@traits-dev/cli 0.2.0 → 0.3.0

package/dist/traits.js

@@ -211,6 +211,7 @@ import {
 import {
   detectEvalTierAvailability,
   formatValidationResult as formatValidationResult2,
+  loadBuiltInEvalSuite,
   resolveTierExecution,
   runOfflineBaselineScaffold,
   toValidationResultObject as toValidationResultObject2
@@ -224,6 +225,7 @@ function printEvalUsage(out = process.stderr) {
       "Options:",
       "  --model <model>           Model target (required)",
       "  --tier <1|2|3>            Highest tier to run (default: highest available)",
+      "  --suite <name>            Built-in baseline suite: support|healthcare|developer",
       "  --provider <name>         Judge provider for Tier 3: auto|openai|anthropic",
       "  --embedding-model <name>  Embedding model for Tier 2 (OpenAI)",
       "  --judge-model <name>      Judge model for Tier 3 provider",
@@ -236,6 +238,11 @@ function printEvalUsage(out = process.stderr) {
       "  --samples <path>          JSON file with samples: [{ id, response }]",
       "  --scenarios <path>        Alias for --samples in this scaffold",
       "  --json                    Output structured JSON",
+      "  --format <text|json|junit> Output format (default: text)",
+      "  --junit-threshold <num>   Global JUnit pass threshold in [0,1] (default: 0.7)",
+      "  --junit-threshold-tier1 <num> Tier 1 JUnit threshold override",
+      "  --junit-threshold-tier2 <num> Tier 2 JUnit threshold override",
+      "  --junit-threshold-tier3 <num> Tier 3 JUnit threshold override",
       "  --strict                  Treat validation warnings as errors",
       "  --verbose                 Include command metadata output",
       "  --no-color                Disable colorized output",
@@ -251,6 +258,7 @@ function parseEvalArgs(args) {
     profilePath: null,
     model: null,
     tier: null,
+    suite: null,
     provider: "auto",
     embeddingModel: null,
     judgeModel: null,
@@ -260,6 +268,11 @@ function parseEvalArgs(args) {
     maxRetries: null,
     retryBaseMs: null,
     json: false,
+    format: "text",
+    junitThreshold: null,
+    junitThresholdTier1: null,
+    junitThresholdTier2: null,
+    junitThresholdTier3: null,
     strict: false,
     verbose: false,
     noColor: false,
@@ -274,6 +287,7 @@ function parseEvalArgs(args) {
     const arg = args[index];
     if (arg === "--json") {
       result.json = true;
+      result.format = "json";
       continue;
     }
     if (arg === "--strict") {
@@ -300,14 +314,18 @@ function parseEvalArgs(args) {
       result.constraintImpact = true;
       continue;
     }
-    if (arg === "--model" || arg === "--tier" || arg === "--provider" || arg === "--embedding-model" || arg === "--judge-model" || arg === "--openai-base-url" || arg === "--anthropic-base-url" || arg === "--timeout-ms" || arg === "--max-retries" || arg === "--retry-base-ms" || arg === "--response" || arg === "--samples" || arg === "--scenarios") {
+    if (arg === "--model" || arg === "--tier" || arg === "--suite" || arg === "--provider" || arg === "--format" || arg === "--embedding-model" || arg === "--judge-model" || arg === "--openai-base-url" || arg === "--anthropic-base-url" || arg === "--timeout-ms" || arg === "--max-retries" || arg === "--retry-base-ms" || arg === "--junit-threshold" || arg === "--junit-threshold-tier1" || arg === "--junit-threshold-tier2" || arg === "--junit-threshold-tier3" || arg === "--response" || arg === "--samples" || arg === "--scenarios") {
       const value = args[index + 1];
       if (!value) return { error: `Missing value for "${arg}"` };
       if (arg === "--model") result.model = value;
       if (arg === "--tier") result.tier = Number(value);
+      if (arg === "--suite") result.suite = String(value).toLowerCase();
       if (arg === "--provider") {
         result.provider = String(value).toLowerCase();
       }
+      if (arg === "--format") {
+        result.format = String(value).toLowerCase();
+      }
       if (arg === "--embedding-model") result.embeddingModel = value;
       if (arg === "--judge-model") result.judgeModel = value;
       if (arg === "--openai-base-url") result.openaiBaseUrl = value;
@@ -315,6 +333,10 @@ function parseEvalArgs(args) {
       if (arg === "--timeout-ms") result.timeoutMs = Number(value);
       if (arg === "--max-retries") result.maxRetries = Number(value);
       if (arg === "--retry-base-ms") result.retryBaseMs = Number(value);
+      if (arg === "--junit-threshold") result.junitThreshold = Number(value);
+      if (arg === "--junit-threshold-tier1") result.junitThresholdTier1 = Number(value);
+      if (arg === "--junit-threshold-tier2") result.junitThresholdTier2 = Number(value);
+      if (arg === "--junit-threshold-tier3") result.junitThresholdTier3 = Number(value);
       if (arg === "--response") result.responses.push(value);
       if (arg === "--samples" || arg === "--scenarios") result.samplesPath = value;
       index += 1;
@@ -338,6 +360,20 @@ function parseEvalArgs(args) {
   if (!["auto", "openai", "anthropic"].includes(result.provider)) {
     return { error: 'Invalid "--provider" value. Expected auto, openai, or anthropic.' };
   }
+  if (result.suite != null && !["support", "healthcare", "developer"].includes(
+    result.suite
+  )) {
+    return { error: 'Invalid "--suite" value. Expected support, healthcare, or developer.' };
+  }
+  if (result.suite != null && result.samplesPath != null) {
+    return { error: 'Use either "--suite" or "--samples/--scenarios", not both.' };
+  }
+  if (result.suite != null && result.responses.length > 0) {
+    return { error: 'Use either "--suite" or "--response", not both.' };
+  }
+  if (!["text", "json", "junit"].includes(result.format)) {
+    return { error: 'Invalid "--format" value. Expected text, json, or junit.' };
+  }
   if (result.timeoutMs != null && (!Number.isInteger(result.timeoutMs) || result.timeoutMs < 0)) {
     return { error: 'Invalid "--timeout-ms" value. Expected a non-negative integer.' };
   }
@@ -347,9 +383,33 @@ function parseEvalArgs(args) {
   if (result.retryBaseMs != null && (!Number.isInteger(result.retryBaseMs) || result.retryBaseMs < 0)) {
     return { error: 'Invalid "--retry-base-ms" value. Expected a non-negative integer.' };
   }
+  for (const [flag, value] of [
+    ["--junit-threshold", result.junitThreshold],
+    ["--junit-threshold-tier1", result.junitThresholdTier1],
+    ["--junit-threshold-tier2", result.junitThresholdTier2],
+    ["--junit-threshold-tier3", result.junitThresholdTier3]
+  ]) {
+    if (value == null) continue;
+    if (!Number.isFinite(value) || value < 0 || value > 1) {
+      return { error: `Invalid "${flag}" value. Expected a number in [0, 1].` };
+    }
+  }
   return { value: result };
 }
 function loadSamples(options, cwd) {
+  if (options.suite) {
+    const suite = loadBuiltInEvalSuite(options.suite);
+    if (!suite) {
+      throw new Error(
+        `Unknown suite "${options.suite}". Expected support, healthcare, or developer.`
+      );
+    }
+    return suite.scenarios.map((scenario) => ({
+      id: scenario.id,
+      prompt: scenario.messages.map((message) => `${message.role}: ${message.content}`).join("\n"),
+      response: scenario.expected_behavior ?? ""
+    }));
+  }
   if (options.samplesPath) {
     const sampleFile = path2.resolve(cwd, options.samplesPath);
     const parsed = JSON.parse(fs.readFileSync(sampleFile, "utf8"));
@@ -377,10 +437,118 @@ function loadSamples(options, cwd) {
   }));
 }
 function writeProgress(io, options, message) {
-  if (options.json) return;
+  if (options.format !== "text") return;
   io.stderr.write(`${message}
 `);
 }
+function escapeXml(value) {
+  return String(value ?? "").replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;").replace(/"/g, "&quot;").replace(/'/g, "&apos;");
+}
+function resolveJUnitThresholds(options) {
+  const base = options.junitThreshold ?? 0.7;
+  return {
+    tier1: options.junitThresholdTier1 ?? base,
+    tier2: options.junitThresholdTier2 ?? base,
+    tier3: options.junitThresholdTier3 ?? base
+  };
+}
+function buildSampleScoreMap(samples) {
+  return new Map((samples ?? []).map((sample) => [String(sample.id), Number(sample.score)]));
+}
+function collectScenarioIds(reports) {
+  const ids = [];
+  const seen = /* @__PURE__ */ new Set();
+  for (const sample of reports.tier1?.samples ?? []) {
+    const id = String(sample.id);
+    if (seen.has(id)) continue;
+    seen.add(id);
+    ids.push(id);
+  }
+  for (const sample of reports.tier2?.samples ?? []) {
+    const id = String(sample.id);
+    if (seen.has(id)) continue;
+    seen.add(id);
+    ids.push(id);
+  }
+  for (const sample of reports.tier3?.samples ?? []) {
+    const id = String(sample.id);
+    if (seen.has(id)) continue;
+    seen.add(id);
+    ids.push(id);
+  }
+  return ids;
+}
+function buildJUnitReport(args) {
+  const ids = collectScenarioIds(args.tierReports);
+  const tier1Scores = buildSampleScoreMap(args.tierReports.tier1?.samples);
+  const tier2Scores = buildSampleScoreMap(args.tierReports.tier2?.samples);
+  const tier3Scores = buildSampleScoreMap(args.tierReports.tier3?.samples);
+  const className = `traits.eval.${path2.basename(args.profilePath, path2.extname(args.profilePath))}`;
+  let failures = 0;
+  const testCases = [];
+  for (const id of ids) {
+    const reasons = [];
+    const scoreLines = [];
+    const tier1 = tier1Scores.get(id);
+    if (tier1 != null) {
+      scoreLines.push(`tier1=${tier1.toFixed(3)} threshold=${args.thresholds.tier1.toFixed(3)}`);
+      if (tier1 < args.thresholds.tier1) {
+        reasons.push(
+          `Tier 1 score ${tier1.toFixed(3)} below threshold ${args.thresholds.tier1.toFixed(3)}`
+        );
+      }
+    }
+    const tier2 = tier2Scores.get(id);
+    if (tier2 != null) {
+      scoreLines.push(`tier2=${tier2.toFixed(3)} threshold=${args.thresholds.tier2.toFixed(3)}`);
+      if (tier2 < args.thresholds.tier2) {
+        reasons.push(
+          `Tier 2 score ${tier2.toFixed(3)} below threshold ${args.thresholds.tier2.toFixed(3)}`
+        );
+      }
+    }
+    const tier3 = tier3Scores.get(id);
+    if (tier3 != null) {
+      scoreLines.push(`tier3=${tier3.toFixed(3)} threshold=${args.thresholds.tier3.toFixed(3)}`);
+      if (tier3 < args.thresholds.tier3) {
+        reasons.push(
+          `Tier 3 score ${tier3.toFixed(3)} below threshold ${args.thresholds.tier3.toFixed(3)}`
+        );
+      }
+    }
+    const testcase = [];
+    testcase.push(
+      `  <testcase classname="${escapeXml(className)}" name="${escapeXml(id)}" time="0">`
+    );
+    if (reasons.length > 0) {
+      failures += 1;
+      testcase.push(
+        `    <failure message="${escapeXml("traits eval threshold failure")}">${escapeXml(
+          reasons.join(" | ")
+        )}</failure>`
+      );
+    }
+    if (scoreLines.length > 0) {
+      testcase.push(`    <system-out>${escapeXml(scoreLines.join(" | "))}</system-out>`);
+    }
+    testcase.push("  </testcase>");
+    testCases.push(testcase.join("\n"));
+  }
+  const xml = [
+    '<?xml version="1.0" encoding="UTF-8"?>',
+    "<testsuites>",
+    `  <testsuite name="traits.eval" tests="${ids.length}" failures="${failures}" errors="0" skipped="0" time="0">`,
+    `    <properties><property name="profile" value="${escapeXml(args.profilePath)}" /><property name="model" value="${escapeXml(args.model)}" /><property name="threshold_tier1" value="${args.thresholds.tier1.toFixed(3)}" /><property name="threshold_tier2" value="${args.thresholds.tier2.toFixed(3)}" /><property name="threshold_tier3" value="${args.thresholds.tier3.toFixed(3)}" /></properties>`,
+    ...testCases,
+    "  </testsuite>",
+    "</testsuites>"
+  ].join("\n");
+  return {
+    xml,
+    tests: ids.length,
+    failures
+  };
+}
 async function runEval(args, io = process) {
   const parsed = parseEvalArgs(args);
   if ("error" in parsed) {
@@ -515,6 +683,8 @@ async function runEval(args, io = process) {
     const payload = {
       profile: profilePath,
       model: options.model,
+      format: options.format,
+      suite: options.suite,
       tier_requested: requestedTier,
       tier_executed: tierResolution.tier_executed,
       tier_resolution: tierResolution,
@@ -531,11 +701,22 @@ async function runEval(args, io = process) {
         errors: evaluation.validation.errors.length
       };
     }
-    if (options.json) {
+    if (options.format === "json") {
       io.stdout.write(`${JSON.stringify(payload, null, 2)}
 `);
       return 0;
     }
+    if (options.format === "junit") {
+      const junit = buildJUnitReport({
+        profilePath,
+        model: options.model,
+        tierReports,
+        thresholds: resolveJUnitThresholds(options)
+      });
+      io.stdout.write(`${junit.xml}
+`);
+      return junit.failures > 0 ? 1 : 0;
+    }
     if (tierReports.tier1) {
       io.stdout.write(`Tier 1 average score: ${tierReports.tier1.average_score.toFixed(3)}
 `);
@@ -547,10 +728,16 @@ async function runEval(args, io = process) {
     if (tierReports.tier2) {
       io.stdout.write(`Tier 2 average score: ${tierReports.tier2.average_score.toFixed(3)}
 `);
+      io.stdout.write(
+        "Note: Tier 2 embedding scores are directionally useful but sensitive to model granularity.\n"
+      );
     }
     if (tierReports.tier3) {
       io.stdout.write(`Tier 3 average score: ${tierReports.tier3.average_score.toFixed(3)}
 `);
+      io.stdout.write(
+        "Note: Tier 3 judge scores are noisy across runs. Do not use as a sole merge gate.\n"
+      );
     }
     if (baselineReport?.tier1) {
       io.stdout.write(
@@ -578,12 +765,12 @@ async function runEval(args, io = process) {
     return 0;
   } catch (error) {
     const typedError = error;
-    if ((typedError.code === "E_EVAL_TIER2_UNAVAILABLE" || typedError.code === "E_EVAL_TIER3_UNAVAILABLE") && !options.json) {
+    if ((typedError.code === "E_EVAL_TIER2_UNAVAILABLE" || typedError.code === "E_EVAL_TIER3_UNAVAILABLE") && options.format !== "json") {
       io.stderr.write(`Error: ${typedError.message ?? "Evaluation tier unavailable."}
 `);
       return 2;
     }
-    if ((typedError.code === "E_EVAL_TIER2_UNAVAILABLE" || typedError.code === "E_EVAL_TIER3_UNAVAILABLE") && options.json) {
+    if ((typedError.code === "E_EVAL_TIER2_UNAVAILABLE" || typedError.code === "E_EVAL_TIER3_UNAVAILABLE") && options.format === "json") {
       io.stdout.write(
         `${JSON.stringify(
           {
@@ -599,7 +786,7 @@ async function runEval(args, io = process) {
     }
     const validation = typedError.validation;
     if (typedError.code === "E_EVAL_VALIDATION" && validation) {
-      if (options.json) {
+      if (options.format === "json") {
         io.stdout.write(
           `${JSON.stringify(
             {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@traits-dev/cli",
-  "version": "0.2.0",
+  "version": "0.3.0",
   "description": "traits.dev command-line interface for voice profile init, validate, compile, eval, and import workflows.",
   "keywords": [
     "traits-dev",
@@ -41,7 +41,7 @@
     "provenance": true
   },
   "dependencies": {
-    "@traits-dev/core": "^0.2.0"
+    "@traits-dev/core": "^0.3.0"
   },
   "devDependencies": {
     "@types/node": "^25.2.3",