@codeyam/codeyam-cli 0.1.0-staging.7c30edc → 0.1.0-staging.7ec08ee

package/analyzer-template/packages/ai/src/lib/dataStructure/helpers/cleanKnownObjectFunctions.ts

@@ -6,7 +6,7 @@ import { ScopeNode, ScopeDataStructure } from '../ScopeDataStructure';
 
 type PrimitiveKind = 'array' | 'string' | 'number' | 'boolean' | 'date';
 
-const IGNORE_CLASSES_AND_OBJECTS = [
+const IGNORE_CLASSES_AND_OBJECTS_SET = new Set([
   'Math',
   'JSON',
   'Reflect',
@@ -22,7 +22,7 @@ const IGNORE_CLASSES_AND_OBJECTS = [
   'useContext',
   'useReducer',
   'useLayoutEffect',
-];
+]);
 
 /** cache the tokenised representation of every path we touch */
 const partsCache = new Map<string, string[]>();
@@ -147,16 +147,26 @@ function scrub(
   rootTable: Map<string, PrimitiveKind>,
   onRemove: (key: string) => void,
 ): void {
-  const sortedKeys = Object.keys(mapping).sort((a, b) => a.localeCompare(b));
+  // Early exit: if no primitive types exist in the schema, there are no
+  // built-in methods to clean. Saves sorting + walking 10K+ keys.
+  if (rootTable.size === 0) return;
+
+  // Use direct comparison instead of localeCompare (3-5x faster for large arrays)
+  const sortedKeys = Object.keys(mapping).sort((a, b) =>
+    a < b ? -1 : a > b ? 1 : 0,
+  );
 
   for (const keyPath of sortedKeys) {
     const keyParts = getParts(keyPath);
     const prefixParts = buildPrefixParts(keyParts);
 
+    const firstPart = keyParts[0];
+    const firstPartBase = firstPart.includes('(')
+      ? firstPart.split('(')[0]
+      : firstPart;
     if (
-      IGNORE_CLASSES_AND_OBJECTS.find(
-        (m) => keyParts[0] === m || keyParts[0].startsWith(`${m}(`),
-      )
+      IGNORE_CLASSES_AND_OBJECTS_SET.has(firstPart) ||
+      IGNORE_CLASSES_AND_OBJECTS_SET.has(firstPartBase)
     ) {
       delete mapping[keyPath];
       continue;

package/analyzer-template/packages/ai/src/lib/dataStructure/helpers/coercePrimitivesToArraysBySchema.ts

@@ -0,0 +1,62 @@
+/**
+ * Coerces primitive values to empty arrays when the schema expects an array.
+ *
+ * The LLM sometimes generates a primitive (boolean, number, string) where the
+ * schema expects an array. For example:
+ *   Schema: { webapps: [{ name: "string" }] }
+ *   LLM output: { webapps: true }
+ *
+ * This causes runtime errors like `TypeError: config.webapps?.forEach is not a function`
+ * because the code expects an array but gets a boolean.
+ *
+ * Coercion rule:
+ * - primitive → array: replace with []
+ *
+ * Null and undefined values are preserved (they may be intentionally nullable).
+ *
+ * @param data The mock data object to coerce (mutated in place)
+ * @param schema The schema describing expected types
+ */
+export default function coercePrimitivesToArraysBySchema<T>(
+  data: T,
+  schema: unknown,
+): T {
+  if (data === null || data === undefined) {
+    return data;
+  }
+
+  if (typeof data !== 'object') {
+    return data;
+  }
+
+  if (Array.isArray(data)) {
+    const itemSchema = Array.isArray(schema) ? schema[0] : undefined;
+    for (let i = 0; i < data.length; i++) {
+      data[i] = coercePrimitivesToArraysBySchema(data[i], itemSchema);
+    }
+    return data;
+  }
+
+  const dataObj = data as Record<string, unknown>;
+  const schemaObj = schema as Record<string, unknown> | undefined;
+
+  for (const key of Object.keys(dataObj)) {
+    const value = dataObj[key];
+    const fieldSchema = schemaObj?.[key];
+
+    if (value === null || value === undefined) {
+      continue;
+    }
+
+    if (typeof value !== 'object' && Array.isArray(fieldSchema)) {
+      // Schema expects an array but the value is a primitive.
+      // Replace with an empty array.
+      dataObj[key] = [];
+    } else if (typeof value === 'object' && value !== null) {
+      // Both value and schema are objects — recurse
+      coercePrimitivesToArraysBySchema(value, fieldSchema);
+    }
+  }
+
+  return data;
+}

package/analyzer-template/packages/ai/src/lib/dataStructure/helpers/convertDotNotation.ts

@@ -14,35 +14,46 @@ const isStandaloneIndex = (s?: string) => !!s && STANDALONE_INDEX_RE.test(s);
 // The regex matches any path ending with .length that has [] somewhere before it
 const DYNAMIC_LENGTH_RE = /\[\].*\.length$/;
 
-// Treat these as structural placeholders (don't commit them as concrete leaves)
-function isSkippableLeafType(t: string) {
-  // 'unknown' by itself is a placeholder (but 'boolean | unknown' is not)
-  // Also handles optional variants like 'object | undefined', 'object | null',
-  // 'array | undefined', etc. — these are still structural placeholders that
-  // should not overwrite already-populated structures.
+// Cache for type string analysis to avoid repeated split/filter operations.
+// These functions are called multiple times per path segment across thousands of paths.
+const typeAnalysisCache = new Map<
+  string,
+  { isSkippable: boolean; baseType: string; isNullable: boolean }
+>();
+
+function getTypeAnalysis(t: string) {
+  const cached = typeAnalysisCache.get(t);
+  if (cached) return cached;
   const parts = t.split('|').map((s) => s.trim());
   const base = parts.filter((s) => s !== 'undefined' && s !== 'null');
-  return (
-    base.length === 1 &&
-    (base[0] === 'object' ||
-      base[0] === 'array' ||
-      base[0] === 'function' ||
-      base[0] === 'unknown')
-  );
+  const result = {
+    isSkippable:
+      base.length === 1 &&
+      (base[0] === 'object' ||
+        base[0] === 'array' ||
+        base[0] === 'function' ||
+        base[0] === 'unknown'),
+    baseType: base[0],
+    isNullable: parts.includes('undefined') || parts.includes('null'),
+  };
+  typeAnalysisCache.set(t, result);
+  return result;
+}
+
+// Treat these as structural placeholders (don't commit them as concrete leaves)
+function isSkippableLeafType(t: string) {
+  return getTypeAnalysis(t).isSkippable;
 }
 
 // Extract the base structural type from a potentially nullable type string.
 // e.g., 'object | undefined' → 'object', 'array | null' → 'array'
 function getBaseSkippableType(t: string): string {
-  const parts = t.split('|').map((s) => s.trim());
-  const base = parts.filter((s) => s !== 'undefined' && s !== 'null');
-  return base[0];
+  return getTypeAnalysis(t).baseType;
 }
 
 // Check if a type string has nullable annotations (| undefined or | null)
 function isNullableType(t: string): boolean {
-  const parts = t.split('|').map((s) => s.trim());
-  return parts.includes('undefined') || parts.includes('null');
+  return getTypeAnalysis(t).isNullable;
 }
 
 // Matches paths containing [][] — e.g., "items[][]" or "items[][].text"

package/analyzer-template/packages/ai/src/lib/dataStructure/helpers/stripNullableMarkers.ts

@@ -0,0 +1,35 @@
+/**
+ * Recursively strips `_nullable` keys from LLM-generated mock data.
+ *
+ * The `_nullable` marker is an internal CodeYam concept used in type definitions
+ * to indicate that a field can be null/undefined. The LLM sometimes includes
+ * these markers in its generated scenario data, which causes runtime errors
+ * when code iterates over object keys (e.g., `Object.keys(importedBy)` picks
+ * up `_nullable` as a key alongside real data).
+ *
+ * This function mutates the input object in-place, consistent with other
+ * post-processing helpers like `convertNullToUndefinedBySchema`.
+ */
+export default function stripNullableMarkers(
+  data: Record<string, unknown>,
+): void {
+  if (data == null || typeof data !== 'object') return;
+
+  // Delete _nullable from this level
+  if ('_nullable' in data) {
+    delete data._nullable;
+  }
+
+  // Recurse into nested objects and arrays
+  for (const value of Object.values(data)) {
+    if (Array.isArray(value)) {
+      for (const item of value) {
+        if (item !== null && typeof item === 'object' && !Array.isArray(item)) {
+          stripNullableMarkers(item as Record<string, unknown>);
+        }
+      }
+    } else if (value !== null && typeof value === 'object') {
+      stripNullableMarkers(value as Record<string, unknown>);
+    }
+  }
+}

package/analyzer-template/packages/ai/src/lib/dataStructureChunking.ts

@@ -4,10 +4,10 @@ import type { ExecutionFlow, ScenariosDataStructure } from '~codeyam/types';
 type RequiredValue = NonNullable<ExecutionFlow['requiredValues']>[number];
 
 const DEFAULT_MAX_CHUNK_SIZE = 10_000; // ~10K chars per chunk
+const DEFAULT_MAX_KEYS_PER_CHUNK = 20;
 
 export interface ChunkOptions {
   maxChunkSize?: number;
-  /** @deprecated Use maxChunkSize instead. Kept for backward compatibility in tests. */
   maxKeysPerChunk?: number;
 }
 
@@ -33,16 +33,19 @@ export function chunkDataStructure(
     return [dataForMocks];
   }
 
-  const { maxChunkSize = DEFAULT_MAX_CHUNK_SIZE } = options;
+  const {
+    maxChunkSize = DEFAULT_MAX_CHUNK_SIZE,
+    maxKeysPerChunk = DEFAULT_MAX_KEYS_PER_CHUNK,
+  } = options;
   const keys = Object.keys(dataForMocks);
 
   // Calculate total size to see if chunking is needed
   const totalSize = JSON.stringify(dataForMocks).length;
-  if (totalSize <= maxChunkSize) {
+  if (totalSize <= maxChunkSize && keys.length <= maxKeysPerChunk) {
     return [dataForMocks];
   }
 
-  // Greedily pack keys into chunks by cumulative size
+  // Greedily pack keys into chunks by cumulative size and key count
   const chunks: Array<ScenariosDataStructure['dataForMocks']> = [];
   let currentChunk: Record<string, unknown> = {};
   let currentSize = 0;
@@ -52,8 +55,14 @@ export function chunkDataStructure(
       (dataForMocks as Record<string, unknown>)[key],
     ).length;
 
-    // If adding this key would exceed the limit AND chunk isn't empty, start a new chunk
-    if (currentSize > 0 && currentSize + keySize > maxChunkSize) {
+    const currentKeyCount = Object.keys(currentChunk).length;
+
+    // If adding this key would exceed size or key count limit AND chunk isn't empty, start a new chunk
+    if (
+      currentSize > 0 &&
+      (currentSize + keySize > maxChunkSize ||
+        currentKeyCount >= maxKeysPerChunk)
+    ) {
       chunks.push(currentChunk as ScenariosDataStructure['dataForMocks']);
       currentChunk = {};
       currentSize = 0;

package/analyzer-template/packages/ai/src/lib/generateEntityScenarioData.ts

@@ -19,6 +19,8 @@ import convertNullToUndefinedBySchema from './dataStructure/helpers/convertNullT
 import convertTypeAnnotationsToValues from './dataStructure/helpers/convertTypeAnnotationsToValues';
 import fixNullIdsBySchema from './dataStructure/helpers/fixNullIdsBySchema';
 import coerceObjectsToPrimitivesBySchema from './dataStructure/helpers/coerceObjectsToPrimitivesBySchema';
+import coercePrimitivesToArraysBySchema from './dataStructure/helpers/coercePrimitivesToArraysBySchema';
+import stripNullableMarkers from './dataStructure/helpers/stripNullableMarkers';
 import { JsonTypeDefinition } from '~codeyam/types';
 import { deepMerge } from '~codeyam/generate';
 import {
@@ -172,6 +174,46 @@ function findKeyPath(
   return null;
 }
 
+/**
+ * Strip primitive values from `source` when the `destination` already has an
+ * array or object at the same key. Returns a cleaned copy of source (or null
+ * if nothing remains after stripping).
+ *
+ * This prevents misplaced LLM data (e.g., `{ webapps: true }`) from
+ * overwriting correct complex values (e.g., `{ webapps: [...] }`) during
+ * deep merge in relocateMisplacedNestedKeys.
+ */
+function stripPrimitivesOverwritingComplexValues(
+  destination: Record<string, unknown>,
+  source: unknown,
+): Record<string, unknown> | null {
+  if (typeof source !== 'object' || source === null || Array.isArray(source)) {
+    return null;
+  }
+
+  const result: Record<string, unknown> = {};
+  let hasKeys = false;
+
+  for (const key of Object.keys(source as Record<string, unknown>)) {
+    const srcVal = (source as Record<string, unknown>)[key];
+    const dstVal = destination[key];
+
+    // If destination has an array or object but source has a primitive, skip it
+    if (
+      typeof dstVal === 'object' &&
+      dstVal !== null &&
+      typeof srcVal !== 'object'
+    ) {
+      continue;
+    }
+
+    result[key] = srcVal;
+    hasKeys = true;
+  }
+
+  return hasKeys ? result : null;
+}
+
 /**
  * Relocate misplaced nested keys in mockData to their correct position
  * based on the dataForMocks structure.
@@ -233,12 +275,22 @@ function relocateMisplacedNestedKeys(
     }
 
     // Deep merge the value into the correct location
-    // Use deep merge to preserve existing data at that location
+    // Use deep merge to preserve existing data at that location.
+    // Before merging, strip primitives from the misplaced value that would
+    // overwrite arrays/objects at the destination. The misplaced data is lower
+    // quality — the LLM put it in the wrong place — so primitives like `true`
+    // should not overwrite correct complex values like arrays.
     if (current[key] !== undefined && typeof current[key] === 'object') {
-      current[key] = deepMerge(
+      const safeValue = stripPrimitivesOverwritingComplexValues(
         current[key] as Record<string, unknown>,
         value as Record<string, unknown>,
       );
+      if (safeValue !== null) {
+        current[key] = deepMerge(
+          current[key] as Record<string, unknown>,
+          safeValue,
+        );
+      }
     } else {
       current[key] = value;
     }
@@ -1063,6 +1115,18 @@ export async function generateDataForScenario({
     );
   }
 
+  // Strip _nullable markers from LLM-generated mock data.
+  // The _nullable marker is an internal CodeYam concept used in type definitions
+  // to indicate that a field can be null/undefined. The LLM sometimes includes
+  // these markers in its generated data, which causes runtime errors when code
+  // iterates over object keys (e.g., Object.keys(importedBy) picks up "_nullable",
+  // then Object.keys(importedBy["_nullable"]) calls Object.keys(null) and throws).
+  if (fullScenarioData.data.mockData) {
+    stripNullableMarkers(
+      fullScenarioData.data.mockData as Record<string, unknown>,
+    );
+  }
+
   // Relocate misplaced nested keys to their correct position.
   // The LLM sometimes places nested keys at root level instead of inside their
   // parent object (e.g., 'fastener' at root instead of inside 'trpc').
@@ -1098,6 +1162,18 @@ export async function generateDataForScenario({
     );
   }
 
+  // Coerce primitives to empty arrays when the schema expects an array.
+  // The LLM sometimes generates a primitive (e.g., `webapps: true`) where the
+  // schema expects an array (e.g., `webapps: [{ name: "string" }]`).
+  // This causes runtime errors like "TypeError: config.webapps?.forEach is not a function".
+  // Must run AFTER coerceObjectsToPrimitivesBySchema and BEFORE convertCommaSeparatedStringsToArrays.
+  if (structure.dataForMocks && fullScenarioData.data.mockData) {
+    coercePrimitivesToArraysBySchema(
+      fullScenarioData.data.mockData,
+      structure.dataForMocks,
+    );
+  }
+
   // Convert comma-separated strings to arrays when appropriate.
   // The LLM sometimes generates strings like "color,size" instead of arrays
   // like ["color", "size"] when the schema type is incorrectly inferred as

package/analyzer-template/packages/ai/src/lib/generateExecutionFlows.ts

@@ -32,6 +32,30 @@ import resolvePathToControllable from './resolvePathToControllable';
 import { AI } from '~codeyam/ai';
 import type { JsxRenderingUsage, ConditionalUsage } from './astScopes/types';
 
+/**
+ * Patterns that indicate a child component is a modal/overlay/blocking element.
+ * Same patterns used in generateExecutionFlowsFromConditionalEffects.ts for
+ * state variable names — here applied to child component names.
+ */
+const BLOCKING_CHILD_PATTERNS = [
+  /modal/i,
+  /overlay/i,
+  /dialog/i,
+  /popup/i,
+  /drawer/i,
+  /sheet/i,
+  /lightbox/i,
+  /backdrop/i,
+];
+
+/**
+ * Check if a child component name suggests it's a blocking/modal element.
+ * Exported for testing.
+ */
+export function isBlockingChildComponent(componentName: string): boolean {
+  return BLOCKING_CHILD_PATTERNS.some((pattern) => pattern.test(componentName));
+}
+
 interface GenerateExecutionFlowsArgs {
   entity: Pick<Entity, 'sha' | 'name' | 'filePath' | 'code' | 'metadata'>;
   mergedDataStructure: Omit<DataStructure, 'equivalentSignatureVariables'>;
@@ -105,19 +129,6 @@ export default function generateExecutionFlows({
     fullToShortPathMap[fullPath] = shortPath;
   }
 
-  console.log(
-    `[generateExecutionFlows] ${entity.name}: attributesMap has ${Object.keys(resolvedAttributesMap).length} keys, fullToShortPathMap has ${Object.keys(fullToShortPathMap).length} entries`,
-  );
-  console.log(
-    `[generateExecutionFlows] ${entity.name}: attributesMap keys: [${Object.keys(resolvedAttributesMap).join(', ')}]`,
-  );
-  console.log(
-    `[generateExecutionFlows] ${entity.name}: fullToShortPathMap: ${JSON.stringify(fullToShortPathMap)}`,
-  );
-  console.log(
-    `[generateExecutionFlows] ${entity.name}: equivalentSignatureVariables: ${JSON.stringify(equivalentSignatureVariables)}`,
-  );
-
   // Get pre-computed conditional data from metadata
   const conditionalUsages: Record<string, ConditionalUsage[]> =
     entity.metadata?.isolatedDataStructure?.conditionalUsages ?? {};
@@ -162,26 +173,6 @@ export default function generateExecutionFlows({
     }
   }
 
-  console.log(
-    `[generateExecutionFlows] ${entity.name}: conditionalUsages keys: [${Object.keys(conditionalUsages).join(', ')}]`,
-  );
-  console.log(
-    `[generateExecutionFlows] ${entity.name}: conditionalUsages sourceDataPaths: ${JSON.stringify(
-      Object.fromEntries(
-        Object.entries(conditionalUsages).map(([k, v]) => [
-          k,
-          (v as any[]).map((u: any) => u.sourceDataPath ?? '(none)'),
-        ]),
-      ),
-    )}`,
-  );
-  console.log(
-    `[generateExecutionFlows] ${entity.name}: compoundConditionals count: ${compoundConditionals.length}`,
-  );
-  console.log(
-    `[generateExecutionFlows] ${entity.name}: sourceEquivalencies keys count: ${Object.keys(mergedDataStructure.sourceEquivalencies ?? {}).length}`,
-  );
-
   const conditionalsBasedFlows = generateExecutionFlowsFromConditionals({
     conditionalUsages,
     compoundConditionals,
@@ -273,6 +264,11 @@ export default function generateExecutionFlows({
     calculateFlowPriority(flow, gatingPathToChildren);
   }
 
+  // Assign exclusiveGroup for truthy/falsy flow pairs that control child components.
+  // When the same gating path produces both a truthy flow (show component) and a
+  // falsy flow (hide component), they are mutually exclusive and should be grouped.
+  assignExclusiveGroupsForGatingPairs(allFlows);
+
   // Sort flows by priority (highest first)
   allFlows.sort((a, b) => (b.priority ?? 0) - (a.priority ?? 0));
 
@@ -361,6 +357,60 @@ function buildGatingPathToChildrenMap(
   return gatingPathToChildren;
 }
 
+/**
+ * Assign exclusiveGroup for truthy/falsy flow pairs that control child components.
+ *
+ * When the same attribute path produces both a truthy and falsy flow (e.g.,
+ * {diffView && <DiffModal/>}), they are mutually exclusive — only one can be
+ * active at a time. Grouping them ensures the scenario generator treats them
+ * as alternative states rather than combining them.
+ */
+function assignExclusiveGroupsForGatingPairs(flows: ExecutionFlow[]): void {
+  // Build a map: attributePath -> flows with truthy or falsy comparison for that path
+  const pathToFlows = new Map<
+    string,
+    { truthy: ExecutionFlow[]; falsy: ExecutionFlow[] }
+  >();
+
+  for (const flow of flows) {
+    // Only consider flows that control child components
+    if (!flow.childComponentsControlled?.length) continue;
+
+    for (const rv of flow.requiredValues) {
+      if (rv.comparison === 'truthy' || rv.comparison === 'falsy') {
+        if (!pathToFlows.has(rv.attributePath)) {
+          pathToFlows.set(rv.attributePath, { truthy: [], falsy: [] });
+        }
+        pathToFlows.get(rv.attributePath)![rv.comparison].push(flow);
+      }
+    }
+  }
+
+  // Also check falsy flows that DON'T have childComponentsControlled —
+  // they still pair with truthy flows that do
+  for (const flow of flows) {
+    if (flow.childComponentsControlled?.length) continue; // Already processed
+
+    for (const rv of flow.requiredValues) {
+      if (rv.comparison === 'falsy' && pathToFlows.has(rv.attributePath)) {
+        pathToFlows.get(rv.attributePath)!.falsy.push(flow);
+      }
+    }
+  }
+
+  // Assign exclusiveGroup where both truthy and falsy flows exist
+  for (const [attributePath, { truthy, falsy }] of pathToFlows) {
+    if (truthy.length > 0 && falsy.length > 0) {
+      const groupName = `gating:${attributePath}`;
+      for (const flow of [...truthy, ...falsy]) {
+        if (!flow.exclusiveGroup) {
+          flow.exclusiveGroup = groupName;
+        }
+      }
+    }
+  }
+}
+
 /**
  * Calculate priority and childComponentsControlled for a flow.
  *
@@ -415,6 +465,19 @@ function calculateFlowPriority(
     if (flow.impact !== 'high') {
       flow.impact = 'high';
     }
+
+    // Check if any controlled child component is a modal/dialog/overlay.
+    // Only truthy flows (which SHOW the component) should be marked as blocking.
+    // Falsy flows (which HIDE the component) should not be blocking.
+    const isTruthyFlow = flow.requiredValues.some(
+      (rv) => rv.comparison === 'truthy' || rv.comparison === 'equals',
+    );
+    if (
+      isTruthyFlow &&
+      Array.from(childComponentsControlled).some(isBlockingChildComponent)
+    ) {
+      flow.blocksOtherFlows = true;
+    }
   }
 
   // Calculate priority score

package/analyzer-template/packages/analyze/index.ts

@@ -36,7 +36,10 @@ export { getAllEntityNodes } from './src/lib/asts/sourceFiles/getAllEntityNodes'
 export { default as fileAnalyzerFromCode } from './src/lib/files/fileAnalyzerFromCode';
 
 export { default as isolateDataStructure } from './src/lib/files/scenarios/isolateDataStructure';
-export { default as mergeInDependentDataStructure } from './src/lib/files/scenarios/mergeInDependentDataStructure';
+export {
+  default as mergeInDependentDataStructure,
+  DataStructureTimeoutError,
+} from './src/lib/files/scenarios/mergeInDependentDataStructure';
 export { default as mergeValidatedDataStructures } from './src/lib/files/scenarios/mergeValidatedDataStructures';
 
 export { discoverDirectDependencies } from './src/lib/files/analyze/dependencyResolver';

package/analyzer-template/packages/analyze/src/lib/ProjectAnalyzer.ts

@@ -46,6 +46,12 @@ export class ProjectAnalyzer {
       lib.asts.createProgramFromTsConfigPath(configPath, excludePatterns),
     );
 
+    if (this.programs.length === 0) {
+      throw new Error(
+        'No TypeScript programs could be created — no tsconfig.json found',
+      );
+    }
+
     // Primary program is the first one (usually root) for backward compatibility
     this.program = this.programs[0];
     this.typeChecker = this.program.getTypeChecker();
@@ -334,17 +340,23 @@ export class ProjectAnalyzer {
    * Refresh the TypeScript program to pick up file changes
    */
   refreshProgram(): void {
-    // Memory logging: Capture state before refresh
-    const memBefore = process.memoryUsage();
-
     // Apply the same unapprovedPaths filtering when refreshing
     const excludePatterns =
       this.project.metadata?.unapprovedPaths?.filter(Boolean) || [];
 
-    // Refresh all programs
-    this.programs = this.tsConfigPaths.map((configPath) =>
-      lib.asts.createProgramFromTsConfigPath(configPath, excludePatterns),
-    );
+    // Release old programs and caches BEFORE creating new ones to reduce peak memory.
+    // Without this, both old (~1.7GB) and new (~1.7GB) programs coexist during creation,
+    // causing OOM on monorepos with multiple tsconfigs.
+    this.programs = [];
+    this.sourceFileByPath.clear();
+    this.typeCheckerByPath.clear();
+
+    // Create programs one at a time so each old program can be GC'd before the next
+    for (const configPath of this.tsConfigPaths) {
+      this.programs.push(
+        lib.asts.createProgramFromTsConfigPath(configPath, excludePatterns),
+      );
+    }
 
     // Update primary program for backward compatibility
     this.program = this.programs[0];

package/analyzer-template/packages/analyze/src/lib/asts/index.ts

@@ -95,8 +95,13 @@ export function createProgramFromTsConfigPath(
     throw new Error('Could not parse tsconfig.json');
   }
 
-  // Filter file names based on exclude patterns to reduce memory usage
-  let fileNames = parsedCommandLine.fileNames;
+  // Always filter out node_modules from root files to prevent OOM on large projects.
+  // This only removes them as root compilation targets — TypeScript's module resolution
+  // still follows imports into node_modules for type info when processing source files.
+  let fileNames = parsedCommandLine.fileNames.filter(
+    (f) => !f.includes('/node_modules/'),
+  );
+
   if (excludePatterns && excludePatterns.length > 0) {
     const regexPatterns = excludePatterns
       .map((pattern) => {

package/analyzer-template/packages/analyze/src/lib/asts/nodes/getNodeType.ts

@@ -39,6 +39,7 @@ export function getNodeType(node: ts.Node, sourceFile: ts.SourceFile) {
     case 'StringLiteral':
     case 'NumericLiteral':
     case 'ObjectLiteralExpression':
+    case 'ArrayLiteralExpression':
       nodeType = 'data';
       break;
     case 'TypeAliasDeclaration':