sentinelayer-cli 0.1.2 → 0.4.4

package/src/memory/retrieval.js

@@ -1,581 +1,581 @@
-import fsp from "node:fs/promises";
-import path from "node:path";
-
-function normalizeString(value) {
-  return String(value || "").trim();
-}
-
-function normalizeNumber(value, fallback = 0) {
-  const parsed = Number(value);
-  if (!Number.isFinite(parsed)) {
-    return fallback;
-  }
-  return parsed;
-}
-
-function normalizeSeverity(value) {
-  const severity = normalizeString(value).toUpperCase();
-  if (severity === "P0" || severity === "P1" || severity === "P2" || severity === "P3") {
-    return severity;
-  }
-  return "P3";
-}
-
-export function tokenize(value) {
-  return Array.from(
-    new Set(
-      normalizeString(value)
-        .toLowerCase()
-        .split(/[^a-z0-9_]+/g)
-        .map((token) => token.trim())
-        .filter((token) => token.length >= 2)
-    )
-  );
-}
-
-function tokenizeWithFrequency(value) {
-  const tokens = normalizeString(value)
-    .toLowerCase()
-    .split(/[^a-z0-9_]+/g)
-    .map((token) => token.trim())
-    .filter((token) => token.length >= 2);
-  const frequency = new Map();
-  for (const token of tokens) {
-    frequency.set(token, (frequency.get(token) || 0) + 1);
-  }
-  return {
-    tokens,
-    uniqueTokens: Array.from(new Set(tokens)),
-    frequency,
-  };
-}
-
-function toPosixPath(value) {
-  return normalizeString(value).replace(/\\/g, "/");
-}
-
-function clipSnippet(text, maxLength = 180) {
-  const normalized = normalizeString(text).replace(/\s+/g, " ");
-  if (normalized.length <= maxLength) {
-    return normalized;
-  }
-  return `${normalized.slice(0, maxLength - 3)}...`;
-}
-
-const SEVERITY_SCORE = Object.freeze({
-  P0: 1,
-  P1: 0.9,
-  P2: 0.65,
-  P3: 0.4,
-});
-
-function parseTimestamp(value) {
-  const millis = Date.parse(value);
-  if (!Number.isFinite(millis)) {
-    return 0;
-  }
-  return millis;
-}
-
-function buildDocumentVector({ frequency, tokenCount, idfByToken }) {
-  const vector = new Map();
-  const denominator = Math.max(1, tokenCount);
-  for (const [token, count] of frequency.entries()) {
-    const idf = idfByToken.get(token) || 0;
-    const tf = count / denominator;
-    const weight = tf * idf;
-    if (weight > 0) {
-      vector.set(token, weight);
-    }
-  }
-  return vector;
-}
-
-function vectorMagnitude(vector) {
-  let sum = 0;
-  for (const value of vector.values()) {
-    sum += value * value;
-  }
-  return Math.sqrt(sum);
-}
-
-function cosineSimilarity(leftVector, leftMagnitude, rightVector, rightMagnitude) {
-  if (!leftVector || !rightVector || leftMagnitude <= 0 || rightMagnitude <= 0) {
-    return 0;
-  }
-  const [smaller, larger] =
-    leftVector.size <= rightVector.size ? [leftVector, rightVector] : [rightVector, leftVector];
-  let dot = 0;
-  for (const [token, value] of smaller.entries()) {
-    const counterpart = larger.get(token);
-    if (counterpart) {
-      dot += value * counterpart;
-    }
-  }
-  if (dot <= 0) {
-    return 0;
-  }
-  return dot / (leftMagnitude * rightMagnitude);
-}
-
-function buildDeterministicScore({ document, queryTokens, cosine, recency }) {
-  const tokenSet = document.tokenSet;
-  const overlapCount = queryTokens.filter((token) => tokenSet.has(token)).length;
-  const tokenOverlap = queryTokens.length > 0 ? overlapCount / queryTokens.length : 0;
-  const queryText = normalizeString(document.queryTextForExact).toLowerCase();
-  const exactMatch =
-    queryTokens.length > 0 &&
-    queryTokens.every((token) => queryText.includes(token))
-      ? 1
-      : 0;
-  const severityScore = SEVERITY_SCORE[normalizeSeverity(document.severity)] || 0.35;
-  return {
-    exactMatch,
-    tokenOverlap,
-    cosine,
-    recency,
-    severityScore,
-    score:
-      exactMatch * 0.25 +
-      tokenOverlap * 0.25 +
-      cosine * 0.3 +
-      recency * 0.1 +
-      severityScore * 0.1,
-  };
-}
-
-function resolveSpecCandidates({ targetPath, specFile }) {
-  if (normalizeString(specFile)) {
-    const absoluteCandidate = path.isAbsolute(specFile)
-      ? specFile
-      : path.join(targetPath, specFile);
-    return [absoluteCandidate];
-  }
-  return [path.join(targetPath, "SPEC.md"), path.join(targetPath, "docs", "spec.md")];
-}
-
-async function loadSpecDocument({ targetPath, specFile = "" }) {
-  const candidates = resolveSpecCandidates({ targetPath, specFile });
-  for (const candidate of candidates) {
-    try {
-      const text = await fsp.readFile(candidate, "utf-8");
-      const normalizedPath = toPosixPath(path.resolve(candidate));
-      return {
-        documentId: `spec:${normalizedPath}`,
-        sourceType: "spec",
-        sourcePath: normalizedPath,
-        severity: "P3",
-        updatedAt: new Date().toISOString(),
-        text,
-        metadata: {
-          category: "spec",
-        },
-      };
-    } catch {
-      // Continue to next candidate.
-    }
-  }
-  return null;
-}
-
-async function loadAuditHistoryDocuments({ outputRoot, excludeRunId = "", maxAuditRuns = 3 }) {
-  const auditsDirectory = path.join(outputRoot, "audits");
-  let runEntries = [];
-  try {
-    runEntries = await fsp.readdir(auditsDirectory, { withFileTypes: true });
-  } catch {
-    return [];
-  }
-  const runIds = runEntries
-    .filter((entry) => entry.isDirectory() && entry.name.startsWith("audit-"))
-    .map((entry) => entry.name)
-    .filter((runId) => runId !== normalizeString(excludeRunId))
-    .sort((left, right) => right.localeCompare(left))
-    .slice(0, Math.max(0, Math.floor(Number(maxAuditRuns || 0))));
-
-  const documents = [];
-  for (const runId of runIds) {
-    const reportPath = path.join(auditsDirectory, runId, "AUDIT_REPORT.json");
-    try {
-      const report = JSON.parse(await fsp.readFile(reportPath, "utf-8"));
-      documents.push({
-        documentId: `audit:${runId}:summary`,
-        sourceType: "audit-history",
-        sourcePath: toPosixPath(reportPath),
-        severity: report.summary?.blocking ? "P1" : "P3",
-        updatedAt: normalizeString(report.generatedAt) || new Date().toISOString(),
-        text: [
-          `run_id ${runId}`,
-          `summary p0 ${report.summary?.P0 || 0}`,
-          `p1 ${report.summary?.P1 || 0}`,
-          `p2 ${report.summary?.P2 || 0}`,
-          `p3 ${report.summary?.P3 || 0}`,
-          `risk_surfaces ${(report.ingest?.riskSurfaces || []).join(" ")}`,
-        ].join(" "),
-        metadata: {
-          category: "audit-summary",
-          runId,
-        },
-      });
-
-      let findingIndex = 0;
-      for (const agent of report.agentResults || []) {
-        for (const finding of agent.findings || []) {
-          if (findingIndex >= 60) {
-            break;
-          }
-          findingIndex += 1;
-          documents.push({
-            documentId: `audit:${runId}:finding:${findingIndex}`,
-            sourceType: "audit-history",
-            sourcePath: toPosixPath(reportPath),
-            severity: normalizeSeverity(finding.severity),
-            updatedAt: normalizeString(report.generatedAt) || new Date().toISOString(),
-            text: [finding.file, finding.layer, finding.ruleId, finding.message].filter(Boolean).join(" "),
-            metadata: {
-              category: "audit-finding",
-              runId,
-              agentId: agent.agentId,
-              file: toPosixPath(finding.file),
-              line: normalizeNumber(finding.line, 0),
-            },
-          });
-        }
-      }
-    } catch {
-      // Ignore malformed historical run artifacts.
-    }
-  }
-  return documents;
-}
-
-function buildIngestDocuments(ingest = {}) {
-  const summary = ingest.summary || {};
-  const frameworkList = Array.isArray(ingest.frameworks) ? ingest.frameworks : [];
-  const riskSurfaces = Array.isArray(ingest.riskSurfaces) ? ingest.riskSurfaces : [];
-  const documents = [
-    {
-      documentId: "ingest:summary",
-      sourceType: "ingest",
-      sourcePath: "",
-      severity: "P3",
-      updatedAt: new Date().toISOString(),
-      text: [
-        `files_scanned ${summary.filesScanned || 0}`,
-        `total_loc ${summary.totalLoc || 0}`,
-        `frameworks ${frameworkList.join(" ")}`,
-      ].join(" "),
-      metadata: {
-        category: "ingest-summary",
-      },
-    },
-  ];
-  for (const [index, surface] of riskSurfaces.entries()) {
-    const surfaceName = normalizeString(surface?.surface || surface);
-    if (!surfaceName) {
-      continue;
-    }
-    documents.push({
-      documentId: `ingest:surface:${index + 1}`,
-      sourceType: "ingest",
-      sourcePath: "",
-      severity: "P2",
-      updatedAt: new Date().toISOString(),
-      text: `risk_surface ${surfaceName}`,
-      metadata: {
-        category: "ingest-risk-surface",
-        surface: surfaceName,
-      },
-    });
-  }
-  return documents;
-}
-
-function buildLocalDocument(raw = {}, index = 0) {
-  const text = normalizeString(raw.text);
-  const tokenized = tokenizeWithFrequency(text);
-  const updatedAt = normalizeString(raw.updatedAt) || new Date().toISOString();
-  const sequence = index + 1;
-  return {
-    documentId: normalizeString(raw.documentId) || `doc-${String(sequence).padStart(6, "0")}`,
-    sourceType: normalizeString(raw.sourceType) || "memory",
-    sourcePath: toPosixPath(raw.sourcePath),
-    severity: normalizeSeverity(raw.severity),
-    updatedAt,
-    text,
-    snippet: clipSnippet(text),
-    metadata: raw.metadata && typeof raw.metadata === "object" ? raw.metadata : {},
-    sequence,
-    tokenSet: new Set(tokenized.uniqueTokens),
-    tokenFrequency: tokenized.frequency,
-    tokenCount: tokenized.tokens.length,
-    queryTextForExact: text,
-    timestampMs: parseTimestamp(updatedAt),
-  };
-}
-
-export function buildDocumentsFromBlackboardEntries(entries = []) {
-  return (entries || []).map((entry, index) => ({
-    documentId: `blackboard:${normalizeString(entry.entryId) || index + 1}`,
-    sourceType: "blackboard",
-    sourcePath: "",
-    severity: normalizeSeverity(entry.severity),
-    updatedAt: normalizeString(entry.timestamp) || new Date().toISOString(),
-    text: [
-      entry.agentId,
-      entry.source,
-      entry.file,
-      entry.layer,
-      entry.ruleId,
-      entry.message,
-      entry.note,
-    ]
-      .filter(Boolean)
-      .join(" "),
-    metadata: {
-      category: "blackboard-entry",
-      entryId: entry.entryId || "",
-      line: normalizeNumber(entry.line, 0),
-      file: toPosixPath(entry.file),
-    },
-  }));
-}
-
-export async function buildSharedMemoryCorpus({
-  outputRoot = "",
-  targetPath = "",
-  ingest = {},
-  specFile = "",
-  excludeRunId = "",
-  maxAuditRuns = 3,
-} = {}) {
-  const resolvedOutputRoot = path.resolve(String(outputRoot || "."));
-  const resolvedTargetPath = path.resolve(String(targetPath || "."));
-
-  const [historyDocuments, specDocument] = await Promise.all([
-    loadAuditHistoryDocuments({
-      outputRoot: resolvedOutputRoot,
-      excludeRunId,
-      maxAuditRuns,
-    }),
-    loadSpecDocument({
-      targetPath: resolvedTargetPath,
-      specFile,
-    }),
-  ]);
-
-  const documents = [
-    ...buildIngestDocuments(ingest),
-    ...historyDocuments,
-    ...(specDocument ? [specDocument] : []),
-  ];
-
-  const sourceCounts = documents.reduce(
-    (counts, document) => {
-      const key = normalizeString(document.sourceType) || "unknown";
-      counts[key] = (counts[key] || 0) + 1;
-      return counts;
-    },
-    {}
-  );
-
-  return {
-    documents,
-    sourceCounts,
-    hasSpecDocument: Boolean(specDocument),
-    historyRunDocumentCount: historyDocuments.length,
-  };
-}
-
-export function buildLocalHybridIndex(documents = []) {
-  const preparedDocuments = (documents || [])
-    .map((document, index) => buildLocalDocument(document, index))
-    .filter((document) => document.text.length > 0);
-
-  const documentCount = preparedDocuments.length;
-  const docFrequency = new Map();
-  for (const document of preparedDocuments) {
-    for (const token of document.tokenSet.values()) {
-      docFrequency.set(token, (docFrequency.get(token) || 0) + 1);
-    }
-  }
-
-  const idfByToken = new Map();
-  for (const [token, frequency] of docFrequency.entries()) {
-    const idf = Math.log((documentCount + 1) / (frequency + 1)) + 1;
-    idfByToken.set(token, idf);
-  }
-
-  const timestampValues = preparedDocuments
-    .map((document) => document.timestampMs)
-    .filter((value) => Number.isFinite(value) && value > 0);
-  const minTimestamp = timestampValues.length > 0 ? Math.min(...timestampValues) : 0;
-  const maxTimestamp = timestampValues.length > 0 ? Math.max(...timestampValues) : 0;
-
-  const entries = preparedDocuments.map((document) => {
-    const vector = buildDocumentVector({
-      frequency: document.tokenFrequency,
-      tokenCount: document.tokenCount,
-      idfByToken,
-    });
-    const magnitude = vectorMagnitude(vector);
-    let recency = 0.5;
-    if (maxTimestamp > minTimestamp && document.timestampMs > 0) {
-      recency = (document.timestampMs - minTimestamp) / (maxTimestamp - minTimestamp);
-    }
-    return {
-      ...document,
-      vector,
-      magnitude,
-      recency,
-    };
-  });
-
-  return {
-    schemaVersion: "1.0.0",
-    generatedAt: new Date().toISOString(),
-    documentCount,
-    vocabularySize: idfByToken.size,
-    idfByToken,
-    entries,
-  };
-}
-
-export function queryLocalHybridIndex(index, { query = "", limit = 12, minScore = 0 } = {}) {
-  const queryText = normalizeString(query);
-  const queryTokenized = tokenizeWithFrequency(queryText);
-  const queryTokens = queryTokenized.uniqueTokens;
-  const queryVector = buildDocumentVector({
-    frequency: queryTokenized.frequency,
-    tokenCount: queryTokenized.tokens.length,
-    idfByToken: index.idfByToken || new Map(),
-  });
-  const queryMagnitude = vectorMagnitude(queryVector);
-
-  const normalizedLimit = Math.max(1, Math.floor(normalizeNumber(limit, 12)));
-  const normalizedMinScore = Math.max(0, Math.min(1, normalizeNumber(minScore, 0)));
-  const results = (index.entries || [])
-    .map((document) => {
-      const cosine = cosineSimilarity(queryVector, queryMagnitude, document.vector, document.magnitude);
-      const signals = buildDeterministicScore({
-        document,
-        queryTokens,
-        cosine,
-        recency: document.recency,
-      });
-      return {
-        documentId: document.documentId,
-        sourceType: document.sourceType,
-        sourcePath: document.sourcePath,
-        severity: document.severity,
-        updatedAt: document.updatedAt,
-        snippet: document.snippet,
-        metadata: document.metadata,
-        score: Number(signals.score.toFixed(6)),
-        scoreBreakdown: {
-          exactMatch: Number(signals.exactMatch.toFixed(6)),
-          tokenOverlap: Number(signals.tokenOverlap.toFixed(6)),
-          cosine: Number(signals.cosine.toFixed(6)),
-          recency: Number(signals.recency.toFixed(6)),
-          severity: Number(signals.severityScore.toFixed(6)),
-        },
-      };
-    })
-    .filter((result) => result.score >= normalizedMinScore)
-    .sort((left, right) => {
-      if (right.score !== left.score) {
-        return right.score - left.score;
-      }
-      return left.documentId.localeCompare(right.documentId);
-    })
-    .slice(0, normalizedLimit);
-
-  return {
-    query: queryText,
-    limit: normalizedLimit,
-    minScore: normalizedMinScore,
-    queryTokenCount: queryTokens.length,
-    results,
-  };
-}
-
-export async function queryHybridRetriever({
-  query = "",
-  documents = [],
-  limit = 12,
-  provider = "local",
-  apiEndpoint = "",
-  apiKey = "",
-  fetchImpl = globalThis.fetch,
-} = {}) {
-  const normalizedProvider = normalizeString(provider).toLowerCase() || "local";
-  const localIndex = buildLocalHybridIndex(documents);
-  const localQuery = queryLocalHybridIndex(localIndex, {
-    query,
-    limit,
-  });
-
-  const canUseApi =
-    (normalizedProvider === "api" || normalizedProvider === "auto") &&
-    normalizeString(apiEndpoint) &&
-    typeof fetchImpl === "function";
-  if (!canUseApi) {
-    return {
-      providerRequested: normalizedProvider,
-      providerUsed: "local",
-      apiFallback: false,
-      apiError: "",
-      indexSummary: {
-        documentCount: localIndex.documentCount,
-        vocabularySize: localIndex.vocabularySize,
-      },
-      results: localQuery.results,
-    };
-  }
-
-  try {
-    const response = await fetchImpl(String(apiEndpoint), {
-      method: "POST",
-      headers: {
-        "content-type": "application/json",
-        ...(normalizeString(apiKey) ? { authorization: `Bearer ${apiKey}` } : {}),
-      },
-      body: JSON.stringify({
-        query: normalizeString(query),
-        limit: Math.max(1, Math.floor(normalizeNumber(limit, 12))),
-        documents,
-      }),
-    });
-    if (!response.ok) {
-      throw new Error(`Memory API request failed (${response.status}).`);
-    }
-    const payload = await response.json();
-    const apiResults = Array.isArray(payload?.results) ? payload.results : [];
-    if (apiResults.length === 0) {
-      throw new Error("Memory API response missing results.");
-    }
-    return {
-      providerRequested: normalizedProvider,
-      providerUsed: "api",
-      apiFallback: false,
-      apiError: "",
-      indexSummary: {
-        documentCount: localIndex.documentCount,
-        vocabularySize: localIndex.vocabularySize,
-      },
-      results: apiResults.slice(0, Math.max(1, Math.floor(normalizeNumber(limit, 12)))),
-    };
-  } catch (error) {
-    return {
-      providerRequested: normalizedProvider,
-      providerUsed: "local",
-      apiFallback: true,
-      apiError: normalizeString(error?.message || error),
-      indexSummary: {
-        documentCount: localIndex.documentCount,
-        vocabularySize: localIndex.vocabularySize,
-      },
-      results: localQuery.results,
-    };
-  }
-}
+import fsp from "node:fs/promises";
+import path from "node:path";
+
+function normalizeString(value) {
+  return String(value || "").trim();
+}
+
+function normalizeNumber(value, fallback = 0) {
+  const parsed = Number(value);
+  if (!Number.isFinite(parsed)) {
+    return fallback;
+  }
+  return parsed;
+}
+
+function normalizeSeverity(value) {
+  const severity = normalizeString(value).toUpperCase();
+  if (severity === "P0" || severity === "P1" || severity === "P2" || severity === "P3") {
+    return severity;
+  }
+  return "P3";
+}
+
+export function tokenize(value) {
+  return Array.from(
+    new Set(
+      normalizeString(value)
+        .toLowerCase()
+        .split(/[^a-z0-9_]+/g)
+        .map((token) => token.trim())
+        .filter((token) => token.length >= 2)
+    )
+  );
+}
+
+function tokenizeWithFrequency(value) {
+  const tokens = normalizeString(value)
+    .toLowerCase()
+    .split(/[^a-z0-9_]+/g)
+    .map((token) => token.trim())
+    .filter((token) => token.length >= 2);
+  const frequency = new Map();
+  for (const token of tokens) {
+    frequency.set(token, (frequency.get(token) || 0) + 1);
+  }
+  return {
+    tokens,
+    uniqueTokens: Array.from(new Set(tokens)),
+    frequency,
+  };
+}
+
+function toPosixPath(value) {
+  return normalizeString(value).replace(/\\/g, "/");
+}
+
+function clipSnippet(text, maxLength = 180) {
+  const normalized = normalizeString(text).replace(/\s+/g, " ");
+  if (normalized.length <= maxLength) {
+    return normalized;
+  }
+  return `${normalized.slice(0, maxLength - 3)}...`;
+}
+
+const SEVERITY_SCORE = Object.freeze({
+  P0: 1,
+  P1: 0.9,
+  P2: 0.65,
+  P3: 0.4,
+});
+
+function parseTimestamp(value) {
+  const millis = Date.parse(value);
+  if (!Number.isFinite(millis)) {
+    return 0;
+  }
+  return millis;
+}
+
+function buildDocumentVector({ frequency, tokenCount, idfByToken }) {
+  const vector = new Map();
+  const denominator = Math.max(1, tokenCount);
+  for (const [token, count] of frequency.entries()) {
+    const idf = idfByToken.get(token) || 0;
+    const tf = count / denominator;
+    const weight = tf * idf;
+    if (weight > 0) {
+      vector.set(token, weight);
+    }
+  }
+  return vector;
+}
+
+function vectorMagnitude(vector) {
+  let sum = 0;
+  for (const value of vector.values()) {
+    sum += value * value;
+  }
+  return Math.sqrt(sum);
+}
+
+function cosineSimilarity(leftVector, leftMagnitude, rightVector, rightMagnitude) {
+  if (!leftVector || !rightVector || leftMagnitude <= 0 || rightMagnitude <= 0) {
+    return 0;
+  }
+  const [smaller, larger] =
+    leftVector.size <= rightVector.size ? [leftVector, rightVector] : [rightVector, leftVector];
+  let dot = 0;
+  for (const [token, value] of smaller.entries()) {
+    const counterpart = larger.get(token);
+    if (counterpart) {
+      dot += value * counterpart;
+    }
+  }
+  if (dot <= 0) {
+    return 0;
+  }
+  return dot / (leftMagnitude * rightMagnitude);
+}
+
+function buildDeterministicScore({ document, queryTokens, cosine, recency }) {
+  const tokenSet = document.tokenSet;
+  const overlapCount = queryTokens.filter((token) => tokenSet.has(token)).length;
+  const tokenOverlap = queryTokens.length > 0 ? overlapCount / queryTokens.length : 0;
+  const queryText = normalizeString(document.queryTextForExact).toLowerCase();
+  const exactMatch =
+    queryTokens.length > 0 &&
+    queryTokens.every((token) => queryText.includes(token))
+      ? 1
+      : 0;
+  const severityScore = SEVERITY_SCORE[normalizeSeverity(document.severity)] || 0.35;
+  return {
+    exactMatch,
+    tokenOverlap,
+    cosine,
+    recency,
+    severityScore,
+    score:
+      exactMatch * 0.25 +
+      tokenOverlap * 0.25 +
+      cosine * 0.3 +
+      recency * 0.1 +
+      severityScore * 0.1,
+  };
+}
+
+function resolveSpecCandidates({ targetPath, specFile }) {
+  if (normalizeString(specFile)) {
+    const absoluteCandidate = path.isAbsolute(specFile)
+      ? specFile
+      : path.join(targetPath, specFile);
+    return [absoluteCandidate];
+  }
+  return [path.join(targetPath, "SPEC.md"), path.join(targetPath, "docs", "spec.md")];
+}
+
+async function loadSpecDocument({ targetPath, specFile = "" }) {
+  const candidates = resolveSpecCandidates({ targetPath, specFile });
+  for (const candidate of candidates) {
+    try {
+      const text = await fsp.readFile(candidate, "utf-8");
+      const normalizedPath = toPosixPath(path.resolve(candidate));
+      return {
+        documentId: `spec:${normalizedPath}`,
+        sourceType: "spec",
+        sourcePath: normalizedPath,
+        severity: "P3",
+        updatedAt: new Date().toISOString(),
+        text,
+        metadata: {
+          category: "spec",
+        },
+      };
+    } catch {
+      // Continue to next candidate.
+    }
+  }
+  return null;
+}
+
+async function loadAuditHistoryDocuments({ outputRoot, excludeRunId = "", maxAuditRuns = 3 }) {
+  const auditsDirectory = path.join(outputRoot, "audits");
+  let runEntries = [];
+  try {
+    runEntries = await fsp.readdir(auditsDirectory, { withFileTypes: true });
+  } catch {
+    return [];
+  }
+  const runIds = runEntries
+    .filter((entry) => entry.isDirectory() && entry.name.startsWith("audit-"))
+    .map((entry) => entry.name)
+    .filter((runId) => runId !== normalizeString(excludeRunId))
+    .sort((left, right) => right.localeCompare(left))
+    .slice(0, Math.max(0, Math.floor(Number(maxAuditRuns || 0))));
+
+  const documents = [];
+  for (const runId of runIds) {
+    const reportPath = path.join(auditsDirectory, runId, "AUDIT_REPORT.json");
+    try {
+      const report = JSON.parse(await fsp.readFile(reportPath, "utf-8"));
+      documents.push({
+        documentId: `audit:${runId}:summary`,
+        sourceType: "audit-history",
+        sourcePath: toPosixPath(reportPath),
+        severity: report.summary?.blocking ? "P1" : "P3",
+        updatedAt: normalizeString(report.generatedAt) || new Date().toISOString(),
+        text: [
+          `run_id ${runId}`,
+          `summary p0 ${report.summary?.P0 || 0}`,
+          `p1 ${report.summary?.P1 || 0}`,
+          `p2 ${report.summary?.P2 || 0}`,
+          `p3 ${report.summary?.P3 || 0}`,
+          `risk_surfaces ${(report.ingest?.riskSurfaces || []).join(" ")}`,
+        ].join(" "),
+        metadata: {
+          category: "audit-summary",
+          runId,
+        },
+      });
+
+      let findingIndex = 0;
+      for (const agent of report.agentResults || []) {
+        for (const finding of agent.findings || []) {
+          if (findingIndex >= 60) {
+            break;
+          }
+          findingIndex += 1;
+          documents.push({
+            documentId: `audit:${runId}:finding:${findingIndex}`,
+            sourceType: "audit-history",
+            sourcePath: toPosixPath(reportPath),
+            severity: normalizeSeverity(finding.severity),
+            updatedAt: normalizeString(report.generatedAt) || new Date().toISOString(),
+            text: [finding.file, finding.layer, finding.ruleId, finding.message].filter(Boolean).join(" "),
+            metadata: {
+              category: "audit-finding",
+              runId,
+              agentId: agent.agentId,
+              file: toPosixPath(finding.file),
+              line: normalizeNumber(finding.line, 0),
+            },
+          });
+        }
+      }
+    } catch {
+      // Ignore malformed historical run artifacts.
+    }
+  }
+  return documents;
+}
+
+function buildIngestDocuments(ingest = {}) {
+  const summary = ingest.summary || {};
+  const frameworkList = Array.isArray(ingest.frameworks) ? ingest.frameworks : [];
+  const riskSurfaces = Array.isArray(ingest.riskSurfaces) ? ingest.riskSurfaces : [];
+  const documents = [
+    {
+      documentId: "ingest:summary",
+      sourceType: "ingest",
+      sourcePath: "",
+      severity: "P3",
+      updatedAt: new Date().toISOString(),
+      text: [
+        `files_scanned ${summary.filesScanned || 0}`,
+        `total_loc ${summary.totalLoc || 0}`,
+        `frameworks ${frameworkList.join(" ")}`,
+      ].join(" "),
+      metadata: {
+        category: "ingest-summary",
+      },
+    },
+  ];
+  for (const [index, surface] of riskSurfaces.entries()) {
+    const surfaceName = normalizeString(surface?.surface || surface);
+    if (!surfaceName) {
+      continue;
+    }
+    documents.push({
+      documentId: `ingest:surface:${index + 1}`,
+      sourceType: "ingest",
+      sourcePath: "",
+      severity: "P2",
+      updatedAt: new Date().toISOString(),
+      text: `risk_surface ${surfaceName}`,
+      metadata: {
+        category: "ingest-risk-surface",
+        surface: surfaceName,
+      },
+    });
+  }
+  return documents;
+}
+
+function buildLocalDocument(raw = {}, index = 0) {
+  const text = normalizeString(raw.text);
+  const tokenized = tokenizeWithFrequency(text);
+  const updatedAt = normalizeString(raw.updatedAt) || new Date().toISOString();
+  const sequence = index + 1;
+  return {
+    documentId: normalizeString(raw.documentId) || `doc-${String(sequence).padStart(6, "0")}`,
+    sourceType: normalizeString(raw.sourceType) || "memory",
+    sourcePath: toPosixPath(raw.sourcePath),
+    severity: normalizeSeverity(raw.severity),
+    updatedAt,
+    text,
+    snippet: clipSnippet(text),
+    metadata: raw.metadata && typeof raw.metadata === "object" ? raw.metadata : {},
+    sequence,
+    tokenSet: new Set(tokenized.uniqueTokens),
+    tokenFrequency: tokenized.frequency,
+    tokenCount: tokenized.tokens.length,
+    queryTextForExact: text,
+    timestampMs: parseTimestamp(updatedAt),
+  };
+}
+
+export function buildDocumentsFromBlackboardEntries(entries = []) {
+  return (entries || []).map((entry, index) => ({
+    documentId: `blackboard:${normalizeString(entry.entryId) || index + 1}`,
+    sourceType: "blackboard",
+    sourcePath: "",
+    severity: normalizeSeverity(entry.severity),
+    updatedAt: normalizeString(entry.timestamp) || new Date().toISOString(),
+    text: [
+      entry.agentId,
+      entry.source,
+      entry.file,
+      entry.layer,
+      entry.ruleId,
+      entry.message,
+      entry.note,
+    ]
+      .filter(Boolean)
+      .join(" "),
+    metadata: {
+      category: "blackboard-entry",
+      entryId: entry.entryId || "",
+      line: normalizeNumber(entry.line, 0),
+      file: toPosixPath(entry.file),
+    },
+  }));
+}
+
+export async function buildSharedMemoryCorpus({
+  outputRoot = "",
+  targetPath = "",
+  ingest = {},
+  specFile = "",
+  excludeRunId = "",
+  maxAuditRuns = 3,
+} = {}) {
+  const resolvedOutputRoot = path.resolve(String(outputRoot || "."));
+  const resolvedTargetPath = path.resolve(String(targetPath || "."));
+
+  const [historyDocuments, specDocument] = await Promise.all([
+    loadAuditHistoryDocuments({
+      outputRoot: resolvedOutputRoot,
+      excludeRunId,
+      maxAuditRuns,
+    }),
+    loadSpecDocument({
+      targetPath: resolvedTargetPath,
+      specFile,
+    }),
+  ]);
+
+  const documents = [
+    ...buildIngestDocuments(ingest),
+    ...historyDocuments,
+    ...(specDocument ? [specDocument] : []),
+  ];
+
+  const sourceCounts = documents.reduce(
+    (counts, document) => {
+      const key = normalizeString(document.sourceType) || "unknown";
+      counts[key] = (counts[key] || 0) + 1;
+      return counts;
+    },
+    {}
+  );
+
+  return {
+    documents,
+    sourceCounts,
+    hasSpecDocument: Boolean(specDocument),
+    historyRunDocumentCount: historyDocuments.length,
+  };
+}
+
+export function buildLocalHybridIndex(documents = []) {
+  const preparedDocuments = (documents || [])
+    .map((document, index) => buildLocalDocument(document, index))
+    .filter((document) => document.text.length > 0);
+
+  const documentCount = preparedDocuments.length;
+  const docFrequency = new Map();
+  for (const document of preparedDocuments) {
+    for (const token of document.tokenSet.values()) {
+      docFrequency.set(token, (docFrequency.get(token) || 0) + 1);
+    }
+  }
+
+  const idfByToken = new Map();
+  for (const [token, frequency] of docFrequency.entries()) {
+    const idf = Math.log((documentCount + 1) / (frequency + 1)) + 1;
+    idfByToken.set(token, idf);
+  }
+
+  const timestampValues = preparedDocuments
+    .map((document) => document.timestampMs)
+    .filter((value) => Number.isFinite(value) && value > 0);
+  const minTimestamp = timestampValues.length > 0 ? Math.min(...timestampValues) : 0;
+  const maxTimestamp = timestampValues.length > 0 ? Math.max(...timestampValues) : 0;
+
+  const entries = preparedDocuments.map((document) => {
+    const vector = buildDocumentVector({
+      frequency: document.tokenFrequency,
+      tokenCount: document.tokenCount,
+      idfByToken,
+    });
+    const magnitude = vectorMagnitude(vector);
+    let recency = 0.5;
+    if (maxTimestamp > minTimestamp && document.timestampMs > 0) {
+      recency = (document.timestampMs - minTimestamp) / (maxTimestamp - minTimestamp);
+    }
+    return {
+      ...document,
+      vector,
+      magnitude,
+      recency,
+    };
+  });
+
+  return {
+    schemaVersion: "1.0.0",
+    generatedAt: new Date().toISOString(),
+    documentCount,
+    vocabularySize: idfByToken.size,
+    idfByToken,
+    entries,
+  };
+}
+
+export function queryLocalHybridIndex(index, { query = "", limit = 12, minScore = 0 } = {}) {
+  const queryText = normalizeString(query);
+  const queryTokenized = tokenizeWithFrequency(queryText);
+  const queryTokens = queryTokenized.uniqueTokens;
+  const queryVector = buildDocumentVector({
+    frequency: queryTokenized.frequency,
+    tokenCount: queryTokenized.tokens.length,
+    idfByToken: index.idfByToken || new Map(),
+  });
+  const queryMagnitude = vectorMagnitude(queryVector);
+
+  const normalizedLimit = Math.max(1, Math.floor(normalizeNumber(limit, 12)));
+  const normalizedMinScore = Math.max(0, Math.min(1, normalizeNumber(minScore, 0)));
+  const results = (index.entries || [])
+    .map((document) => {
+      const cosine = cosineSimilarity(queryVector, queryMagnitude, document.vector, document.magnitude);
+      const signals = buildDeterministicScore({
+        document,
+        queryTokens,
+        cosine,
+        recency: document.recency,
+      });
+      return {
+        documentId: document.documentId,
+        sourceType: document.sourceType,
+        sourcePath: document.sourcePath,
+        severity: document.severity,
+        updatedAt: document.updatedAt,
+        snippet: document.snippet,
+        metadata: document.metadata,
+        score: Number(signals.score.toFixed(6)),
+        scoreBreakdown: {
+          exactMatch: Number(signals.exactMatch.toFixed(6)),
+          tokenOverlap: Number(signals.tokenOverlap.toFixed(6)),
+          cosine: Number(signals.cosine.toFixed(6)),
+          recency: Number(signals.recency.toFixed(6)),
+          severity: Number(signals.severityScore.toFixed(6)),
+        },
+      };
+    })
+    .filter((result) => result.score >= normalizedMinScore)
+    .sort((left, right) => {
+      if (right.score !== left.score) {
+        return right.score - left.score;
+      }
+      return left.documentId.localeCompare(right.documentId);
+    })
+    .slice(0, normalizedLimit);
+
+  return {
+    query: queryText,
+    limit: normalizedLimit,
+    minScore: normalizedMinScore,
+    queryTokenCount: queryTokens.length,
+    results,
+  };
+}
+
+export async function queryHybridRetriever({
+  query = "",
+  documents = [],
+  limit = 12,
+  provider = "local",
+  apiEndpoint = "",
+  apiKey = "",
+  fetchImpl = globalThis.fetch,
+} = {}) {
+  const normalizedProvider = normalizeString(provider).toLowerCase() || "local";
+  const localIndex = buildLocalHybridIndex(documents);
+  const localQuery = queryLocalHybridIndex(localIndex, {
+    query,
+    limit,
+  });
+
+  const canUseApi =
+    (normalizedProvider === "api" || normalizedProvider === "auto") &&
+    normalizeString(apiEndpoint) &&
+    typeof fetchImpl === "function";
+  if (!canUseApi) {
+    return {
+      providerRequested: normalizedProvider,
+      providerUsed: "local",
+      apiFallback: false,
+      apiError: "",
+      indexSummary: {
+        documentCount: localIndex.documentCount,
+        vocabularySize: localIndex.vocabularySize,
+      },
+      results: localQuery.results,
+    };
+  }
+
+  try {
+    const response = await fetchImpl(String(apiEndpoint), {
+      method: "POST",
+      headers: {
+        "content-type": "application/json",
+        ...(normalizeString(apiKey) ? { authorization: `Bearer ${apiKey}` } : {}),
+      },
+      body: JSON.stringify({
+        query: normalizeString(query),
+        limit: Math.max(1, Math.floor(normalizeNumber(limit, 12))),
+        documents,
+      }),
+    });
+    if (!response.ok) {
+      throw new Error(`Memory API request failed (${response.status}).`);
+    }
+    const payload = await response.json();
+    const apiResults = Array.isArray(payload?.results) ? payload.results : [];
+    if (apiResults.length === 0) {
+      throw new Error("Memory API response missing results.");
+    }
+    return {
+      providerRequested: normalizedProvider,
+      providerUsed: "api",
+      apiFallback: false,
+      apiError: "",
+      indexSummary: {
+        documentCount: localIndex.documentCount,
+        vocabularySize: localIndex.vocabularySize,
+      },
+      results: apiResults.slice(0, Math.max(1, Math.floor(normalizeNumber(limit, 12)))),
+    };
+  } catch (error) {
+    return {
+      providerRequested: normalizedProvider,
+      providerUsed: "local",
+      apiFallback: true,
+      apiError: normalizeString(error?.message || error),
+      indexSummary: {
+        documentCount: localIndex.documentCount,
+        vocabularySize: localIndex.vocabularySize,
+      },
+      results: localQuery.results,
+    };
+  }
+}