent = {
     type: "message_content";
-    message: string;
 };
 type OutputType = OutputFileContent | OutputMessageContent;
 interface LatticeEvalProjectType {
@@ -170,6 +169,11 @@ interface LatticeEvalCaseRunResult {
     judge_thread_id?: string;
     test_prompt?: string;
     final_output?: string;
+    messages?: Array<{
+        role: string;
+        content: string;
+        id?: string;
+    }>;
     logs: LatticeEvalLogEvent[];
 }
 /**
@@ -185,12 +189,18 @@ declare class LatticeEval {
     private lastTestPrompt?;
     private lastFinalOutput?;
     private lastDurationMs;
+    private lastMessages;
     getLastRunMeta(): {
         duration_ms: number;
         thread_id: string | undefined;
         judge_thread_id: string | undefined;
         test_prompt: string | undefined;
         final_output: string | undefined;
+        messages: {
+            role: string;
+            content: string;
+            id?: string;
+        }[];
     };
     /**
      * Create a new LatticeEval instance
@@ -255,6 +265,11 @@ interface CaseRunResult {
     judge_thread_id?: string;
     test_prompt?: string;
     final_output?: string;
+    messages?: Array<{
+        role: string;
+        content: string;
+        id?: string;
+    }>;
     error_stack?: string;
 }
 /**

package/dist/index.js

@@ -50,6 +50,7 @@ var LatticeEval = class {
   constructor(config) {
     this.inMemoryLogs = [];
     this.lastDurationMs = 0;
+    this.lastMessages = [];
     this.config = config;
     this.baseUrl = this.config.base_url;
     this.verbose = this.config.verbose ?? true;
@@ -60,7 +61,8 @@ var LatticeEval = class {
       thread_id: this.lastThreadId,
       judge_thread_id: this.lastJudgeThreadId,
       test_prompt: this.lastTestPrompt,
-      final_output: this.lastFinalOutput
+      final_output: this.lastFinalOutput,
+      messages: this.lastMessages
     };
   }
   getInMemoryLogs() {
@@ -78,9 +80,19 @@ var LatticeEval = class {
     if (level === "error") {
       const keyInfo = this.getKeyInfo(data);
       console.log(`  \u2717 ${message}${keyInfo ? ` ${keyInfo}` : ""}`);
-    } else if (message.includes("Starting case") || message.includes("Case evaluation completed")) {
+    } else if (message.startsWith("Starting")) {
       const keyInfo = this.getKeyInfo(data);
       console.log(`  ${message}${keyInfo ? ` ${keyInfo}` : ""}`);
+    } else if (message.includes("Case evaluation completed")) {
+      const keyInfo = this.getKeyInfo(data);
+      const pass = data?.pass;
+      const summary = data?.summary;
+      const status = pass ? "\u2713 PASS" : "\u2717 FAIL";
+      const reason = summary || (pass ? "Test passed" : "Test failed");
+      console.log(`  ${status} ${message}${keyInfo ? ` ${keyInfo}` : ""}`);
+      if (reason) {
+        console.log(`    Reason: ${reason}`);
+      }
     }
   }
   log(message, data) {
@@ -216,11 +228,15 @@ var LatticeEval = class {
     this.lastTestPrompt = void 0;
     this.lastFinalOutput = void 0;
     this.lastDurationMs = 0;
-    this.log("Starting case evaluation", {
+    this.lastMessages = [];
+    const contentAssertion = evalCase.eval?.content_assertion || "";
+    const message = contentAssertion ? `Starting: ${contentAssertion}` : "Starting";
+    this.log(message, {
       case_id: evalCase.caseId,
       thread_id: threadId,
       steps_count: evalCase.steps?.length,
-      output_type: evalCase.output?.type
+      output_type: evalCase.output?.type,
+      content_assertion: contentAssertion
     });
     let currentThreadId = threadId;
     let lastResponseData = null;
@@ -233,6 +249,38 @@ var LatticeEval = class {
       );
       currentThreadId = result.threadId;
       lastResponseData = result.responseData;
+      const existingIds = new Set(this.lastMessages.map((m) => m.id).filter(Boolean));
+      if (result.responseData?.messages && Array.isArray(result.responseData.messages)) {
+        for (const msg of result.responseData.messages) {
+          if (msg && typeof msg === "object") {
+            const msgId = msg.id || msg.lc_id || msg._id;
+            if (msgId && existingIds.has(msgId)) {
+              continue;
+            }
+            const role = msg.role || msg.getType?.() || msg.type || "unknown";
+            let content = "";
+            if (typeof msg.content === "string") {
+              content = msg.content;
+            } else if (Array.isArray(msg.content)) {
+              content = msg.content.map(
+                (c) => typeof c === "string" ? c : c?.text || (typeof c === "object" ? JSON.stringify(c) : String(c))
+              ).join("\n");
+            } else if (msg.content && typeof msg.content === "object") {
+              content = msg.content.text || JSON.stringify(msg.content);
+            } else {
+              content = String(msg.content || "");
+            }
+            this.lastMessages.push({
+              role,
+              content,
+              id: msgId
+            });
+            if (msgId) {
+              existingIds.add(msgId);
+            }
+          }
+        }
+      }
     }
     const finalAgentId = evalCase.steps[evalCase.steps.length - 1]?.agent_id || "";
     this.log("All agent steps completed", {
@@ -424,10 +472,12 @@ ${rubricsSection}
         pass
       });
     }
+    const summary = parsedResult.summary || testResultContent || "";
     this.log("Case evaluation completed", {
       case_id: evalCase.caseId,
       pass,
-      final_score: finalScore
+      final_score: finalScore,
+      summary
     });
     const finishedAt = Date.now();
     this.lastDurationMs = finishedAt - startedAt;
@@ -466,6 +516,7 @@ async function evaluateLatticeCaseWithLogs(evalCase, config) {
       judge_thread_id: meta.judge_thread_id,
       test_prompt: meta.test_prompt,
       final_output: meta.final_output,
+      messages: meta.messages,
       logs: evaluator.getInMemoryLogs()
     };
   } catch (error) {
@@ -485,6 +536,7 @@ async function evaluateLatticeCaseWithLogs(evalCase, config) {
       judge_thread_id: meta.judge_thread_id,
       test_prompt: meta.test_prompt,
       final_output: meta.final_output,
+      messages: meta.messages,
       logs: evaluator.getInMemoryLogs()
     };
   }
@@ -632,6 +684,7 @@ var LatticeEvalSuite = class {
         judge_thread_id: run.judge_thread_id,
         test_prompt: run.test_prompt,
         final_output: run.final_output,
+        messages: run.messages,
         logs: run.logs
       };
     } catch (error) {
@@ -668,6 +721,7 @@ var LatticeEvalSuite = class {
           judge_thread_id: run.judge_thread_id,
           test_prompt: run.test_prompt,
           final_output: run.final_output,
+          messages: run.messages,
           logs: run.logs
         };
       } catch (error) {
@@ -936,12 +990,26 @@ Results saved to: ${batch_dir}`);
       lines.push(`\`\`\``);
       lines.push(``);
     }
+    if (payload.messages && Array.isArray(payload.messages) && payload.messages.length > 0) {
+      lines.push(`## Conversation Messages`);
+      lines.push(``);
+      for (let i = 0; i < payload.messages.length; i++) {
+        const msg = payload.messages[i];
+        const role = msg.role || "unknown";
+        const content = msg.content || "";
+        lines.push(`### Message ${i + 1} (${role})`);
+        lines.push(``);
+        lines.push(`\`\`\``);
+        lines.push(content);
+        lines.push(`\`\`\``);
+        lines.push(``);
+      }
+    }
     if (payload.finalOutput) {
       lines.push(`## Final Output`);
       lines.push(``);
       lines.push(`\`\`\``);
-      const output = payload.finalOutput.length > 5e3 ? payload.finalOutput.substring(0, 5e3) + "\n\n... (truncated, see JSON for full output)" : payload.finalOutput;
-      lines.push(output);
+      lines.push(payload.finalOutput);
       lines.push(`\`\`\``);
       lines.push(``);
     }
@@ -949,8 +1017,7 @@ Results saved to: ${batch_dir}`);
       lines.push(`## Test Prompt`);
       lines.push(``);
       lines.push(`\`\`\``);
-      const prompt = payload.testPrompt.length > 5e3 ? payload.testPrompt.substring(0, 5e3) + "\n\n... (truncated, see JSON for full prompt)" : payload.testPrompt;
-      lines.push(prompt);
+      lines.push(payload.testPrompt);
       lines.push(`\`\`\``);
       lines.push(``);
     }
@@ -1056,7 +1123,8 @@ Results saved to: ${batch_dir}`);
           threadId: r.thread_id,
           judgeThreadId: r.judge_thread_id,
           finalOutput: r.final_output,
-          testPrompt: r.test_prompt
+          testPrompt: r.test_prompt,
+          messages: r.messages
         };
         await (0, import_promises.writeFile)(jsonPath, JSON.stringify(payload, null, 2), "utf-8");
         const mdPath = import_path.default.join(individualDir, `${baseFilename}.md`);