@hiveai/core 0.11.0 → 0.12.1

package/dist/index.d.ts

@@ -718,6 +718,55 @@ declare function isGlobPath(p: string): boolean;
 declare function globToRegExp(pattern: string): RegExp;
 declare function relPathFrom(root: string, abs: string): string;
 
+/**
+ * Distinctive-token corroboration for the anti-pattern gate.
+ *
+ * The pre-commit gate used to hard-block whenever a diff shared ANY ≥4-char token
+ * with an anchored gotcha's body. That fires on ubiquitous domain words ("memory",
+ * "sensor", "scope", "input", "version") and on version-bump diffs — blocking agents
+ * for nothing. The fix: a `literal` overlap only corroborates a BLOCK when at least
+ * one shared token is *distinctive* to that gotcha — i.e. rare across the gotcha
+ * corpus (low document frequency), like `BigInt`, `open-in-view`, `rec_7`. Common
+ * words can still surface the warning for review; they just can't hard-block.
+ *
+ * Pure module (no I/O), TF-IDF-style. Unit-tested in `test/distinctive.test.ts`.
+ */
+/**
+ * Language keywords + ubiquitous code words that would match almost any memory body
+ * and so carry no distinguishing signal. Shared by the diff tokenizer and the
+ * distinctiveness check so "literal" stays meaningful.
+ */
+declare const CODE_STOPWORDS: Set<string>;
+/** Minimum token length kept for word-level matching (shorter tokens are too noisy). */
+declare const MIN_WORD_LEN = 4;
+/** Split text into lowercase word tokens (>= MIN_WORD_LEN, excluding code stopwords). */
+declare function tokenizeWords(text: string): string[];
+interface DocFrequency {
+    /** token -> number of documents (memory bodies) it appears in */
+    df: Map<string, number>;
+    /** total number of documents */
+    total: number;
+}
+/** Build per-token document frequency across a corpus of memory bodies. */
+declare function buildDocFrequency(bodies: string[]): DocFrequency;
+/**
+ * Document-frequency cap at/below which a token counts as distinctive. Deliberately
+ * strict — "distinctive" means *rare* (≈ the bottom 10% of the corpus), with a floor
+ * of 1 so a token appearing in a single memory is always distinctive. Strictness is
+ * intentional: blocking is the aggressive action, so we under-block rather than fire
+ * on a word that several gotchas happen to share.
+ */
+declare function distinctiveCap(total: number): number;
+/** True when `token` is distinctive (rare) within the corpus. */
+declare function isDistinctiveToken(token: string, freq: DocFrequency): boolean;
+/**
+ * True when the added diff text shares at least one *distinctive* word token with the
+ * memory body. This is the precise corroboration the block decision should require:
+ * "the change actually contains the specific thing this gotcha warns about", not
+ * "the change happens to mention a common domain word".
+ */
+declare function diffHasDistinctiveOverlap(addedDiffText: string, memoryBody: string, freq: DocFrequency): boolean;
+
 /**
  * Progressive disclosure for `skill` memories.
  *
@@ -1417,4 +1466,4 @@ interface SensorSuggestionOptions {
  */
 declare function suggestSensorFromMemory(body: string, anchorPaths: string[], options?: SensorSuggestionOptions): Sensor | null;
 
-export { AUTOPILOT_DEFAULTS, type Activation, type ActivationContext, ActivationSchema, type Anchor, AnchorSchema, type AntiPatternGate, type AutoPromoteRule, BRIEFING_MARKER_TTL_MS, BRIEFING_PRESET_DEFAULTS, type BreakingChange, type BriefingBudgetNumbers, type BriefingBudgetPreset, type BriefingMarker, type BudgetPart, type BudgetSlice, type BuildCodeMapOptions, CHARS_PER_TOKEN, CODE_MAP_FILE, CONFIG_FILE, type CodeExport, type CodeExportKind, type CodeFileEntry, type CodeMap, type CodeMapQueryOptions, type CollectTimelineOpts, type ConfidenceLevel, type ConfidenceThresholds, type ConflictCandidatePair, type ConflictCandidatesOpts, type ContractDiffResult, type ContractFile, type ContractSnapshot, CrossRepoProvenanceSchema, type CrossRepoReport, type CrossRepoSource, DECAY_DAYS, DEFAULT_AUTO_PROMOTE_RULE, DEFAULT_CONFIDENCE_THRESHOLDS, DEFAULT_CONFIG, DEFAULT_DORMANT_DAYS, type DepChange, type DepTrackResult, type DependencySnapshot, type EvalReport, type EvalSpec, GUESSABLE_THRESHOLD, HAIVE_DIR, type HaiveConfig, type HaivePaths, type ImpactOptions, type ImpactScore, type ImpactSummary, type ImpactTier, type LexicalRankResult, type LoadedMemory, MEMORIES_DIR, type Memory, type MemoryFrontmatter, MemoryFrontmatterSchema, type MemoryScope, MemoryScopeSchema, type MemoryStatus, MemoryStatusSchema, type MemoryType, MemoryTypeSchema, type MemoryUsage, PROJECT_CONTEXT_FILE, RUNTIME_JOURNAL_FILENAME, type ResolveProjectInfo, type RetirementSignal, type RetrievalAggregate, type RetrievalCase, type RetrievalCaseResult, type RuntimeJournalEntry, SESSION_RECAP_TTL_MS, STACK_PACK_TAG, type SelfEvalOptions, type Sensor, type SensorAggregate, type SensorCase, type SensorCaseResult, type SensorHit, SensorSchema, type SensorSuggestionOptions, type SensorTarget, type SkillActivation, type TimelineEntry, type TopicStatusPair, type TruncateOptions, type TruncateResult, USAGE_FILE, USAGE_LOG_DIR, USAGE_LOG_FILE, type UsageAggregate, type UsageEvent, type UsageIndex, type VerifyOptions, type VerifyResult, addedLinesFromDiff, aggregateRetrieval, aggregateSensors, aggregateUsage, allocateBudget, antiPatternGateParams, appendRuntimeJournalEntry, appendUsageEvent, briefingMarkerPath, briefingMarkersDir, buildCodeMap, buildFrontmatter, buildReport, bumpRead, codeMapPath, collectTimelineEntries, compareImpact, compileRegexSensor, computeImpact, configPath, contractLockPath, deriveConfidence, diffContract, emptyUsage, emptyUsageIndex, enforcementDir, estimateTokens, evaluateSkillActivation, extractActionsBriefBody, extractSnippet, findLexicalConflictPairs, findProjectRoot, findTopicStatusConflictPairs, firstMemoryOneLine, getUsage, globToRegExp, hasRecentBriefingMarker, inferModulesFromPaths, isAutoPromoteEligible, isDecaying, isFreshIsoDate, isGlobPath, isLikelyGuessable, isRetiredMemory, isSkill, isSkillSuppressed, isStackPackSeed, listMarkdownFilesRecursive, literalMatchesAllTokens, literalMatchesAnyToken, loadCodeMap, loadConfig, loadConfigSync, loadMemoriesFromDir, loadMemory, loadUsageIndex, memoryFilePath, memoryMatchesAnchorPaths, newMemoryId, normalizeSessionId, overallScore, parseMemory, parseSince, pathsOverlap, pickSnippetNeedle, pullCrossRepoSources, queryCodeMap, rankMemoriesLexical, readRecentBriefingMarker, readRuntimeJournalTail, readUsageEvents, recordApplied, recordRejection, relPathFrom, resolveBriefingBudget, resolveHaivePaths, resolveManifestFiles, resolveProjectInfo, retirementSignal, runRegexSensor, runSensors, runtimeJournalPath, saveCodeMap, saveConfig, saveUsageIndex, scoreRetrievalCase, scoreSensorCase, sensorAppliesToPath, sensorTargetsFromDiff, serializeMemory, snapshotContract, specificityScore, stripPrivate, suggestSensorFromMemory, suggestTopicKey, summarizeImpact, synthesizeSelfEvalCases, titleFromBody, tokenizeQuery, trackDependencies, trackReads, truncateToTokens, usageLogPath, usageLogSize, usagePath, verifyAnchor, watchContracts, writeBriefingMarker };
+export { AUTOPILOT_DEFAULTS, type Activation, type ActivationContext, ActivationSchema, type Anchor, AnchorSchema, type AntiPatternGate, type AutoPromoteRule, BRIEFING_MARKER_TTL_MS, BRIEFING_PRESET_DEFAULTS, type BreakingChange, type BriefingBudgetNumbers, type BriefingBudgetPreset, type BriefingMarker, type BudgetPart, type BudgetSlice, type BuildCodeMapOptions, CHARS_PER_TOKEN, CODE_MAP_FILE, CODE_STOPWORDS, CONFIG_FILE, type CodeExport, type CodeExportKind, type CodeFileEntry, type CodeMap, type CodeMapQueryOptions, type CollectTimelineOpts, type ConfidenceLevel, type ConfidenceThresholds, type ConflictCandidatePair, type ConflictCandidatesOpts, type ContractDiffResult, type ContractFile, type ContractSnapshot, CrossRepoProvenanceSchema, type CrossRepoReport, type CrossRepoSource, DECAY_DAYS, DEFAULT_AUTO_PROMOTE_RULE, DEFAULT_CONFIDENCE_THRESHOLDS, DEFAULT_CONFIG, DEFAULT_DORMANT_DAYS, type DepChange, type DepTrackResult, type DependencySnapshot, type DocFrequency, type EvalReport, type EvalSpec, GUESSABLE_THRESHOLD, HAIVE_DIR, type HaiveConfig, type HaivePaths, type ImpactOptions, type ImpactScore, type ImpactSummary, type ImpactTier, type LexicalRankResult, type LoadedMemory, MEMORIES_DIR, MIN_WORD_LEN, type Memory, type MemoryFrontmatter, MemoryFrontmatterSchema, type MemoryScope, MemoryScopeSchema, type MemoryStatus, MemoryStatusSchema, type MemoryType, MemoryTypeSchema, type MemoryUsage, PROJECT_CONTEXT_FILE, RUNTIME_JOURNAL_FILENAME, type ResolveProjectInfo, type RetirementSignal, type RetrievalAggregate, type RetrievalCase, type RetrievalCaseResult, type RuntimeJournalEntry, SESSION_RECAP_TTL_MS, STACK_PACK_TAG, type SelfEvalOptions, type Sensor, type SensorAggregate, type SensorCase, type SensorCaseResult, type SensorHit, SensorSchema, type SensorSuggestionOptions, type SensorTarget, type SkillActivation, type TimelineEntry, type TopicStatusPair, type TruncateOptions, type TruncateResult, USAGE_FILE, USAGE_LOG_DIR, USAGE_LOG_FILE, type UsageAggregate, type UsageEvent, type UsageIndex, type VerifyOptions, type VerifyResult, addedLinesFromDiff, aggregateRetrieval, aggregateSensors, aggregateUsage, allocateBudget, antiPatternGateParams, appendRuntimeJournalEntry, appendUsageEvent, briefingMarkerPath, briefingMarkersDir, buildCodeMap, buildDocFrequency, buildFrontmatter, buildReport, bumpRead, codeMapPath, collectTimelineEntries, compareImpact, compileRegexSensor, computeImpact, configPath, contractLockPath, deriveConfidence, diffContract, diffHasDistinctiveOverlap, distinctiveCap, emptyUsage, emptyUsageIndex, enforcementDir, estimateTokens, evaluateSkillActivation, extractActionsBriefBody, extractSnippet, findLexicalConflictPairs, findProjectRoot, findTopicStatusConflictPairs, firstMemoryOneLine, getUsage, globToRegExp, hasRecentBriefingMarker, inferModulesFromPaths, isAutoPromoteEligible, isDecaying, isDistinctiveToken, isFreshIsoDate, isGlobPath, isLikelyGuessable, isRetiredMemory, isSkill, isSkillSuppressed, isStackPackSeed, listMarkdownFilesRecursive, literalMatchesAllTokens, literalMatchesAnyToken, loadCodeMap, loadConfig, loadConfigSync, loadMemoriesFromDir, loadMemory, loadUsageIndex, memoryFilePath, memoryMatchesAnchorPaths, newMemoryId, normalizeSessionId, overallScore, parseMemory, parseSince, pathsOverlap, pickSnippetNeedle, pullCrossRepoSources, queryCodeMap, rankMemoriesLexical, readRecentBriefingMarker, readRuntimeJournalTail, readUsageEvents, recordApplied, recordRejection, relPathFrom, resolveBriefingBudget, resolveHaivePaths, resolveManifestFiles, resolveProjectInfo, retirementSignal, runRegexSensor, runSensors, runtimeJournalPath, saveCodeMap, saveConfig, saveUsageIndex, scoreRetrievalCase, scoreSensorCase, sensorAppliesToPath, sensorTargetsFromDiff, serializeMemory, snapshotContract, specificityScore, stripPrivate, suggestSensorFromMemory, suggestTopicKey, summarizeImpact, synthesizeSelfEvalCases, titleFromBody, tokenizeQuery, tokenizeWords, trackDependencies, trackReads, truncateToTokens, usageLogPath, usageLogSize, usagePath, verifyAnchor, watchContracts, writeBriefingMarker };

package/dist/index.js

@@ -926,6 +926,85 @@ function isAutoPromoteEligible(fm, usage, rule = DEFAULT_AUTO_PROMOTE_RULE) {
   return usage.read_count >= rule.minReads;
 }
 
+// src/distinctive.ts
+var CODE_STOPWORDS = /* @__PURE__ */ new Set([
+  "import",
+  "export",
+  "function",
+  "return",
+  "const",
+  "let",
+  "var",
+  "class",
+  "public",
+  "private",
+  "protected",
+  "static",
+  "this",
+  "true",
+  "false",
+  "null",
+  "undefined",
+  "void",
+  "async",
+  "await",
+  "from",
+  "type",
+  "interface",
+  "extends",
+  "implements",
+  "number",
+  "string",
+  "boolean",
+  "value",
+  "default",
+  "case",
+  "break",
+  "continue",
+  "throw",
+  "catch",
+  "finally",
+  "else",
+  "while",
+  "for",
+  "new",
+  "super",
+  "yield",
+  "module",
+  "require",
+  "console"
+]);
+var MIN_WORD_LEN = 4;
+function tokenizeWords(text) {
+  return text.toLowerCase().split(/[^a-z0-9]+/).filter((t) => t.length >= MIN_WORD_LEN && !CODE_STOPWORDS.has(t));
+}
+function buildDocFrequency(bodies) {
+  const df = /* @__PURE__ */ new Map();
+  for (const body of bodies) {
+    const unique = new Set(tokenizeWords(body));
+    for (const tok of unique) df.set(tok, (df.get(tok) ?? 0) + 1);
+  }
+  return { df, total: bodies.length };
+}
+function distinctiveCap(total) {
+  return Math.max(1, Math.floor(0.1 * total));
+}
+function isDistinctiveToken(token, freq) {
+  const tok = token.toLowerCase();
+  if (tok.length < MIN_WORD_LEN || CODE_STOPWORDS.has(tok)) return false;
+  const df = freq.df.get(tok);
+  if (df === void 0) return true;
+  return df <= distinctiveCap(freq.total);
+}
+function diffHasDistinctiveOverlap(addedDiffText, memoryBody, freq) {
+  const memoryTokens = new Set(tokenizeWords(memoryBody));
+  if (memoryTokens.size === 0) return false;
+  for (const tok of new Set(tokenizeWords(addedDiffText))) {
+    if (memoryTokens.has(tok) && isDistinctiveToken(tok, freq)) return true;
+  }
+  return false;
+}
+
 // src/skill-activation.ts
 function isSkill(fm) {
   return fm.type === "skill";
@@ -2894,7 +2973,7 @@ function pickLowercaseValuePattern(text) {
   for (const match of text.matchAll(/\blowercase\s+([A-Za-z][A-Za-z0-9_.:-]{2,79})\s+([a-z][a-z0-9_.:-]{1,40})\b/g)) {
     const key = match[1] ?? "";
     const value = match[2] ?? "";
-    if (!isDistinctiveToken(key, true) || isBoringValue(value)) continue;
+    if (!isDistinctiveToken2(key, true) || isBoringValue(value)) continue;
     candidates.push({
       label: `${key}=${value}`,
       pattern: `${escapeRegExp(key)}\\s*[:=]\\s*["']?${escapeRegExp(value)}["']?`,
@@ -2910,7 +2989,7 @@ function pickAssignmentPattern(text) {
       const key = match[1] ?? "";
       const operator = match[2] ?? "";
       const value = match[3] ?? "";
-      if (!isDistinctiveToken(key, true) || isBoringValue(value)) continue;
+      if (!isDistinctiveToken2(key, true) || isBoringValue(value)) continue;
       const label = `${key}${operator}${value}`;
       candidates.push({
         label,
@@ -2941,7 +3020,7 @@ function pickDistinctiveToken(text) {
     const raw = (match[1] ?? match[2] ?? match[3] ?? "").trim();
     const token = raw.replace(/^[^\w.-]+|[^\w.-]+$/g, "");
     const isCodeLike = Boolean(match[1] ?? match[2]);
-    if (!isDistinctiveToken(token, isCodeLike)) continue;
+    if (!isDistinctiveToken2(token, isCodeLike)) continue;
     const key = token.toLowerCase();
     const codeSpanBonus = match[1] ? 20 : match[2] ? 8 : 0;
     const shapeBonus = /[-_.:]/.test(token) ? 3 : /[A-Z]/.test(token.slice(1)) ? 2 : /\d/.test(token) ? 1 : 0;
@@ -2952,7 +3031,7 @@ function pickDistinctiveToken(text) {
   const best = [...candidates.values()].sort((a, b) => b.score - a.score)[0];
   return best?.raw ?? null;
 }
-function isDistinctiveToken(token, isCodeLike) {
+function isDistinctiveToken2(token, isCodeLike) {
   if (token.length < 4 || token.length > 80) return false;
   if (/^https?:\/\//i.test(token)) return false;
   if (/^\d+$/.test(token)) return false;
@@ -2985,6 +3064,7 @@ export {
   BRIEFING_PRESET_DEFAULTS,
   CHARS_PER_TOKEN,
   CODE_MAP_FILE,
+  CODE_STOPWORDS,
   CONFIG_FILE,
   CrossRepoProvenanceSchema,
   DECAY_DAYS,
@@ -2995,6 +3075,7 @@ export {
   GUESSABLE_THRESHOLD,
   HAIVE_DIR,
   MEMORIES_DIR,
+  MIN_WORD_LEN,
   MemoryFrontmatterSchema,
   MemoryScopeSchema,
   MemoryStatusSchema,
@@ -3018,6 +3099,7 @@ export {
   briefingMarkerPath,
   briefingMarkersDir,
   buildCodeMap,
+  buildDocFrequency,
   buildFrontmatter,
   buildReport,
   bumpRead,
@@ -3030,6 +3112,8 @@ export {
   contractLockPath,
   deriveConfidence,
   diffContract,
+  diffHasDistinctiveOverlap,
+  distinctiveCap,
   emptyUsage,
   emptyUsageIndex,
   enforcementDir,
@@ -3047,6 +3131,7 @@ export {
   inferModulesFromPaths,
   isAutoPromoteEligible,
   isDecaying,
+  isDistinctiveToken,
   isFreshIsoDate,
   isGlobPath,
   isLikelyGuessable,
@@ -3106,6 +3191,7 @@ export {
   synthesizeSelfEvalCases,
   titleFromBody,
   tokenizeQuery,
+  tokenizeWords,
   trackDependencies,
   trackReads,
   truncateToTokens,