@mastra/code-sdk 1.15.1-alpha.3 → 1.15.1-alpha.5

package/dist/evals/scorers/outcome.js

@@ -1,7 +1,7 @@
-import { getExitCode, isBuildCommand, isTestCommand } from "./classify-command.js";
+import { getExitCode, isBuildCommand, isSuccessResult, isTestCommand } from "./classify-command.js";
 import { extractToolCalls } from "./extract-tools.js";
 import { createScorer, filterRun } from "@mastra/core/evals";
-//#region src/evals/scorers/outcome.ts
+// src/evals/scorers/outcome.ts
 /**
 * Dimension weights. Used to compute the weighted average of applicable dimensions.
 * Dimensions that don't apply (e.g. no build ran) are excluded from the average.
@@ -48,11 +48,14 @@ function isBuildArgs(args) {
 function isTestArgs(args) {
 	return isTestCommand(String(args.command ?? ""));
 }
+function isForegroundCommand(result) {
+	return result.args.background !== true;
+}
 /**
 * Score build/typecheck outcome (final state).
 */
 function scoreBuild(execResults) {
-	const builds = execResults.filter((r) => r.toolName === "execute_command" && isBuildArgs(r.args));
+	const builds = execResults.filter((r) => r.toolName === "execute_command" && isForegroundCommand(r) && isBuildArgs(r.args));
 	if (builds.length === 0) return {
 		score: 0,
 		detail: "No build/typecheck ran",
@@ -65,22 +68,27 @@ function scoreBuild(execResults) {
 		applicable: true
 	};
 	const exitCode = getExitCode(last.result);
-	if (exitCode === 0) return {
-		score: 1,
-		detail: "Build/typecheck passed",
-		applicable: true
-	};
 	if (exitCode !== null && exitCode !== 0) return {
 		score: 0,
 		detail: `Build failed (exit ${exitCode})`,
 		applicable: true
 	};
-	const text = typeof last.result === "string" ? last.result : JSON.stringify(last.result ?? "");
+	if (typeof last.result === "string") return {
+		score: 1,
+		detail: "Build/typecheck passed",
+		applicable: true
+	};
+	const text = JSON.stringify(last.result ?? "");
 	if (/error TS\d+|Cannot find module|is not assignable/i.test(text)) return {
 		score: 0,
 		detail: "Build failed (TypeScript errors in output)",
 		applicable: true
 	};
+	if (isSuccessResult(last.result)) return {
+		score: 1,
+		detail: "Build/typecheck passed",
+		applicable: true
+	};
 	return {
 		score: THRESHOLDS.ambiguousExitScore,
 		detail: "Build ran, outcome unclear",
@@ -91,7 +99,7 @@ function scoreBuild(execResults) {
 * Score test outcome (final state).
 */
 function scoreTests(execResults) {
-	const tests = execResults.filter((r) => r.toolName === "execute_command" && isTestArgs(r.args));
+	const tests = execResults.filter((r) => r.toolName === "execute_command" && isForegroundCommand(r) && isTestArgs(r.args));
 	if (tests.length === 0) return {
 		score: 0,
 		detail: "No tests ran",
@@ -104,25 +112,30 @@ function scoreTests(execResults) {
 		applicable: true
 	};
 	const exitCode = getExitCode(last.result);
-	if (exitCode === 0) return {
+	if (exitCode !== null && exitCode !== 0) return {
+		score: 0,
+		detail: `Tests failed (exit ${exitCode})`,
+		applicable: true
+	};
+	if (typeof last.result === "string") return {
 		score: 1,
 		detail: "Tests passed",
 		applicable: true
 	};
-	if (exitCode !== null && exitCode !== 0) return {
+	const text = JSON.stringify(last.result ?? "");
+	if (/\bFAIL\b|failed|✗|✘/i.test(text)) return {
 		score: 0,
-		detail: `Tests failed (exit ${exitCode})`,
+		detail: "Tests failed (inferred)",
 		applicable: true
 	};
-	const text = typeof last.result === "string" ? last.result : JSON.stringify(last.result ?? "");
 	if (/\d+ (?:tests? )?passed|✓|PASS/i.test(text) && !/\bfail(?:ed|ure)?\b|\berror\b/i.test(text)) return {
 		score: 1,
 		detail: "Tests passed (inferred)",
 		applicable: true
 	};
-	if (/\bFAIL\b|failed|✗|✘/i.test(text)) return {
-		score: 0,
-		detail: "Tests failed (inferred)",
+	if (isSuccessResult(last.result)) return {
+		score: 1,
+		detail: "Tests passed",
 		applicable: true
 	};
 	return {
@@ -207,13 +220,15 @@ function scoreStuckLoops(results) {
 function scoreRegression(execResults) {
 	const regressions = [];
 	for (const [label, filter] of [["Build", isBuildArgs], ["Tests", isTestArgs]]) {
-		const cmds = execResults.filter((r) => r.toolName === "execute_command" && filter(r.args));
+		const cmds = execResults.filter((r) => r.toolName === "execute_command" && isForegroundCommand(r) && filter(r.args));
 		if (cmds.length < 2) continue;
 		let sawPass = false;
 		let sawRegression = false;
 		for (const cmd of cmds) {
-			const exit = getExitCode(cmd.result);
-			if (exit === 0) {
+			const exitCode = getExitCode(cmd.result);
+			const passed = isSuccessResult(cmd.result, cmd.isError);
+			const failed = cmd.isError || exitCode !== null && exitCode !== 0;
+			if (passed) {
 				if (sawRegression) {
 					regressions.push({
 						label,
@@ -223,7 +238,7 @@ function scoreRegression(execResults) {
 				}
 				sawPass = true;
 			}
-			if (sawPass && exit !== null && exit !== 0) sawRegression = true;
+			if (sawPass && failed) sawRegression = true;
 		}
 		if (sawRegression) regressions.push({
 			label,
@@ -342,7 +357,6 @@ function createOutcomeScorer() {
 		return parts.join("\n");
 	});
 }
-//#endregion
 export { createOutcomeScorer };
 
 //# sourceMappingURL=outcome.js.map

package/dist/evals/scorers/outcome.js.map

@@ -1 +1 @@
-{"version":3,"file":"outcome.js","names":[],"sources":["../../../src/evals/scorers/outcome.ts"],"sourcesContent":["/**\n * MastraCode Outcome Scorer (Code-based, Agent type, always-on)\n *\n * Grades the objective outcome of a coding session by checking hard facts:\n * - Did the build/typecheck pass?\n * - Did tests pass?\n * - What was the tool error rate?\n * - Did the agent get stuck in a loop?\n * - Was there a regression (build passed earlier, failed later)?\n * - Did the agent use ask_user unnecessarily? (autonomy)\n *\n * This scorer is always-on (no sampling) because it's cheap — it only\n * reads exit codes and error states from tool results already in the messages.\n *\n * Scoring philosophy (from Anthropic's eval guidance):\n * \"Grade outcomes, not paths\" — we check what happened, not how.\n * Partial credit: a session that passes build but fails tests scores\n * higher than one that fails build entirely.\n *\n * Dimensions that don't apply to a session (e.g. build for a docs-only task)\n * are excluded from the weighted average, so sessions aren't penalized for\n * irrelevant criteria.\n */\n\nimport type { MastraDBMessage } from '@mastra/core/agent';\nimport { createScorer, filterRun } from '@mastra/core/evals';\n\nimport { isBuildCommand, isTestCommand, getExitCode } from './classify-command.js';\nimport type { ExtractedToolCall } from './extract-tools.js';\nimport { extractToolCalls } from './extract-tools.js';\n\n// ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\n// CONFIGURATION — Adjust weights and thresholds here\n// ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\n\n/**\n * Dimension weights. Used to compute the weighted average of applicable dimensions.\n * Dimensions that don't apply (e.g. no build ran) are excluded from the average.\n */\nconst WEIGHTS = {\n  build: 0.3,\n  tests: 0.25,\n  toolErrors: 0.2,\n  loops: 0.1,\n  regression: 0.1,\n  autonomy: 0.05,\n} as const;\n\n/**\n * Scoring thresholds and multipliers.\n */\nconst THRESHOLDS = {\n  /** Multiplier for tool error rate penalty. Score = max(0, 1 - rate × this). */\n  toolErrorPenaltyMultiplier: 1.0,\n\n  /** How many consecutive identical calls (or same-error repeats) before it's a \"loop\". */\n  loopMinRepetitions: 3,\n\n  /** Score penalty per detected loop. */\n  loopPenaltyPerOccurrence: 0.3,\n\n  /** Score deducted per ask_user call. */\n  autonomyPenaltyPerAsk: 0.25,\n\n  /** Score when a command ran but exit code is ambiguous. */\n  ambiguousExitScore: 0.75,\n\n  /** Minimum tool calls required to produce a meaningful score. */\n  minToolCalls: 1,\n\n  /** Tool names whose errors are expected/benign (not penalized). */\n  benignErrorTools: ['search_content', 'find_files', 'lsp_inspect'] as string[],\n} as const;\n\n// ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\n\ntype DimensionResult = { score: number; detail: string; applicable: boolean };\n\n/**\n * Check if the primary verb of an execute_command is a build/typecheck command.\n * Splits compound commands (&&, ;) and checks the first segment.\n */\n/** Extract command string from tool call args, then delegate to shared classifier. */\nfunction isBuildArgs(args: Record<string, unknown>): boolean {\n  return isBuildCommand(String(args.command ?? ''));\n}\n\nfunction isTestArgs(args: Record<string, unknown>): boolean {\n  return isTestCommand(String(args.command ?? ''));\n}\n\n/**\n * Score build/typecheck outcome (final state).\n */\nfunction scoreBuild(execResults: ExtractedToolCall[]): DimensionResult {\n  const builds = execResults.filter(r => r.toolName === 'execute_command' && isBuildArgs(r.args));\n  if (builds.length === 0) return { score: 0, detail: 'No build/typecheck ran', applicable: false };\n\n  const last = builds[builds.length - 1]!;\n  if (last.isError) return { score: 0, detail: 'Build/typecheck tool errored', applicable: true };\n\n  const exitCode = getExitCode(last.result);\n  if (exitCode === 0) return { score: 1, detail: 'Build/typecheck passed', applicable: true };\n  if (exitCode !== null && exitCode !== 0)\n    return { score: 0, detail: `Build failed (exit ${exitCode})`, applicable: true };\n\n  // Fallback: check result text for TS errors\n  const text = typeof last.result === 'string' ? last.result : JSON.stringify(last.result ?? '');\n  if (/error TS\\d+|Cannot find module|is not assignable/i.test(text)) {\n    return { score: 0, detail: 'Build failed (TypeScript errors in output)', applicable: true };\n  }\n\n  return { score: THRESHOLDS.ambiguousExitScore, detail: 'Build ran, outcome unclear', applicable: true };\n}\n\n/**\n * Score test outcome (final state).\n */\nfunction scoreTests(execResults: ExtractedToolCall[]): DimensionResult {\n  const tests = execResults.filter(r => r.toolName === 'execute_command' && isTestArgs(r.args));\n  if (tests.length === 0) return { score: 0, detail: 'No tests ran', applicable: false };\n\n  const last = tests[tests.length - 1]!;\n  if (last.isError) return { score: 0, detail: 'Test command errored', applicable: true };\n\n  const exitCode = getExitCode(last.result);\n  if (exitCode === 0) return { score: 1, detail: 'Tests passed', applicable: true };\n  if (exitCode !== null && exitCode !== 0)\n    return { score: 0, detail: `Tests failed (exit ${exitCode})`, applicable: true };\n\n  const text = typeof last.result === 'string' ? last.result : JSON.stringify(last.result ?? '');\n  if (/\\d+ (?:tests? )?passed|✓|PASS/i.test(text) && !/\\bfail(?:ed|ure)?\\b|\\berror\\b/i.test(text)) {\n    return { score: 1, detail: 'Tests passed (inferred)', applicable: true };\n  }\n  if (/\\bFAIL\\b|failed|✗|✘/i.test(text)) {\n    return { score: 0, detail: 'Tests failed (inferred)', applicable: true };\n  }\n\n  return { score: THRESHOLDS.ambiguousExitScore, detail: 'Tests ran, outcome unclear', applicable: true };\n}\n\n/**\n * Score tool error rate, excluding benign errors from expected-to-fail tools.\n */\nfunction scoreToolErrors(results: ExtractedToolCall[]): DimensionResult {\n  if (results.length === 0) return { score: 1, detail: 'No tool calls', applicable: false };\n\n  // Only count errors from non-benign tools\n  const nonBenign = results.filter(r => !THRESHOLDS.benignErrorTools.includes(r.toolName));\n  if (nonBenign.length === 0) return { score: 1, detail: 'All tool calls are benign-error tools', applicable: true };\n\n  const errors = nonBenign.filter(r => r.isError);\n  const rate = errors.length / nonBenign.length;\n\n  if (rate === 0) return { score: 1, detail: 'No tool errors', applicable: true };\n\n  const score = Math.max(0, 1 - rate * THRESHOLDS.toolErrorPenaltyMultiplier);\n  return {\n    score,\n    detail: `${errors.length}/${nonBenign.length} tools errored (${(rate * 100).toFixed(0)}%)`,\n    applicable: true,\n  };\n}\n\n/**\n * Detect stuck loops: same tool + same args appearing 3+ times consecutively,\n * or same tool + same args erroring 3+ times total.\n */\nfunction scoreStuckLoops(results: ExtractedToolCall[]): DimensionResult {\n  if (results.length < 3) return { score: 1, detail: 'Too few calls for loop detection', applicable: true };\n\n  const fingerprint = (r: ExtractedToolCall) => `${r.toolName}:${JSON.stringify(r.args)}`;\n\n  // Check consecutive identical calls\n  let maxConsecutive = 1;\n  let currentRun = 1;\n  for (let i = 1; i < results.length; i++) {\n    if (fingerprint(results[i]!) === fingerprint(results[i - 1]!)) {\n      currentRun++;\n      maxConsecutive = Math.max(maxConsecutive, currentRun);\n    } else {\n      currentRun = 1;\n    }\n  }\n\n  // Check repeated error fingerprints\n  const errorCounts = new Map<string, number>();\n  for (const r of results) {\n    if (!r.isError) continue;\n    const fp = fingerprint(r);\n    errorCounts.set(fp, (errorCounts.get(fp) ?? 0) + 1);\n  }\n  const maxErrorRepeat = Math.max(0, ...[...errorCounts.values()]);\n\n  const minReps = THRESHOLDS.loopMinRepetitions;\n  const loopSeverity = Math.max(\n    maxConsecutive >= minReps ? maxConsecutive - (minReps - 1) : 0,\n    maxErrorRepeat >= minReps ? maxErrorRepeat - (minReps - 1) : 0,\n  );\n\n  if (loopSeverity === 0) return { score: 1, detail: 'No stuck loops', applicable: true };\n\n  const score = Math.max(0, 1 - loopSeverity * THRESHOLDS.loopPenaltyPerOccurrence);\n  const details: string[] = [];\n  if (maxConsecutive >= minReps) details.push(`${maxConsecutive} consecutive identical calls`);\n  if (maxErrorRepeat >= minReps) details.push(`same error repeated ${maxErrorRepeat}x`);\n  return { score, detail: details.join('; '), applicable: true };\n}\n\n/**\n * Detect regressions: build/test passed at some point, then failed later.\n * Gives partial credit if the agent recovered (final state passes).\n */\nfunction scoreRegression(execResults: ExtractedToolCall[]): DimensionResult {\n  const regressions: { label: string; recovered: boolean }[] = [];\n\n  for (const [label, filter] of [\n    ['Build', isBuildArgs],\n    ['Tests', isTestArgs],\n  ] as const) {\n    const cmds = execResults.filter(r => r.toolName === 'execute_command' && filter(r.args));\n    if (cmds.length < 2) continue;\n\n    let sawPass = false;\n    let sawRegression = false;\n    for (const cmd of cmds) {\n      const exit = getExitCode(cmd.result);\n      if (exit === 0) {\n        if (sawRegression) {\n          // Recovered from regression\n          regressions.push({ label, recovered: true });\n          sawRegression = false;\n        }\n        sawPass = true;\n      }\n      if (sawPass && exit !== null && exit !== 0) {\n        sawRegression = true;\n      }\n    }\n\n    if (sawRegression) {\n      // Regression persisted at end of session\n      regressions.push({ label, recovered: false });\n    }\n  }\n\n  if (regressions.length === 0) return { score: 1, detail: 'No regressions', applicable: true };\n\n  // Persistent regressions are severe (0), recovered regressions get partial credit (0.5)\n  const persistentCount = regressions.filter(r => !r.recovered).length;\n  const recoveredCount = regressions.filter(r => r.recovered).length;\n  const score = Math.max(0, 1 - persistentCount * 0.5 - recoveredCount * 0.1);\n  const details = regressions.map(\n    r => `${r.label} ${r.recovered ? 'regressed then recovered' : 'regressed (persisted)'}`,\n  );\n  return { score, detail: details.join('; '), applicable: true };\n}\n\n/**\n * Score autonomy: penalize ask_user calls.\n * 0 calls = 1.0, each call reduces by 0.25 (up to a floor of 0).\n */\nfunction scoreAutonomy(results: ExtractedToolCall[]): DimensionResult {\n  const askUserCalls = results.filter(r => r.toolName === 'ask_user');\n  const count = askUserCalls.length;\n\n  if (count === 0) return { score: 1, detail: 'No ask_user calls (fully autonomous)', applicable: true };\n\n  const score = Math.max(0, 1 - count * THRESHOLDS.autonomyPenaltyPerAsk);\n  return { score, detail: `${count} ask_user call${count > 1 ? 's' : ''}`, applicable: true };\n}\n\n/**\n * Compute weighted average of only applicable dimensions.\n * Returns null if no dimensions are applicable.\n */\nfunction weightedAverage(dimensions: Record<string, DimensionResult>, weights: Record<string, number>): number | null {\n  let totalWeight = 0;\n  let weightedSum = 0;\n\n  for (const [key, result] of Object.entries(dimensions)) {\n    if (!result.applicable) continue;\n    const weight = weights[key] ?? 0;\n    totalWeight += weight;\n    weightedSum += result.score * weight;\n  }\n\n  if (totalWeight === 0) return null;\n  return weightedSum / totalWeight;\n}\n\nexport function createOutcomeScorer() {\n  return createScorer({\n    id: 'mastracode-outcome',\n    name: 'MastraCode Outcome',\n    description:\n      'Grades coding session outcomes: build/test pass, tool errors, stuck loops, regressions, autonomy. Always-on.',\n    type: 'agent',\n    prepareRun: filterRun({ partTypes: ['tool-invocation', 'text'] }),\n  })\n    .preprocess(async ({ run }) => {\n      const messages = (run.output ?? []) as MastraDBMessage[];\n      const allResults = extractToolCalls(messages);\n      const totalCalls = allResults.length;\n\n      // H1: Empty sessions or sessions with no tool calls → score 0\n      if (totalCalls < THRESHOLDS.minToolCalls) {\n        const empty: DimensionResult = { score: 0, detail: 'No tool calls', applicable: false };\n        return {\n          empty: true as const,\n          build: empty,\n          tests: empty,\n          toolErrors: empty,\n          loops: empty,\n          regression: empty,\n          autonomy: empty,\n          totalCalls,\n        };\n      }\n\n      const execResults = allResults.filter(r => r.toolName === 'execute_command');\n\n      return {\n        empty: false as const,\n        build: scoreBuild(execResults),\n        tests: scoreTests(execResults),\n        toolErrors: scoreToolErrors(allResults),\n        loops: scoreStuckLoops(allResults),\n        regression: scoreRegression(execResults),\n        autonomy: scoreAutonomy(allResults),\n        totalCalls,\n      };\n    })\n    .generateScore(({ results }) => {\n      const p = results.preprocessStepResult;\n      if (p.empty) return 0;\n\n      const score = weightedAverage(\n        {\n          build: p.build,\n          tests: p.tests,\n          toolErrors: p.toolErrors,\n          loops: p.loops,\n          regression: p.regression,\n          autonomy: p.autonomy,\n        },\n        WEIGHTS,\n      );\n\n      return score === null ? 0 : Math.round(score * 100) / 100;\n    })\n    .generateReason(({ results, score }) => {\n      const p = results.preprocessStepResult;\n      if (p.empty) {\n        return `Score: 0 (${p.totalCalls} tool calls — below minimum threshold for scoring)`;\n      }\n\n      const pct = (w: number) => `${(w * 100).toFixed(0)}%`;\n      const dimLine = (name: string, weight: number, r: DimensionResult) =>\n        r.applicable\n          ? `${name} (${pct(weight)}): ${r.detail} [${r.score}]`\n          : `${name}: ${r.detail} [N/A — excluded from average]`;\n\n      const parts: string[] = [`Score: ${score} (${p.totalCalls} tool calls total)`];\n      parts.push(dimLine('Build', WEIGHTS.build, p.build));\n      parts.push(dimLine('Tests', WEIGHTS.tests, p.tests));\n      parts.push(dimLine('Tool errors', WEIGHTS.toolErrors, p.toolErrors));\n      parts.push(dimLine('Loops', WEIGHTS.loops, p.loops));\n      parts.push(dimLine('Regression', WEIGHTS.regression, p.regression));\n      parts.push(dimLine('Autonomy', WEIGHTS.autonomy, p.autonomy));\n\n      return parts.join('\\n');\n    });\n}\n"],"mappings":";;;;;;;;AAuCA,MAAM,UAAU;CACd,OAAO;CACP,OAAO;CACP,YAAY;CACZ,OAAO;CACP,YAAY;CACZ,UAAU;AACZ;;;;AAKA,MAAM,aAAa;;CAEjB,4BAA4B;;CAG5B,oBAAoB;;CAGpB,0BAA0B;;CAG1B,uBAAuB;;CAGvB,oBAAoB;;CAGpB,cAAc;;CAGd,kBAAkB;EAAC;EAAkB;EAAc;CAAa;AAClE;;;;;;AAWA,SAAS,YAAY,MAAwC;CAC3D,OAAO,eAAe,OAAO,KAAK,WAAW,EAAE,CAAC;AAClD;AAEA,SAAS,WAAW,MAAwC;CAC1D,OAAO,cAAc,OAAO,KAAK,WAAW,EAAE,CAAC;AACjD;;;;AAKA,SAAS,WAAW,aAAmD;CACrE,MAAM,SAAS,YAAY,QAAO,MAAK,EAAE,aAAa,qBAAqB,YAAY,EAAE,IAAI,CAAC;CAC9F,IAAI,OAAO,WAAW,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAA0B,YAAY;CAAM;CAEhG,MAAM,OAAO,OAAO,OAAO,SAAS;CACpC,IAAI,KAAK,SAAS,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAgC,YAAY;CAAK;CAE9F,MAAM,WAAW,YAAY,KAAK,MAAM;CACxC,IAAI,aAAa,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAA0B,YAAY;CAAK;CAC1F,IAAI,aAAa,QAAQ,aAAa,GACpC,OAAO;EAAE,OAAO;EAAG,QAAQ,sBAAsB,SAAS;EAAI,YAAY;CAAK;CAGjF,MAAM,OAAO,OAAO,KAAK,WAAW,WAAW,KAAK,SAAS,KAAK,UAAU,KAAK,UAAU,EAAE;CAC7F,IAAI,oDAAoD,KAAK,IAAI,GAC/D,OAAO;EAAE,OAAO;EAAG,QAAQ;EAA8C,YAAY;CAAK;CAG5F,OAAO;EAAE,OAAO,WAAW;EAAoB,QAAQ;EAA8B,YAAY;CAAK;AACxG;;;;AAKA,SAAS,WAAW,aAAmD;CACrE,MAAM,QAAQ,YAAY,QAAO,MAAK,EAAE,aAAa,qBAAqB,WAAW,EAAE,IAAI,CAAC;CAC5F,IAAI,MAAM,WAAW,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAgB,YAAY;CAAM;CAErF,MAAM,OAAO,MAAM,MAAM,SAAS;CAClC,IAAI,KAAK,SAAS,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAwB,YAAY;CAAK;CAEtF,MAAM,WAAW,YAAY,KAAK,MAAM;CACxC,IAAI,aAAa,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAgB,YAAY;CAAK;CAChF,IAAI,aAAa,QAAQ,aAAa,GACpC,OAAO;EAAE,OAAO;EAAG,QAAQ,sBAAsB,SAAS;EAAI,YAAY;CAAK;CAEjF,MAAM,OAAO,OAAO,KAAK,WAAW,WAAW,KAAK,SAAS,KAAK,UAAU,KAAK,UAAU,EAAE;CAC7F,IAAI,iCAAiC,KAAK,IAAI,KAAK,CAAC,iCAAiC,KAAK,IAAI,GAC5F,OAAO;EAAE,OAAO;EAAG,QAAQ;EAA2B,YAAY;CAAK;CAEzE,IAAI,uBAAuB,KAAK,IAAI,GAClC,OAAO;EAAE,OAAO;EAAG,QAAQ;EAA2B,YAAY;CAAK;CAGzE,OAAO;EAAE,OAAO,WAAW;EAAoB,QAAQ;EAA8B,YAAY;CAAK;AACxG;;;;AAKA,SAAS,gBAAgB,SAA+C;CACtE,IAAI,QAAQ,WAAW,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAiB,YAAY;CAAM;CAGxF,MAAM,YAAY,QAAQ,QAAO,MAAK,CAAC,WAAW,iBAAiB,SAAS,EAAE,QAAQ,CAAC;CACvF,IAAI,UAAU,WAAW,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAyC,YAAY;CAAK;CAEjH,MAAM,SAAS,UAAU,QAAO,MAAK,EAAE,OAAO;CAC9C,MAAM,OAAO,OAAO,SAAS,UAAU;CAEvC,IAAI,SAAS,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAkB,YAAY;CAAK;CAG9E,OAAO;EACL,OAFY,KAAK,IAAI,GAAG,IAAI,OAAO,WAAW,0BAE1C;EACJ,QAAQ,GAAG,OAAO,OAAO,GAAG,UAAU,OAAO,mBAAmB,OAAO,IAAA,CAAK,QAAQ,CAAC,EAAE;EACvF,YAAY;CACd;AACF;;;;;AAMA,SAAS,gBAAgB,SAA+C;CACtE,IAAI,QAAQ,SAAS,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAoC,YAAY;CAAK;CAExG,MAAM,eAAe,MAAyB,GAAG,EAAE,SAAS,GAAG,KAAK,UAAU,EAAE,IAAI;CAGpF,IAAI,iBAAiB;CACrB,IAAI,aAAa;CACjB,KAAK,IAAI,IAAI,GAAG,IAAI,QAAQ,QAAQ,KAClC,IAAI,YAAY,QAAQ,EAAG,MAAM,YAAY,QAAQ,IAAI,EAAG,GAAG;EAC7D;EACA,iBAAiB,KAAK,IAAI,gBAAgB,UAAU;CACtD,OACE,aAAa;CAKjB,MAAM,8BAAc,IAAI,IAAoB;CAC5C,KAAK,MAAM,KAAK,SAAS;EACvB,IAAI,CAAC,EAAE,SAAS;EAChB,MAAM,KAAK,YAAY,CAAC;EACxB,YAAY,IAAI,KAAK,YAAY,IAAI,EAAE,KAAK,KAAK,CAAC;CACpD;CACA,MAAM,iBAAiB,KAAK,IAAI,GAAG,GAAG,CAAC,GAAG,YAAY,OAAO,CAAC,CAAC;CAE/D,MAAM,UAAU,WAAW;CAC3B,MAAM,eAAe,KAAK,IACxB,kBAAkB,UAAU,kBAAkB,UAAU,KAAK,GAC7D,kBAAkB,UAAU,kBAAkB,UAAU,KAAK,CAC/D;CAEA,IAAI,iBAAiB,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAkB,YAAY;CAAK;CAEtF,MAAM,QAAQ,KAAK,IAAI,GAAG,IAAI,eAAe,WAAW,wBAAwB;CAChF,MAAM,UAAoB,CAAC;CAC3B,IAAI,kBAAkB,SAAS,QAAQ,KAAK,GAAG,eAAe,6BAA6B;CAC3F,IAAI,kBAAkB,SAAS,QAAQ,KAAK,uBAAuB,eAAe,EAAE;CACpF,OAAO;EAAE;EAAO,QAAQ,QAAQ,KAAK,IAAI;EAAG,YAAY;CAAK;AAC/D;;;;;AAMA,SAAS,gBAAgB,aAAmD;CAC1E,MAAM,cAAuD,CAAC;CAE9D,KAAK,MAAM,CAAC,OAAO,WAAW,CAC5B,CAAC,SAAS,WAAW,GACrB,CAAC,SAAS,UAAU,CACtB,GAAY;EACV,MAAM,OAAO,YAAY,QAAO,MAAK,EAAE,aAAa,qBAAqB,OAAO,EAAE,IAAI,CAAC;EACvF,IAAI,KAAK,SAAS,GAAG;EAErB,IAAI,UAAU;EACd,IAAI,gBAAgB;EACpB,KAAK,MAAM,OAAO,MAAM;GACtB,MAAM,OAAO,YAAY,IAAI,MAAM;GACnC,IAAI,SAAS,GAAG;IACd,IAAI,eAAe;KAEjB,YAAY,KAAK;MAAE;MAAO,WAAW;KAAK,CAAC;KAC3C,gBAAgB;IAClB;IACA,UAAU;GACZ;GACA,IAAI,WAAW,SAAS,QAAQ,SAAS,GACvC,gBAAgB;EAEpB;EAEA,IAAI,eAEF,YAAY,KAAK;GAAE;GAAO,WAAW;EAAM,CAAC;CAEhD;CAEA,IAAI,YAAY,WAAW,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAkB,YAAY;CAAK;CAG5F,MAAM,kBAAkB,YAAY,QAAO,MAAK,CAAC,EAAE,SAAS,CAAC,CAAC;CAC9D,MAAM,iBAAiB,YAAY,QAAO,MAAK,EAAE,SAAS,CAAC,CAAC;CAK5D,OAAO;EAAE,OAJK,KAAK,IAAI,GAAG,IAAI,kBAAkB,KAAM,iBAAiB,EAI1D;EAAG,QAHA,YAAY,KAC1B,MAAK,GAAG,EAAE,MAAM,GAAG,EAAE,YAAY,6BAA6B,yBAElC,CAAC,CAAC,KAAK,IAAI;EAAG,YAAY;CAAK;AAC/D;;;;;AAMA,SAAS,cAAc,SAA+C;CAEpE,MAAM,QADe,QAAQ,QAAO,MAAK,EAAE,aAAa,UAC/B,CAAC,CAAC;CAE3B,IAAI,UAAU,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAwC,YAAY;CAAK;CAGrG,OAAO;EAAE,OADK,KAAK,IAAI,GAAG,IAAI,QAAQ,WAAW,qBACpC;EAAG,QAAQ,GAAG,MAAM,gBAAgB,QAAQ,IAAI,MAAM;EAAM,YAAY;CAAK;AAC5F;;;;;AAMA,SAAS,gBAAgB,YAA6C,SAAgD;CACpH,IAAI,cAAc;CAClB,IAAI,cAAc;CAElB,KAAK,MAAM,CAAC,KAAK,WAAW,OAAO,QAAQ,UAAU,GAAG;EACtD,IAAI,CAAC,OAAO,YAAY;EACxB,MAAM,SAAS,QAAQ,QAAQ;EAC/B,eAAe;EACf,eAAe,OAAO,QAAQ;CAChC;CAEA,IAAI,gBAAgB,GAAG,OAAO;CAC9B,OAAO,cAAc;AACvB;AAEA,SAAgB,sBAAsB;CACpC,OAAO,aAAa;EAClB,IAAI;EACJ,MAAM;EACN,aACE;EACF,MAAM;EACN,YAAY,UAAU,EAAE,WAAW,CAAC,mBAAmB,MAAM,EAAE,CAAC;CAClE,CAAC,CAAC,CACC,WAAW,OAAO,EAAE,UAAU;EAC7B,MAAM,WAAY,IAAI,UAAU,CAAC;EACjC,MAAM,aAAa,iBAAiB,QAAQ;EAC5C,MAAM,aAAa,WAAW;EAG9B,IAAI,aAAa,WAAW,cAAc;GACxC,MAAM,QAAyB;IAAE,OAAO;IAAG,QAAQ;IAAiB,YAAY;GAAM;GACtF,OAAO;IACL,OAAO;IACP,OAAO;IACP,OAAO;IACP,YAAY;IACZ,OAAO;IACP,YAAY;IACZ,UAAU;IACV;GACF;EACF;EAEA,MAAM,cAAc,WAAW,QAAO,MAAK,EAAE,aAAa,iBAAiB;EAE3E,OAAO;GACL,OAAO;GACP,OAAO,WAAW,WAAW;GAC7B,OAAO,WAAW,WAAW;GAC7B,YAAY,gBAAgB,UAAU;GACtC,OAAO,gBAAgB,UAAU;GACjC,YAAY,gBAAgB,WAAW;GACvC,UAAU,cAAc,UAAU;GAClC;EACF;CACF,CAAC,CAAC,CACD,eAAe,EAAE,cAAc;EAC9B,MAAM,IAAI,QAAQ;EAClB,IAAI,EAAE,OAAO,OAAO;EAEpB,MAAM,QAAQ,gBACZ;GACE,OAAO,EAAE;GACT,OAAO,EAAE;GACT,YAAY,EAAE;GACd,OAAO,EAAE;GACT,YAAY,EAAE;GACd,UAAU,EAAE;EACd,GACA,OACF;EAEA,OAAO,UAAU,OAAO,IAAI,KAAK,MAAM,QAAQ,GAAG,IAAI;CACxD,CAAC,CAAC,CACD,gBAAgB,EAAE,SAAS,YAAY;EACtC,MAAM,IAAI,QAAQ;EAClB,IAAI,EAAE,OACJ,OAAO,aAAa,EAAE,WAAW;EAGnC,MAAM,OAAO,MAAc,IAAI,IAAI,IAAA,CAAK,QAAQ,CAAC,EAAE;EACnD,MAAM,WAAW,MAAc,QAAgB,MAC7C,EAAE,aACE,GAAG,KAAK,IAAI,IAAI,MAAM,EAAE,KAAK,EAAE,OAAO,IAAI,EAAE,MAAM,KAClD,GAAG,KAAK,IAAI,EAAE,OAAO;EAE3B,MAAM,QAAkB,CAAC,UAAU,MAAM,IAAI,EAAE,WAAW,mBAAmB;EAC7E,MAAM,KAAK,QAAQ,SAAS,QAAQ,OAAO,EAAE,KAAK,CAAC;EACnD,MAAM,KAAK,QAAQ,SAAS,QAAQ,OAAO,EAAE,KAAK,CAAC;EACnD,MAAM,KAAK,QAAQ,eAAe,QAAQ,YAAY,EAAE,UAAU,CAAC;EACnE,MAAM,KAAK,QAAQ,SAAS,QAAQ,OAAO,EAAE,KAAK,CAAC;EACnD,MAAM,KAAK,QAAQ,cAAc,QAAQ,YAAY,EAAE,UAAU,CAAC;EAClE,MAAM,KAAK,QAAQ,YAAY,QAAQ,UAAU,EAAE,QAAQ,CAAC;EAE5D,OAAO,MAAM,KAAK,IAAI;CACxB,CAAC;AACL"}
+{"version":3,"file":"outcome.js","names":[],"sources":["../../../src/evals/scorers/outcome.ts"],"sourcesContent":["/**\n * MastraCode Outcome Scorer (Code-based, Agent type, always-on)\n *\n * Grades the objective outcome of a coding session by checking hard facts:\n * - Did the build/typecheck pass?\n * - Did tests pass?\n * - What was the tool error rate?\n * - Did the agent get stuck in a loop?\n * - Was there a regression (build passed earlier, failed later)?\n * - Did the agent use ask_user unnecessarily? (autonomy)\n *\n * This scorer is always-on (no sampling) because it's cheap — it only\n * reads exit codes and error states from tool results already in the messages.\n *\n * Scoring philosophy (from Anthropic's eval guidance):\n * \"Grade outcomes, not paths\" — we check what happened, not how.\n * Partial credit: a session that passes build but fails tests scores\n * higher than one that fails build entirely.\n *\n * Dimensions that don't apply to a session (e.g. build for a docs-only task)\n * are excluded from the weighted average, so sessions aren't penalized for\n * irrelevant criteria.\n */\n\nimport type { MastraDBMessage } from '@mastra/core/agent';\nimport { createScorer, filterRun } from '@mastra/core/evals';\n\nimport { isBuildCommand, isTestCommand, getExitCode, isSuccessResult } from './classify-command.js';\nimport type { ExtractedToolCall } from './extract-tools.js';\nimport { extractToolCalls } from './extract-tools.js';\n\n// ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\n// CONFIGURATION — Adjust weights and thresholds here\n// ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\n\n/**\n * Dimension weights. Used to compute the weighted average of applicable dimensions.\n * Dimensions that don't apply (e.g. no build ran) are excluded from the average.\n */\nconst WEIGHTS = {\n  build: 0.3,\n  tests: 0.25,\n  toolErrors: 0.2,\n  loops: 0.1,\n  regression: 0.1,\n  autonomy: 0.05,\n} as const;\n\n/**\n * Scoring thresholds and multipliers.\n */\nconst THRESHOLDS = {\n  /** Multiplier for tool error rate penalty. Score = max(0, 1 - rate × this). */\n  toolErrorPenaltyMultiplier: 1.0,\n\n  /** How many consecutive identical calls (or same-error repeats) before it's a \"loop\". */\n  loopMinRepetitions: 3,\n\n  /** Score penalty per detected loop. */\n  loopPenaltyPerOccurrence: 0.3,\n\n  /** Score deducted per ask_user call. */\n  autonomyPenaltyPerAsk: 0.25,\n\n  /** Score when a command ran but exit code is ambiguous. */\n  ambiguousExitScore: 0.75,\n\n  /** Minimum tool calls required to produce a meaningful score. */\n  minToolCalls: 1,\n\n  /** Tool names whose errors are expected/benign (not penalized). */\n  benignErrorTools: ['search_content', 'find_files', 'lsp_inspect'] as string[],\n} as const;\n\n// ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\n\ntype DimensionResult = { score: number; detail: string; applicable: boolean };\n\n/**\n * Check if the primary verb of an execute_command is a build/typecheck command.\n * Splits compound commands (&&, ;) and checks the first segment.\n */\n/** Extract command string from tool call args, then delegate to shared classifier. */\nfunction isBuildArgs(args: Record<string, unknown>): boolean {\n  return isBuildCommand(String(args.command ?? ''));\n}\n\nfunction isTestArgs(args: Record<string, unknown>): boolean {\n  return isTestCommand(String(args.command ?? ''));\n}\n\nfunction isForegroundCommand(result: ExtractedToolCall): boolean {\n  return result.args.background !== true;\n}\n\n/**\n * Score build/typecheck outcome (final state).\n */\nfunction scoreBuild(execResults: ExtractedToolCall[]): DimensionResult {\n  const builds = execResults.filter(\n    r => r.toolName === 'execute_command' && isForegroundCommand(r) && isBuildArgs(r.args),\n  );\n  if (builds.length === 0) return { score: 0, detail: 'No build/typecheck ran', applicable: false };\n\n  const last = builds[builds.length - 1]!;\n  if (last.isError) return { score: 0, detail: 'Build/typecheck tool errored', applicable: true };\n\n  const exitCode = getExitCode(last.result);\n  if (exitCode !== null && exitCode !== 0)\n    return { score: 0, detail: `Build failed (exit ${exitCode})`, applicable: true };\n  if (typeof last.result === 'string') return { score: 1, detail: 'Build/typecheck passed', applicable: true };\n\n  const text = JSON.stringify(last.result ?? '');\n  if (/error TS\\d+|Cannot find module|is not assignable/i.test(text)) {\n    return { score: 0, detail: 'Build failed (TypeScript errors in output)', applicable: true };\n  }\n  if (isSuccessResult(last.result)) return { score: 1, detail: 'Build/typecheck passed', applicable: true };\n\n  return { score: THRESHOLDS.ambiguousExitScore, detail: 'Build ran, outcome unclear', applicable: true };\n}\n\n/**\n * Score test outcome (final state).\n */\nfunction scoreTests(execResults: ExtractedToolCall[]): DimensionResult {\n  const tests = execResults.filter(\n    r => r.toolName === 'execute_command' && isForegroundCommand(r) && isTestArgs(r.args),\n  );\n  if (tests.length === 0) return { score: 0, detail: 'No tests ran', applicable: false };\n\n  const last = tests[tests.length - 1]!;\n  if (last.isError) return { score: 0, detail: 'Test command errored', applicable: true };\n\n  const exitCode = getExitCode(last.result);\n  if (exitCode !== null && exitCode !== 0)\n    return { score: 0, detail: `Tests failed (exit ${exitCode})`, applicable: true };\n  if (typeof last.result === 'string') return { score: 1, detail: 'Tests passed', applicable: true };\n\n  const text = JSON.stringify(last.result ?? '');\n  if (/\\bFAIL\\b|failed|✗|✘/i.test(text)) {\n    return { score: 0, detail: 'Tests failed (inferred)', applicable: true };\n  }\n  if (/\\d+ (?:tests? )?passed|✓|PASS/i.test(text) && !/\\bfail(?:ed|ure)?\\b|\\berror\\b/i.test(text)) {\n    return { score: 1, detail: 'Tests passed (inferred)', applicable: true };\n  }\n  if (isSuccessResult(last.result)) return { score: 1, detail: 'Tests passed', applicable: true };\n\n  return { score: THRESHOLDS.ambiguousExitScore, detail: 'Tests ran, outcome unclear', applicable: true };\n}\n\n/**\n * Score tool error rate, excluding benign errors from expected-to-fail tools.\n */\nfunction scoreToolErrors(results: ExtractedToolCall[]): DimensionResult {\n  if (results.length === 0) return { score: 1, detail: 'No tool calls', applicable: false };\n\n  // Only count errors from non-benign tools\n  const nonBenign = results.filter(r => !THRESHOLDS.benignErrorTools.includes(r.toolName));\n  if (nonBenign.length === 0) return { score: 1, detail: 'All tool calls are benign-error tools', applicable: true };\n\n  const errors = nonBenign.filter(r => r.isError);\n  const rate = errors.length / nonBenign.length;\n\n  if (rate === 0) return { score: 1, detail: 'No tool errors', applicable: true };\n\n  const score = Math.max(0, 1 - rate * THRESHOLDS.toolErrorPenaltyMultiplier);\n  return {\n    score,\n    detail: `${errors.length}/${nonBenign.length} tools errored (${(rate * 100).toFixed(0)}%)`,\n    applicable: true,\n  };\n}\n\n/**\n * Detect stuck loops: same tool + same args appearing 3+ times consecutively,\n * or same tool + same args erroring 3+ times total.\n */\nfunction scoreStuckLoops(results: ExtractedToolCall[]): DimensionResult {\n  if (results.length < 3) return { score: 1, detail: 'Too few calls for loop detection', applicable: true };\n\n  const fingerprint = (r: ExtractedToolCall) => `${r.toolName}:${JSON.stringify(r.args)}`;\n\n  // Check consecutive identical calls\n  let maxConsecutive = 1;\n  let currentRun = 1;\n  for (let i = 1; i < results.length; i++) {\n    if (fingerprint(results[i]!) === fingerprint(results[i - 1]!)) {\n      currentRun++;\n      maxConsecutive = Math.max(maxConsecutive, currentRun);\n    } else {\n      currentRun = 1;\n    }\n  }\n\n  // Check repeated error fingerprints\n  const errorCounts = new Map<string, number>();\n  for (const r of results) {\n    if (!r.isError) continue;\n    const fp = fingerprint(r);\n    errorCounts.set(fp, (errorCounts.get(fp) ?? 0) + 1);\n  }\n  const maxErrorRepeat = Math.max(0, ...[...errorCounts.values()]);\n\n  const minReps = THRESHOLDS.loopMinRepetitions;\n  const loopSeverity = Math.max(\n    maxConsecutive >= minReps ? maxConsecutive - (minReps - 1) : 0,\n    maxErrorRepeat >= minReps ? maxErrorRepeat - (minReps - 1) : 0,\n  );\n\n  if (loopSeverity === 0) return { score: 1, detail: 'No stuck loops', applicable: true };\n\n  const score = Math.max(0, 1 - loopSeverity * THRESHOLDS.loopPenaltyPerOccurrence);\n  const details: string[] = [];\n  if (maxConsecutive >= minReps) details.push(`${maxConsecutive} consecutive identical calls`);\n  if (maxErrorRepeat >= minReps) details.push(`same error repeated ${maxErrorRepeat}x`);\n  return { score, detail: details.join('; '), applicable: true };\n}\n\n/**\n * Detect regressions: build/test passed at some point, then failed later.\n * Gives partial credit if the agent recovered (final state passes).\n */\nfunction scoreRegression(execResults: ExtractedToolCall[]): DimensionResult {\n  const regressions: { label: string; recovered: boolean }[] = [];\n\n  for (const [label, filter] of [\n    ['Build', isBuildArgs],\n    ['Tests', isTestArgs],\n  ] as const) {\n    const cmds = execResults.filter(r => r.toolName === 'execute_command' && isForegroundCommand(r) && filter(r.args));\n    if (cmds.length < 2) continue;\n\n    let sawPass = false;\n    let sawRegression = false;\n    for (const cmd of cmds) {\n      const exitCode = getExitCode(cmd.result);\n      const passed = isSuccessResult(cmd.result, cmd.isError);\n      const failed = cmd.isError || (exitCode !== null && exitCode !== 0);\n\n      if (passed) {\n        if (sawRegression) {\n          // Recovered from regression\n          regressions.push({ label, recovered: true });\n          sawRegression = false;\n        }\n        sawPass = true;\n      }\n      if (sawPass && failed) {\n        sawRegression = true;\n      }\n    }\n\n    if (sawRegression) {\n      // Regression persisted at end of session\n      regressions.push({ label, recovered: false });\n    }\n  }\n\n  if (regressions.length === 0) return { score: 1, detail: 'No regressions', applicable: true };\n\n  // Persistent regressions are severe (0), recovered regressions get partial credit (0.5)\n  const persistentCount = regressions.filter(r => !r.recovered).length;\n  const recoveredCount = regressions.filter(r => r.recovered).length;\n  const score = Math.max(0, 1 - persistentCount * 0.5 - recoveredCount * 0.1);\n  const details = regressions.map(\n    r => `${r.label} ${r.recovered ? 'regressed then recovered' : 'regressed (persisted)'}`,\n  );\n  return { score, detail: details.join('; '), applicable: true };\n}\n\n/**\n * Score autonomy: penalize ask_user calls.\n * 0 calls = 1.0, each call reduces by 0.25 (up to a floor of 0).\n */\nfunction scoreAutonomy(results: ExtractedToolCall[]): DimensionResult {\n  const askUserCalls = results.filter(r => r.toolName === 'ask_user');\n  const count = askUserCalls.length;\n\n  if (count === 0) return { score: 1, detail: 'No ask_user calls (fully autonomous)', applicable: true };\n\n  const score = Math.max(0, 1 - count * THRESHOLDS.autonomyPenaltyPerAsk);\n  return { score, detail: `${count} ask_user call${count > 1 ? 's' : ''}`, applicable: true };\n}\n\n/**\n * Compute weighted average of only applicable dimensions.\n * Returns null if no dimensions are applicable.\n */\nfunction weightedAverage(dimensions: Record<string, DimensionResult>, weights: Record<string, number>): number | null {\n  let totalWeight = 0;\n  let weightedSum = 0;\n\n  for (const [key, result] of Object.entries(dimensions)) {\n    if (!result.applicable) continue;\n    const weight = weights[key] ?? 0;\n    totalWeight += weight;\n    weightedSum += result.score * weight;\n  }\n\n  if (totalWeight === 0) return null;\n  return weightedSum / totalWeight;\n}\n\nexport function createOutcomeScorer() {\n  return createScorer({\n    id: 'mastracode-outcome',\n    name: 'MastraCode Outcome',\n    description:\n      'Grades coding session outcomes: build/test pass, tool errors, stuck loops, regressions, autonomy. Always-on.',\n    type: 'agent',\n    prepareRun: filterRun({ partTypes: ['tool-invocation', 'text'] }),\n  })\n    .preprocess(async ({ run }) => {\n      const messages = (run.output ?? []) as MastraDBMessage[];\n      const allResults = extractToolCalls(messages);\n      const totalCalls = allResults.length;\n\n      // H1: Empty sessions or sessions with no tool calls → score 0\n      if (totalCalls < THRESHOLDS.minToolCalls) {\n        const empty: DimensionResult = { score: 0, detail: 'No tool calls', applicable: false };\n        return {\n          empty: true as const,\n          build: empty,\n          tests: empty,\n          toolErrors: empty,\n          loops: empty,\n          regression: empty,\n          autonomy: empty,\n          totalCalls,\n        };\n      }\n\n      const execResults = allResults.filter(r => r.toolName === 'execute_command');\n\n      return {\n        empty: false as const,\n        build: scoreBuild(execResults),\n        tests: scoreTests(execResults),\n        toolErrors: scoreToolErrors(allResults),\n        loops: scoreStuckLoops(allResults),\n        regression: scoreRegression(execResults),\n        autonomy: scoreAutonomy(allResults),\n        totalCalls,\n      };\n    })\n    .generateScore(({ results }) => {\n      const p = results.preprocessStepResult;\n      if (p.empty) return 0;\n\n      const score = weightedAverage(\n        {\n          build: p.build,\n          tests: p.tests,\n          toolErrors: p.toolErrors,\n          loops: p.loops,\n          regression: p.regression,\n          autonomy: p.autonomy,\n        },\n        WEIGHTS,\n      );\n\n      return score === null ? 0 : Math.round(score * 100) / 100;\n    })\n    .generateReason(({ results, score }) => {\n      const p = results.preprocessStepResult;\n      if (p.empty) {\n        return `Score: 0 (${p.totalCalls} tool calls — below minimum threshold for scoring)`;\n      }\n\n      const pct = (w: number) => `${(w * 100).toFixed(0)}%`;\n      const dimLine = (name: string, weight: number, r: DimensionResult) =>\n        r.applicable\n          ? `${name} (${pct(weight)}): ${r.detail} [${r.score}]`\n          : `${name}: ${r.detail} [N/A — excluded from average]`;\n\n      const parts: string[] = [`Score: ${score} (${p.totalCalls} tool calls total)`];\n      parts.push(dimLine('Build', WEIGHTS.build, p.build));\n      parts.push(dimLine('Tests', WEIGHTS.tests, p.tests));\n      parts.push(dimLine('Tool errors', WEIGHTS.toolErrors, p.toolErrors));\n      parts.push(dimLine('Loops', WEIGHTS.loops, p.loops));\n      parts.push(dimLine('Regression', WEIGHTS.regression, p.regression));\n      parts.push(dimLine('Autonomy', WEIGHTS.autonomy, p.autonomy));\n\n      return parts.join('\\n');\n    });\n}\n"],"mappings":";;;;;;;;AAuCA,MAAM,UAAU;CACd,OAAO;CACP,OAAO;CACP,YAAY;CACZ,OAAO;CACP,YAAY;CACZ,UAAU;AACZ;;;;AAKA,MAAM,aAAa;;CAEjB,4BAA4B;;CAG5B,oBAAoB;;CAGpB,0BAA0B;;CAG1B,uBAAuB;;CAGvB,oBAAoB;;CAGpB,cAAc;;CAGd,kBAAkB;EAAC;EAAkB;EAAc;CAAa;AAClE;;;;;;AAWA,SAAS,YAAY,MAAwC;CAC3D,OAAO,eAAe,OAAO,KAAK,WAAW,EAAE,CAAC;AAClD;AAEA,SAAS,WAAW,MAAwC;CAC1D,OAAO,cAAc,OAAO,KAAK,WAAW,EAAE,CAAC;AACjD;AAEA,SAAS,oBAAoB,QAAoC;CAC/D,OAAO,OAAO,KAAK,eAAe;AACpC;;;;AAKA,SAAS,WAAW,aAAmD;CACrE,MAAM,SAAS,YAAY,QACzB,MAAK,EAAE,aAAa,qBAAqB,oBAAoB,CAAC,KAAK,YAAY,EAAE,IAAI,CACvF;CACA,IAAI,OAAO,WAAW,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAA0B,YAAY;CAAM;CAEhG,MAAM,OAAO,OAAO,OAAO,SAAS;CACpC,IAAI,KAAK,SAAS,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAgC,YAAY;CAAK;CAE9F,MAAM,WAAW,YAAY,KAAK,MAAM;CACxC,IAAI,aAAa,QAAQ,aAAa,GACpC,OAAO;EAAE,OAAO;EAAG,QAAQ,sBAAsB,SAAS;EAAI,YAAY;CAAK;CACjF,IAAI,OAAO,KAAK,WAAW,UAAU,OAAO;EAAE,OAAO;EAAG,QAAQ;EAA0B,YAAY;CAAK;CAE3G,MAAM,OAAO,KAAK,UAAU,KAAK,UAAU,EAAE;CAC7C,IAAI,oDAAoD,KAAK,IAAI,GAC/D,OAAO;EAAE,OAAO;EAAG,QAAQ;EAA8C,YAAY;CAAK;CAE5F,IAAI,gBAAgB,KAAK,MAAM,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAA0B,YAAY;CAAK;CAExG,OAAO;EAAE,OAAO,WAAW;EAAoB,QAAQ;EAA8B,YAAY;CAAK;AACxG;;;;AAKA,SAAS,WAAW,aAAmD;CACrE,MAAM,QAAQ,YAAY,QACxB,MAAK,EAAE,aAAa,qBAAqB,oBAAoB,CAAC,KAAK,WAAW,EAAE,IAAI,CACtF;CACA,IAAI,MAAM,WAAW,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAgB,YAAY;CAAM;CAErF,MAAM,OAAO,MAAM,MAAM,SAAS;CAClC,IAAI,KAAK,SAAS,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAwB,YAAY;CAAK;CAEtF,MAAM,WAAW,YAAY,KAAK,MAAM;CACxC,IAAI,aAAa,QAAQ,aAAa,GACpC,OAAO;EAAE,OAAO;EAAG,QAAQ,sBAAsB,SAAS;EAAI,YAAY;CAAK;CACjF,IAAI,OAAO,KAAK,WAAW,UAAU,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAgB,YAAY;CAAK;CAEjG,MAAM,OAAO,KAAK,UAAU,KAAK,UAAU,EAAE;CAC7C,IAAI,uBAAuB,KAAK,IAAI,GAClC,OAAO;EAAE,OAAO;EAAG,QAAQ;EAA2B,YAAY;CAAK;CAEzE,IAAI,iCAAiC,KAAK,IAAI,KAAK,CAAC,iCAAiC,KAAK,IAAI,GAC5F,OAAO;EAAE,OAAO;EAAG,QAAQ;EAA2B,YAAY;CAAK;CAEzE,IAAI,gBAAgB,KAAK,MAAM,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAgB,YAAY;CAAK;CAE9F,OAAO;EAAE,OAAO,WAAW;EAAoB,QAAQ;EAA8B,YAAY;CAAK;AACxG;;;;AAKA,SAAS,gBAAgB,SAA+C;CACtE,IAAI,QAAQ,WAAW,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAiB,YAAY;CAAM;CAGxF,MAAM,YAAY,QAAQ,QAAO,MAAK,CAAC,WAAW,iBAAiB,SAAS,EAAE,QAAQ,CAAC;CACvF,IAAI,UAAU,WAAW,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAyC,YAAY;CAAK;CAEjH,MAAM,SAAS,UAAU,QAAO,MAAK,EAAE,OAAO;CAC9C,MAAM,OAAO,OAAO,SAAS,UAAU;CAEvC,IAAI,SAAS,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAkB,YAAY;CAAK;CAG9E,OAAO;EACL,OAFY,KAAK,IAAI,GAAG,IAAI,OAAO,WAAW,0BAE1C;EACJ,QAAQ,GAAG,OAAO,OAAO,GAAG,UAAU,OAAO,mBAAmB,OAAO,IAAA,CAAK,QAAQ,CAAC,EAAE;EACvF,YAAY;CACd;AACF;;;;;AAMA,SAAS,gBAAgB,SAA+C;CACtE,IAAI,QAAQ,SAAS,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAoC,YAAY;CAAK;CAExG,MAAM,eAAe,MAAyB,GAAG,EAAE,SAAS,GAAG,KAAK,UAAU,EAAE,IAAI;CAGpF,IAAI,iBAAiB;CACrB,IAAI,aAAa;CACjB,KAAK,IAAI,IAAI,GAAG,IAAI,QAAQ,QAAQ,KAClC,IAAI,YAAY,QAAQ,EAAG,MAAM,YAAY,QAAQ,IAAI,EAAG,GAAG;EAC7D;EACA,iBAAiB,KAAK,IAAI,gBAAgB,UAAU;CACtD,OACE,aAAa;CAKjB,MAAM,8BAAc,IAAI,IAAoB;CAC5C,KAAK,MAAM,KAAK,SAAS;EACvB,IAAI,CAAC,EAAE,SAAS;EAChB,MAAM,KAAK,YAAY,CAAC;EACxB,YAAY,IAAI,KAAK,YAAY,IAAI,EAAE,KAAK,KAAK,CAAC;CACpD;CACA,MAAM,iBAAiB,KAAK,IAAI,GAAG,GAAG,CAAC,GAAG,YAAY,OAAO,CAAC,CAAC;CAE/D,MAAM,UAAU,WAAW;CAC3B,MAAM,eAAe,KAAK,IACxB,kBAAkB,UAAU,kBAAkB,UAAU,KAAK,GAC7D,kBAAkB,UAAU,kBAAkB,UAAU,KAAK,CAC/D;CAEA,IAAI,iBAAiB,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAkB,YAAY;CAAK;CAEtF,MAAM,QAAQ,KAAK,IAAI,GAAG,IAAI,eAAe,WAAW,wBAAwB;CAChF,MAAM,UAAoB,CAAC;CAC3B,IAAI,kBAAkB,SAAS,QAAQ,KAAK,GAAG,eAAe,6BAA6B;CAC3F,IAAI,kBAAkB,SAAS,QAAQ,KAAK,uBAAuB,eAAe,EAAE;CACpF,OAAO;EAAE;EAAO,QAAQ,QAAQ,KAAK,IAAI;EAAG,YAAY;CAAK;AAC/D;;;;;AAMA,SAAS,gBAAgB,aAAmD;CAC1E,MAAM,cAAuD,CAAC;CAE9D,KAAK,MAAM,CAAC,OAAO,WAAW,CAC5B,CAAC,SAAS,WAAW,GACrB,CAAC,SAAS,UAAU,CACtB,GAAY;EACV,MAAM,OAAO,YAAY,QAAO,MAAK,EAAE,aAAa,qBAAqB,oBAAoB,CAAC,KAAK,OAAO,EAAE,IAAI,CAAC;EACjH,IAAI,KAAK,SAAS,GAAG;EAErB,IAAI,UAAU;EACd,IAAI,gBAAgB;EACpB,KAAK,MAAM,OAAO,MAAM;GACtB,MAAM,WAAW,YAAY,IAAI,MAAM;GACvC,MAAM,SAAS,gBAAgB,IAAI,QAAQ,IAAI,OAAO;GACtD,MAAM,SAAS,IAAI,WAAY,aAAa,QAAQ,aAAa;GAEjE,IAAI,QAAQ;IACV,IAAI,eAAe;KAEjB,YAAY,KAAK;MAAE;MAAO,WAAW;KAAK,CAAC;KAC3C,gBAAgB;IAClB;IACA,UAAU;GACZ;GACA,IAAI,WAAW,QACb,gBAAgB;EAEpB;EAEA,IAAI,eAEF,YAAY,KAAK;GAAE;GAAO,WAAW;EAAM,CAAC;CAEhD;CAEA,IAAI,YAAY,WAAW,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAkB,YAAY;CAAK;CAG5F,MAAM,kBAAkB,YAAY,QAAO,MAAK,CAAC,EAAE,SAAS,CAAC,CAAC;CAC9D,MAAM,iBAAiB,YAAY,QAAO,MAAK,EAAE,SAAS,CAAC,CAAC;CAK5D,OAAO;EAAE,OAJK,KAAK,IAAI,GAAG,IAAI,kBAAkB,KAAM,iBAAiB,EAI1D;EAAG,QAHA,YAAY,KAC1B,MAAK,GAAG,EAAE,MAAM,GAAG,EAAE,YAAY,6BAA6B,yBAElC,CAAC,CAAC,KAAK,IAAI;EAAG,YAAY;CAAK;AAC/D;;;;;AAMA,SAAS,cAAc,SAA+C;CAEpE,MAAM,QADe,QAAQ,QAAO,MAAK,EAAE,aAAa,UAC/B,CAAC,CAAC;CAE3B,IAAI,UAAU,GAAG,OAAO;EAAE,OAAO;EAAG,QAAQ;EAAwC,YAAY;CAAK;CAGrG,OAAO;EAAE,OADK,KAAK,IAAI,GAAG,IAAI,QAAQ,WAAW,qBACpC;EAAG,QAAQ,GAAG,MAAM,gBAAgB,QAAQ,IAAI,MAAM;EAAM,YAAY;CAAK;AAC5F;;;;;AAMA,SAAS,gBAAgB,YAA6C,SAAgD;CACpH,IAAI,cAAc;CAClB,IAAI,cAAc;CAElB,KAAK,MAAM,CAAC,KAAK,WAAW,OAAO,QAAQ,UAAU,GAAG;EACtD,IAAI,CAAC,OAAO,YAAY;EACxB,MAAM,SAAS,QAAQ,QAAQ;EAC/B,eAAe;EACf,eAAe,OAAO,QAAQ;CAChC;CAEA,IAAI,gBAAgB,GAAG,OAAO;CAC9B,OAAO,cAAc;AACvB;AAEA,SAAgB,sBAAsB;CACpC,OAAO,aAAa;EAClB,IAAI;EACJ,MAAM;EACN,aACE;EACF,MAAM;EACN,YAAY,UAAU,EAAE,WAAW,CAAC,mBAAmB,MAAM,EAAE,CAAC;CAClE,CAAC,CAAC,CACC,WAAW,OAAO,EAAE,UAAU;EAC7B,MAAM,WAAY,IAAI,UAAU,CAAC;EACjC,MAAM,aAAa,iBAAiB,QAAQ;EAC5C,MAAM,aAAa,WAAW;EAG9B,IAAI,aAAa,WAAW,cAAc;GACxC,MAAM,QAAyB;IAAE,OAAO;IAAG,QAAQ;IAAiB,YAAY;GAAM;GACtF,OAAO;IACL,OAAO;IACP,OAAO;IACP,OAAO;IACP,YAAY;IACZ,OAAO;IACP,YAAY;IACZ,UAAU;IACV;GACF;EACF;EAEA,MAAM,cAAc,WAAW,QAAO,MAAK,EAAE,aAAa,iBAAiB;EAE3E,OAAO;GACL,OAAO;GACP,OAAO,WAAW,WAAW;GAC7B,OAAO,WAAW,WAAW;GAC7B,YAAY,gBAAgB,UAAU;GACtC,OAAO,gBAAgB,UAAU;GACjC,YAAY,gBAAgB,WAAW;GACvC,UAAU,cAAc,UAAU;GAClC;EACF;CACF,CAAC,CAAC,CACD,eAAe,EAAE,cAAc;EAC9B,MAAM,IAAI,QAAQ;EAClB,IAAI,EAAE,OAAO,OAAO;EAEpB,MAAM,QAAQ,gBACZ;GACE,OAAO,EAAE;GACT,OAAO,EAAE;GACT,YAAY,EAAE;GACd,OAAO,EAAE;GACT,YAAY,EAAE;GACd,UAAU,EAAE;EACd,GACA,OACF;EAEA,OAAO,UAAU,OAAO,IAAI,KAAK,MAAM,QAAQ,GAAG,IAAI;CACxD,CAAC,CAAC,CACD,gBAAgB,EAAE,SAAS,YAAY;EACtC,MAAM,IAAI,QAAQ;EAClB,IAAI,EAAE,OACJ,OAAO,aAAa,EAAE,WAAW;EAGnC,MAAM,OAAO,MAAc,IAAI,IAAI,IAAA,CAAK,QAAQ,CAAC,EAAE;EACnD,MAAM,WAAW,MAAc,QAAgB,MAC7C,EAAE,aACE,GAAG,KAAK,IAAI,IAAI,MAAM,EAAE,KAAK,EAAE,OAAO,IAAI,EAAE,MAAM,KAClD,GAAG,KAAK,IAAI,EAAE,OAAO;EAE3B,MAAM,QAAkB,CAAC,UAAU,MAAM,IAAI,EAAE,WAAW,mBAAmB;EAC7E,MAAM,KAAK,QAAQ,SAAS,QAAQ,OAAO,EAAE,KAAK,CAAC;EACnD,MAAM,KAAK,QAAQ,SAAS,QAAQ,OAAO,EAAE,KAAK,CAAC;EACnD,MAAM,KAAK,QAAQ,eAAe,QAAQ,YAAY,EAAE,UAAU,CAAC;EACnE,MAAM,KAAK,QAAQ,SAAS,QAAQ,OAAO,EAAE,KAAK,CAAC;EACnD,MAAM,KAAK,QAAQ,cAAc,QAAQ,YAAY,EAAE,UAAU,CAAC;EAClE,MAAM,KAAK,QAAQ,YAAY,QAAQ,UAAU,EAAE,QAAQ,CAAC;EAE5D,OAAO,MAAM,KAAK,IAAI;CACxB,CAAC;AACL"}

package/dist/experimental-agent.js

@@ -1,6 +1,6 @@
 import { parseExperimentalAgentSetting } from "./onboarding/settings.js";
 import { createDurableAgent, createEventedAgent, isDurableAgent, isEventedAgent } from "@mastra/core/agent/durable";
-//#region src/experimental-agent.ts
+// src/experimental-agent.ts
 const ACCEPTED_VALUES = "\"durable\", \"evented\", or unset";
 function parseExperimentalAgentEnvironment(env = process.env) {
 	const value = env.MASTRACODE_EXPERIMENTAL_AGENT;
@@ -30,7 +30,6 @@ function validateExperimentalAgent(agent, mastra, report = console.info) {
 	if (engineType !== expectedEngine) throw new Error(`Experimental agent "${selection}" resolved workflow engine ${JSON.stringify(engineType)} instead of "${expectedEngine}".`);
 	report(`Experimental agent: ${selection} (workflow engine: ${engineType})`);
 }
-//#endregion
 export { parseExperimentalAgentEnvironment, resolveExperimentalAgent, validateExperimentalAgent, wrapExperimentalAgent };
 
 //# sourceMappingURL=experimental-agent.js.map

package/dist/factory/client.js

@@ -1,6 +1,6 @@
 import { getPlatformSession } from "../platform/index.js";
 import { MastraClient } from "@mastra/client-js";
-//#region src/factory/client.ts
+// src/factory/client.ts
 /** The agent controller every Factory serves its coding sessions under. */
 const FACTORY_CONTROLLER_ID = "code";
 const ORGANIZATION_ID_HEADER = "X-Mastra-Organization-Id";
@@ -92,7 +92,6 @@ function createFactoryClient(address) {
 		fetch: refreshingFetch(address)
 	});
 }
-//#endregion
 export { FACTORY_CONTROLLER_ID, createFactoryClient, createFactoryUserSession, factoryHeaders, listFactoryProjects, refreshingFetch };
 
 //# sourceMappingURL=client.js.map

package/dist/factory/remote-controller.js

@@ -1,5 +1,5 @@
 import { FACTORY_CONTROLLER_ID, createFactoryClient } from "./client.js";
-//#region src/factory/remote-controller.ts
+// src/factory/remote-controller.ts
 /**
 * The Factory-side counterpart to the local agent controller. Modes, models, and
 * provider auth come from the Factory, never from local settings or auth storage.
@@ -37,7 +37,6 @@ var RemoteController = class {
 		} : { hasAuth: true };
 	}
 };
-//#endregion
 export { RemoteController };
 
 //# sourceMappingURL=remote-controller.js.map

package/dist/factory/remote-session.js

@@ -1,7 +1,7 @@
 import { createFactoryClient } from "./client.js";
 import { randomUUID } from "crypto";
 import { defaultDisplayState } from "@mastra/core/agent-controller";
-//#region src/factory/remote-session.ts
+// src/factory/remote-session.ts
 const MAP_FIELDS = [
 	"activeTools",
 	"toolInputBuffers",
@@ -412,7 +412,6 @@ var RemoteSession = class {
 		this.abort();
 	}
 };
-//#endregion
 export { RemoteSession };
 
 //# sourceMappingURL=remote-session.js.map

package/dist/goal-manager.js

@@ -1,6 +1,6 @@
 import { loadSettings } from "./onboarding/settings.js";
 import { getGoalActivityDurationMs } from "@mastra/core/agent";
-//#region src/goal-manager.ts
+// src/goal-manager.ts
 /**
 * GoalManager — persistent cross-turn goals, backed by the Agent's native goal
 * mechanism.
@@ -348,7 +348,6 @@ var GoalManager = class {
 		};
 	}
 };
-//#endregion
 export { DEFAULT_MAX_TURNS, GoalManager };
 
 //# sourceMappingURL=goal-manager.js.map

package/dist/goal-signal.js

@@ -1,4 +1,4 @@
-//#region src/goal-signal.ts
+// src/goal-signal.ts
 function createGoalReminderSignal(goal) {
 	return {
 		type: "system-reminder",
@@ -11,7 +11,6 @@ function createGoalReminderSignal(goal) {
 		}
 	};
 }
-//#endregion
 export { createGoalReminderSignal };
 
 //# sourceMappingURL=goal-signal.js.map

package/dist/headless/cli.js

@@ -11,7 +11,7 @@ import { FLAGS, buildParseArgsOptions, renderFlagUsage } from "./flags.js";
 import { createMastraCode } from "../index.js";
 import { existsSync } from "fs";
 import { parseArgs } from "util";
-//#region src/headless/cli.ts
+// src/headless/cli.ts
 /**
 * CLI adapter for headless MastraCode runs.
 *
@@ -250,7 +250,6 @@ async function runMCCli(predrainedInput, options) {
 	if (!stdoutFlushed && exitCode === 0) exitCode = 1;
 	process.exit(exitCode);
 }
-//#endregion
 export { flushWritable, hasHeadlessFlag, parseHeadlessArgs, printHeadlessUsage, runMCCli };
 
 //# sourceMappingURL=cli.js.map

package/dist/headless/flags.js

@@ -1,5 +1,5 @@
 import { VALID_MODES, VALID_PERMISSION_MODES, VALID_THINKING_LEVELS } from "./types.js";
-//#region src/headless/flags.ts
+// src/headless/flags.ts
 const VALID_OUTPUTS = [
 	"human",
 	"json",
@@ -198,7 +198,6 @@ function renderFlagUsage() {
 		return [`${label.padEnd(width)}${lines[0]}`, ...lines.slice(1).map((line) => `${" ".repeat(width)}${line}`)].join("\n");
 	}).join("\n");
 }
-//#endregion
 export { FLAGS, VALID_OUTPUTS, buildParseArgsOptions, renderFlagUsage };
 
 //# sourceMappingURL=flags.js.map

package/dist/headless/format.js

@@ -1,4 +1,4 @@
-//#region src/headless/format.ts
+// src/headless/format.ts
 /** Truncate a string to `max` characters, appending "..." if truncated. */
 function truncate(s, max) {
 	return s.length > max ? s.slice(0, max) + "..." : s;
@@ -91,7 +91,6 @@ function renderJsonResult(result) {
 		exitCode: result.exitCode
 	}) + "\n";
 }
-//#endregion
 export { createHumanFormatState, formatHuman, formatJsonl, isJsonlEvent, renderJsonResult, renderTextResult, truncate };
 
 //# sourceMappingURL=format.js.map

package/dist/headless/policy.js

@@ -1,4 +1,4 @@
-//#region src/headless/policy.ts
+// src/headless/policy.ts
 /**
 * Default policy — reproduces the historical headless behavior:
 *  - approve every tool approval request,
@@ -36,7 +36,6 @@ function permissionModeToPolicy(mode) {
 		default: return autoApprovePolicy;
 	}
 }
-//#endregion
 export { autoApprovePolicy, denyPolicy, permissionModeToPolicy };
 
 //# sourceMappingURL=policy.js.map

package/dist/headless/run-mc.js

@@ -4,7 +4,7 @@ import { createGoalReminderSignal } from "../goal-signal.js";
 import { MODEL_FALLBACK_STATE_KEY } from "../auth/account-rotation-processor.js";
 import { autoApprovePolicy } from "./policy.js";
 import { isSessionStartupCancelledError } from "@mastra/core/agent-controller";
-//#region src/headless/run-mc.ts
+// src/headless/run-mc.ts
 /**
 * Core programmatic runner for MastraCode headless runs.
 *
@@ -430,7 +430,6 @@ function runMC(options) {
 		}
 	};
 }
-//#endregion
 export { runMC };
 
 //# sourceMappingURL=run-mc.js.map

package/dist/headless/types.js

@@ -1,4 +1,4 @@
-//#region src/headless/types.ts
+// src/headless/types.ts
 const VALID_MODES = [
 	"build",
 	"plan",
@@ -13,7 +13,6 @@ const VALID_THINKING_LEVELS = [
 	"max"
 ];
 const VALID_PERMISSION_MODES = ["auto", "deny"];
-//#endregion
 export { VALID_MODES, VALID_PERMISSION_MODES, VALID_THINKING_LEVELS };
 
 //# sourceMappingURL=types.js.map

package/dist/hooks/config.js

@@ -2,7 +2,7 @@ import { DEFAULT_CONFIG_DIR } from "../constants.js";
 import * as path$1 from "path";
 import * as fs$1 from "fs";
 import * as os$1 from "os";
-//#region src/hooks/config.ts
+// src/hooks/config.ts
 /**
 * Hook configuration loading from filesystem.
 * Loads from global (~/.mastracode/hooks.json) and project (.mastracode/hooks.json).
@@ -74,7 +74,6 @@ function mergeConfigs(global, project) {
 	}
 	return merged;
 }
-//#endregion
 export { VALID_EVENTS, getGlobalHooksPath, getProjectHooksPath, loadHooksConfig };
 
 //# sourceMappingURL=config.js.map

package/dist/hooks/executor.js

@@ -1,6 +1,6 @@
 import { isBlockingEvent } from "./types.js";
 import { spawn } from "child_process";
-//#region src/hooks/executor.ts
+// src/hooks/executor.ts
 /**
 * Hook process execution.
 * Spawns shell commands, handles stdin/stdout/exit-code protocol.
@@ -125,7 +125,6 @@ async function runHooksForEvent(hooks, stdinPayload, matchContext = {}) {
 		warnings
 	};
 }
-//#endregion
 export { executeHook, matchesHook, runHooksForEvent };
 
 //# sourceMappingURL=executor.js.map

package/dist/hooks/manager.js

@@ -1,7 +1,7 @@
 import { DEFAULT_CONFIG_DIR } from "../constants.js";
 import { getGlobalHooksPath, getProjectHooksPath, loadHooksConfig } from "./config.js";
 import { runHooksForEvent } from "./executor.js";
-//#region src/hooks/manager.ts
+// src/hooks/manager.ts
 /**
 * HookManager — high-level orchestration for the hooks system.
 * Created once at startup, provides methods for each lifecycle event.
@@ -249,7 +249,6 @@ var HookManager = class {
 		return runHooksForEvent(hooks, stdin);
 	}
 };
-//#endregion
 export { HookManager };
 
 //# sourceMappingURL=manager.js.map

package/dist/hooks/types.js

@@ -1,8 +1,7 @@
-//#region src/hooks/types.ts
+// src/hooks/types.ts
 function isBlockingEvent(event) {
 	return event === "PreToolUse" || event === "Stop" || event === "UserPromptSubmit";
 }
-//#endregion
 export { isBlockingEvent };
 
 //# sourceMappingURL=types.js.map

package/dist/index.js

@@ -87,7 +87,7 @@ import { DEFAULT_GOAL_JUDGE_PROMPT } from "@mastra/core/tools";
 import { GithubSignals } from "@mastra/github-signals";
 import { LibSQLStore, LibSQLVector } from "@mastra/libsql";
 import { MastraPlatformExporter, MastraStorageExporter, Observability, SensitiveDataFilter } from "@mastra/observability";
-//#region src/index.ts
+// src/index.ts
 const CODE_AGENT_ID = "code-agent";
 const MASTRACODE_TRANSIENT_CONNECTION_MAX_RETRIES = 10;
 const MASTRACODE_MAX_PROCESSOR_RETRIES = 64;
@@ -1365,7 +1365,6 @@ async function prepareAgentControllerMount(config) {
 * case: `bootLocalAgentController` (local) or {@link mountAgentControllerOnMastra} (server).
 */
 const createMastraCode = bootLocalAgentController;
-//#endregion
 export { KnowledgeInspectorError, LOCAL_KNOWLEDGE_ORG_ID, autoApprovePolicy, bootLocalAgentController, createAuthStorage, createKnowledgeInspector, createMastraCode, createMastraCodeAgentController, denyPolicy, formatHuman, formatJsonl, hasHeadlessFlag, mountAgentControllerOnMastra, permissionModeToPolicy, prepareAgentControllerMount, renderJsonResult, renderTextResult, resolveCloudObservabilityConfig, runMC, runMCCli, wireSessionConcerns };
 
 //# sourceMappingURL=index.js.map

package/dist/ipc/ipc-reporter.js

@@ -1,4 +1,4 @@
-//#region src/ipc/ipc-reporter.ts
+// src/ipc/ipc-reporter.ts
 var IPCReporter = class {
 	enabled;
 	constructor() {
@@ -15,7 +15,6 @@ var IPCReporter = class {
 	}
 };
 const ipcReporter = new IPCReporter();
-//#endregion
 export { ipcReporter };
 
 //# sourceMappingURL=ipc-reporter.js.map

package/dist/knowledge-inspector.js

@@ -1,6 +1,6 @@
 import { resolveKnowledgeScopeIdentity } from "./knowledge-scope.js";
 import { createKnowledgeNodeCursor, isKnowledgeScopeVisible, parseKnowledgeWikilinks } from "@mastra/core/storage";
-//#region src/knowledge-inspector.ts
+// src/knowledge-inspector.ts
 var KnowledgeInspectorError = class extends Error {
 	code;
 	constructor(code, message) {
@@ -510,7 +510,6 @@ async function createKnowledgeInspector(input) {
 		session: input.session
 	}) : void 0;
 }
-//#endregion
 export { KnowledgeInspectorError, createKnowledgeInspector };
 
 //# sourceMappingURL=knowledge-inspector.js.map

package/dist/knowledge-scope.js

@@ -1,4 +1,4 @@
-//#region src/knowledge-scope.ts
+// src/knowledge-scope.ts
 /**
 * The organization rung local (TUI/studio) knowledge is curated under. A fixed
 * literal on purpose: deriving it from a hostname, machine id, or path would
@@ -41,7 +41,6 @@ function resolveKnowledgeScopeIdentity(state) {
 		organizationId: LOCAL_KNOWLEDGE_ORG_ID
 	};
 }
-//#endregion
 export { LOCAL_KNOWLEDGE_ORG_ID, resolveKnowledgeScopeIdentity };
 
 //# sourceMappingURL=knowledge-scope.js.map

package/dist/lsp/client.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"client.d.ts","sourceRoot":"","sources":["../../src/lsp/client.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,gCAAgC,CAAC;AACjE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAEjD;;GAEG;AACH,qBAAa,SAAS;IACpB,OAAO,CAAC,UAAU,CAAkC;IACpD,OAAO,CAAC,OAAO,CAA6B;IAC5C,OAAO,CAAC,UAAU,CAAgB;IAClC,OAAO,CAAC,aAAa,CAAS;IAC9B,OAAO,CAAC,WAAW,CAAwC;IAC3D,OAAO,CAAC,qBAAqB,CAAa;gBAE9B,UAAU,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM;IAK5D;;OAEG;IACG,UAAU,IAAI,OAAO,CAAC,IAAI,CAAC;IAkLjC;;OAEG;IACH,UAAU,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,GAAG,IAAI;IAkBvE;;OAEG;IACH,YAAY,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,IAAI;IAYtE;;;OAGG;IACG,kBAAkB,CACtB,QAAQ,EAAE,MAAM,EAChB,SAAS,GAAE,MAAa,EACxB,aAAa,GAAE,OAAe,GAC7B,OAAO,CAAC,UAAU,EAAE,CAAC;IA+BxB;;OAEG;IACG,QAAQ,CAAC,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC;IAa/E;;OAEG;IACH,cAAc,CAAC,QAAQ,EAAE,MAAM,GAAG,UAAU,EAAE;IAK9C;;OAEG;IACH,iBAAiB,IAAI,UAAU,EAAE;IAQjC;;OAEG;IACH,WAAW,CAAC,QAAQ,EAAE,MAAM,GAAG,IAAI;IAcnC;;OAEG;IACG,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC;IAO9B;;OAEG;IACG,QAAQ,IAAI,OAAO,CAAC,IAAI,CAAC;CAoChC"}
+{"version":3,"file":"client.d.ts","sourceRoot":"","sources":["../../src/lsp/client.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,gCAAgC,CAAC;AACjE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAEjD;;GAEG;AACH,qBAAa,SAAS;IACpB,OAAO,CAAC,UAAU,CAAkC;IACpD,OAAO,CAAC,OAAO,CAA6B;IAC5C,OAAO,CAAC,UAAU,CAAgB;IAClC,OAAO,CAAC,aAAa,CAAS;IAC9B,OAAO,CAAC,WAAW,CAAwC;IAC3D,OAAO,CAAC,qBAAqB,CAAa;gBAE9B,UAAU,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM;IAK5D;;OAEG;IACG,UAAU,IAAI,OAAO,CAAC,IAAI,CAAC;IAkLjC;;OAEG;IACH,UAAU,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,GAAG,IAAI;IAkBvE;;OAEG;IACH,YAAY,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,IAAI;IAYtE;;;OAGG;IACG,kBAAkB,CACtB,QAAQ,EAAE,MAAM,EAChB,SAAS,GAAE,MAAa,EACxB,aAAa,GAAE,OAAe,GAC7B,OAAO,CAAC,UAAU,EAAE,CAAC;IA8BxB;;OAEG;IACG,QAAQ,CAAC,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC;IAa/E;;OAEG;IACH,cAAc,CAAC,QAAQ,EAAE,MAAM,GAAG,UAAU,EAAE;IAK9C;;OAEG;IACH,iBAAiB,IAAI,UAAU,EAAE;IAQjC;;OAEG;IACH,WAAW,CAAC,QAAQ,EAAE,MAAM,GAAG,IAAI;IAcnC;;OAEG;IACG,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC;IAO9B;;OAEG;IACG,QAAQ,IAAI,OAAO,CAAC,IAAI,CAAC;CAoChC"}

package/dist/lsp/client.js

@@ -1,6 +1,6 @@
 import { StreamMessageReader, StreamMessageWriter, createMessageConnection } from "vscode-jsonrpc/node.js";
 import { Position, TextDocumentIdentifier } from "vscode-languageserver-protocol";
-//#region src/lsp/client.ts
+// src/lsp/client.ts
 /**
 * LSP Client wrapper for JSON-RPC communication
 */
@@ -245,7 +245,6 @@ var LSPClient = class {
 		this.diagnostics = /* @__PURE__ */ new Map();
 	}
 };
-//#endregion
 export { LSPClient };
 
 //# sourceMappingURL=client.js.map