akm-cli 0.7.4 → 0.8.0-rc.10

package/dist/indexer/indexer.js

@@ -1,14 +1,18 @@
+// This Source Code Form is subject to the terms of the Mozilla Public
+// License, v. 2.0. If a copy of the MPL was not distributed with this
+// file, You can obtain one at https://mozilla.org/MPL/2.0/.
 import fs from "node:fs";
 import path from "node:path";
+import { SCRIPT_EXTENSIONS } from "../core/asset-spec";
 import { isHttpUrl, resolveStashDir, toErrorMessage } from "../core/common";
+import { concurrentMap } from "../core/concurrent";
 import { getDbPath } from "../core/paths";
 import { isVerbose, warn, warnVerbose } from "../core/warn";
 import { resolveIndexPassLLM } from "../llm/index-passes";
 import { takeWorkflowDocument } from "../workflows/document-cache";
-import { closeDatabase, deleteEntriesByDir, deleteEntriesByStashDir, deleteIndexDirStatesByStashDir, getEmbeddingCount, getEntriesByDir, getEntryCount, getIndexDirState, getMeta, isVecAvailable, openDatabase, openExistingDatabase, rebuildFts, setMeta, upsertEmbedding, upsertEntry, upsertIndexDirState, upsertUtilityScore, warnIfVecMissing, } from "./db";
-import { runGraphExtractionPass } from "./graph-extraction";
-import { runMemoryInferencePass } from "./memory-inference";
-import { applyCuratedFrontmatter, applyWikiFrontmatter, generateMetadataFlat, isWorkflowSkipWarning, loadStashFile, shouldIndexStashFile, } from "./metadata";
+import { clearStaleCacheEntries, closeDatabase, deleteEntriesByDir, deleteEntriesByIds, deleteEntriesByStashDir, deleteIndexDirStatesByStashDir, getAllEntriesForEmbedding, getEmbeddingCount, getEntriesByDir, getEntryCount, getIndexDirState, getMeta, isVecAvailable, openDatabase, openExistingDatabase, rebuildFts, relinkUsageEvents, setMeta, upsertEmbedding, upsertEntry, upsertIndexDirState, upsertUtilityScore, upsertWorkflowDocument, warnIfVecMissing, } from "./db";
+import { deleteStoredGraph } from "./graph-db";
+import { applyCuratedFrontmatter, applyWikiFrontmatter, generateMetadataFlat, isEnrichmentComplete, isWorkflowSkipWarning, loadStashFile, shouldIndexStashFile, } from "./metadata";
 import { buildSearchText } from "./search-fields";
 import { classifySemanticFailure, clearSemanticStatus, deriveSemanticProviderFingerprint, writeSemanticStatus, } from "./semantic-status";
 import { ensureUsageEventsSchema, purgeOldUsageEvents } from "./usage-events";
@@ -18,19 +22,222 @@ function throwIfAborted(signal) {
         throw signal.reason instanceof Error ? signal.reason : new Error("index interrupted");
     }
 }
+function getDefaultLlmConcurrency(llmConfig) {
+    if (typeof llmConfig?.concurrency === "number")
+        return llmConfig.concurrency;
+    if (!llmConfig?.endpoint)
+        return 1;
+    try {
+        const url = new URL(llmConfig.endpoint);
+        const host = url.hostname.toLowerCase();
+        if (host === "localhost" || host === "127.0.0.1" || host === "::1" || host.endsWith(".localhost"))
+            return 1;
+    }
+    catch {
+        return 1;
+    }
+    return 4;
+}
+// ── Phase functions ──────────────────────────────────────────────────────────
+/**
+ * Source cache phase: ensure git stash caches are up to date and purge orphaned
+ * entries from removed sources (incremental only).
+ */
+async function runSourceCachePhase(ctx) {
+    const { db, config, sourceDirs, isIncremental, full } = ctx;
+    if (isIncremental && !full) {
+        // Purge entries from stash dirs that have been removed since the last run
+        // (e.g. after `akm remove`) so orphaned entries don't linger.
+        const prevStashDirsJson = getMeta(db, "stashDirs");
+        if (prevStashDirsJson) {
+            let prevStashDirs = [];
+            try {
+                const parsed = JSON.parse(prevStashDirsJson);
+                if (Array.isArray(parsed)) {
+                    prevStashDirs = parsed.filter((d) => typeof d === "string");
+                }
+                else {
+                    warn("index_meta stashDirs value is not an array — treating as empty");
+                }
+            }
+            catch {
+                warn("index_meta stashDirs value is corrupt JSON — treating as empty");
+            }
+            const currentSet = new Set(sourceDirs);
+            for (const dir of prevStashDirs) {
+                if (!currentSet.has(dir)) {
+                    ctx.hadRemovedSources = true;
+                    deleteEntriesByStashDir(db, dir);
+                    deleteIndexDirStatesByStashDir(db, dir);
+                    deleteStoredGraph(db, dir);
+                }
+            }
+        }
+    }
+    // Source caches are hydrated before akmIndex() calls this phase; nothing
+    // further to do here. The flag is exposed on ctx for runWalkPhase().
+    void config;
+}
+/**
+ * Walk phase: scan the filesystem, generate metadata, and persist entries to
+ * the database. Also kicks off LLM enrichment for directories that need it.
+ *
+ * Writes `ctx.scannedDirs`, `ctx.skippedDirs`, `ctx.generatedCount`,
+ * `ctx.walkWarnings`, and `ctx.dirsNeedingLlm` for downstream phases.
+ */
+async function runWalkPhase(ctx) {
+    const { db, sources, isIncremental, builtAtMs, hadRemovedSources, full, reEnrich, signal, onProgress, config } = ctx;
+    throwIfAborted(signal);
+    ctx.timing.tWalkStart = Date.now();
+    const doFullDelete = full || !isIncremental;
+    const { scannedDirs, skippedDirs, generatedCount, dirsNeedingLlm, warnings } = await indexEntries(db, sources, isIncremental, builtAtMs, hadRemovedSources, doFullDelete, onProgress);
+    ctx.scannedDirs = scannedDirs;
+    ctx.skippedDirs = skippedDirs;
+    ctx.generatedCount = generatedCount;
+    ctx.walkWarnings = warnings;
+    ctx.dirsNeedingLlm = dirsNeedingLlm;
+    onProgress({
+        phase: "scan",
+        message: `Scanned ${scannedDirs} ${scannedDirs === 1 ? "directory" : "directories"} and skipped ${skippedDirs}.`,
+    });
+    // Workflow validation noise gate (issue #273): suppress per-spec stderr lines
+    // at default verbosity and emit a single summary instead.
+    // In verbose mode the per-spec lines are already printed by
+    // buildMetadataSkipWarning at generation time — no second pass needed here.
+    if (!isVerbose()) {
+        const workflowSkipWarnings = warnings.filter(isWorkflowSkipWarning);
+        const skippedWorkflowCount = workflowSkipWarnings.length;
+        if (skippedWorkflowCount > 0) {
+            const noun = skippedWorkflowCount === 1 ? "workflow spec" : "workflow specs";
+            warn(`${skippedWorkflowCount} ${noun} skipped due to validation errors; ` +
+                "rerun with --verbose (or AKM_VERBOSE=1) to see details.");
+        }
+    }
+    ctx.timing.tWalkEnd = Date.now();
+    throwIfAborted(signal);
+    // LLM enrichment for directories that need it
+    await enhanceDirsWithLlm(db, config, dirsNeedingLlm, onProgress, signal, true, reEnrich);
+    onProgress({
+        phase: "llm",
+        message: resolveIndexPassLLM("enrichment", config)
+            ? `LLM enhancement reviewed ${dirsNeedingLlm.length} ${dirsNeedingLlm.length === 1 ? "directory" : "directories"}.`
+            : "LLM enhancement disabled.",
+    });
+    ctx.timing.tLlmEnd = Date.now();
+}
+/**
+ * Embedding phase: generate and store vector embeddings for all unembedded
+ * entries. Writes `ctx.embeddingResult` for the finalize phase.
+ */
+async function runEmbeddingPhase(ctx) {
+    const { db, config, signal, onProgress } = ctx;
+    throwIfAborted(signal);
+    ctx.embeddingResult = await generateEmbeddingsForDb(db, config, onProgress);
+    ctx.timing.tEmbedEnd = Date.now();
+}
+/**
+ * Finalize phase: rebuild FTS, re-link usage events, recompute utility scores,
+ * regenerate wiki indexes, update index metadata, and emit the verify event.
+ */
+async function runFinalizePhase(ctx) {
+    const { db, config, sources, sourceDirs, isIncremental, stashDir, signal, onProgress } = ctx;
+    // Rebuild FTS after all inserts. Use incremental mode when this whole
+    // index run is incremental — only entries touched by `upsertEntry`
+    // since the last rebuild are re-indexed.
+    rebuildFts(db, { incremental: isIncremental });
+    onProgress({
+        phase: "fts",
+        message: isIncremental ? "Rebuilt full-text search index (dirty rows only)." : "Rebuilt full-text search index.",
+    });
+    ctx.timing.tFtsEnd = Date.now();
+    // Re-link detached usage_events and recompute utility scores.
+    relinkUsageEvents(db);
+    recomputeUtilityScores(db);
+    // Purge LLM cache entries for assets that no longer exist in the index.
+    try {
+        clearStaleCacheEntries(db);
+    }
+    catch {
+        /* ignore */
+    }
+    // Regenerate each wiki's index.md from its pages' frontmatter. Best-effort.
+    try {
+        const { regenerateAllWikiIndexes } = await import("../wiki/wiki.js");
+        regenerateAllWikiIndexes(stashDir);
+    }
+    catch {
+        /* best-effort */
+    }
+    throwIfAborted(signal);
+    // Update index metadata
+    const embeddingResult = ctx.embeddingResult ?? { success: false };
+    setMeta(db, "builtAt", new Date().toISOString());
+    setMeta(db, "stashDir", stashDir);
+    setMeta(db, "stashDirs", JSON.stringify(sourceDirs));
+    setMeta(db, "hasEmbeddings", embeddingResult.success ? "1" : "0");
+    warnIfVecMissing(db);
+    const totalEntries = getEntryCount(db);
+    const verification = verifyIndexState(db, config, totalEntries, embeddingResult);
+    if (config.semanticSearchMode === "off") {
+        clearSemanticStatus();
+    }
+    else {
+        writeSemanticStatus({
+            status: verification.semanticStatus === "disabled" ? "pending" : verification.semanticStatus,
+            ...(embeddingResult.reason ? { reason: embeddingResult.reason } : {}),
+            ...(embeddingResult.message ? { message: embeddingResult.message } : {}),
+            providerFingerprint: deriveSemanticProviderFingerprint(config.embedding),
+            lastCheckedAt: new Date().toISOString(),
+            entryCount: verification.entryCount,
+            embeddingCount: verification.embeddingCount,
+        });
+    }
+    onProgress({ phase: "verify", message: verification.message });
+    // Store verification result and totalEntries on ctx for the caller to use
+    ctx._verification = verification;
+    ctx._totalEntries = totalEntries;
+    // suppress unused warning — sources was previously used inline
+    void sources;
+}
+// ── Clean pass ───────────────────────────────────────────────────────────────
+/**
+ * Post-index clean pass: scan the `entries` table for rows whose source file
+ * no longer exists on disk and remove them (unless `dryRun` is true).
+ *
+ * Only rows with a non-empty `file_path` are checked — remote/virtual entries
+ * that have no local path are always skipped.
+ */
+function runCleanPass(db, dryRun) {
+    const allEntries = db.prepare("SELECT id, entry_key AS ref, file_path AS path FROM entries").all();
+    // Only check entries that have a non-empty local path (skip remote/virtual).
+    const localEntries = allEntries.filter((e) => typeof e.path === "string" && e.path.trim() !== "");
+    const missing = localEntries.filter((e) => !fs.existsSync(e.path));
+    if (!dryRun && missing.length > 0) {
+        deleteEntriesByIds(db, missing.map((e) => e.id));
+    }
+    return {
+        checked: localEntries.length,
+        removed: dryRun ? 0 : missing.length,
+        removedRefs: missing.map((e) => e.ref),
+        dryRun,
+    };
+}
 // ── Indexer ──────────────────────────────────────────────────────────────────
 export async function akmIndex(options) {
     const stashDir = options?.stashDir || resolveStashDir();
     const onProgress = options?.onProgress ?? (() => { });
     const signal = options?.signal;
-    const enrich = options?.enrich === true;
+    const reEnrich = options?.reEnrich === true;
+    const full = options?.full === true;
+    const clean = options?.clean === true;
+    const dryRun = options?.dryRun === true;
     // Load config and resolve all stash sources
     const { loadConfig } = await import("../core/config.js");
     const config = loadConfig();
     // Ensure git stash caches are extracted before resolving stash dirs,
     // so their content directories exist on disk for the walker to discover.
     const { ensureSourceCaches, resolveSourceEntries } = await import("./search-source.js");
-    await ensureSourceCaches(config, { force: options?.full === true });
+    await ensureSourceCaches(config, { force: full });
     const allSourceEntries = resolveSourceEntries(stashDir, config);
     const allSourceDirs = allSourceEntries.map((s) => s.path);
     const t0 = Date.now();
@@ -39,11 +246,41 @@ export async function akmIndex(options) {
     const embeddingDim = config.embedding?.dimension;
     const db = openDatabase(dbPath, embeddingDim ? { embeddingDim } : undefined);
     try {
-        // Check if we should do incremental
+        // Determine incremental vs full mode
         const prevStashDir = getMeta(db, "stashDir");
         const prevBuiltAt = getMeta(db, "builtAt");
-        const isIncremental = !options?.full && prevStashDir === stashDir && !!prevBuiltAt;
+        const isIncremental = !full && prevStashDir === stashDir && !!prevBuiltAt;
         const builtAtMs = isIncremental && prevBuiltAt ? new Date(prevBuiltAt).getTime() : 0;
+        // Assemble the run context
+        const ctx = {
+            db,
+            config,
+            sources: allSourceEntries,
+            sourceDirs: allSourceDirs,
+            full,
+            reEnrich,
+            stashDir,
+            onProgress,
+            signal,
+            timing: {
+                t0,
+                tWalkStart: t0,
+                tWalkEnd: t0,
+                tLlmEnd: t0,
+                tFtsEnd: t0,
+                tEmbedEnd: t0,
+            },
+            isIncremental,
+            builtAtMs,
+            hadRemovedSources: false,
+            scannedDirs: 0,
+            skippedDirs: 0,
+            generatedCount: 0,
+            walkWarnings: [],
+            dirsNeedingLlm: [],
+            embeddingResult: null,
+            graphExtractionResult: null,
+        };
         onProgress({
             phase: "summary",
             message: buildIndexSummaryMessage({
@@ -51,219 +288,44 @@ export async function akmIndex(options) {
                 sourcesCount: allSourceDirs.length,
                 semanticSearchMode: config.semanticSearchMode,
                 embeddingProvider: getEmbeddingProvider(config.embedding),
-                llmEnabled: enrich && !!resolveIndexPassLLM("enrichment", config),
+                llmEnabled: !!resolveIndexPassLLM("enrichment", config),
                 vecAvailable: isVecAvailable(db),
             }),
         });
-        let hadRemovedSources = false;
-        if (options?.full || !isIncremental) {
-            // The delete is now merged into the insert transaction inside
-            // indexEntries() so that a reader never sees an empty database between
-            // the wipe and the re-inserts.  The doFullDelete flag signals this path.
+        // ── Phase sequence ───────────────────────────────────────────────────────
+        await runSourceCachePhase(ctx);
+        await runWalkPhase(ctx);
+        await runEmbeddingPhase(ctx);
+        await runFinalizePhase(ctx);
+        // ────────────────────────────────────────────────────────────────────────
+        const { _verification: verification, _totalEntries: totalEntries } = ctx;
+        const { timing } = ctx;
+        // ── Clean pass ───────────────────────────────────────────────────────────
+        // After the normal index completes, remove entries whose source files no
+        // longer exist on disk. Remote entries (empty file_path) are skipped.
+        let cleanResult;
+        if (clean) {
+            cleanResult = runCleanPass(db, dryRun);
         }
-        else {
-            // Incremental: purge entries from stash dirs that have been removed
-            // (e.g. after `akm remove`) so orphaned entries don't linger.
-            const prevStashDirsJson = getMeta(db, "stashDirs");
-            if (prevStashDirsJson) {
-                let prevStashDirs = [];
-                try {
-                    const parsed = JSON.parse(prevStashDirsJson);
-                    if (Array.isArray(parsed)) {
-                        prevStashDirs = parsed.filter((d) => typeof d === "string");
-                    }
-                    else {
-                        warn("index_meta stashDirs value is not an array — treating as empty");
-                    }
-                }
-                catch {
-                    warn("index_meta stashDirs value is corrupt JSON — treating as empty");
-                }
-                const currentSet = new Set(allSourceDirs);
-                for (const dir of prevStashDirs) {
-                    if (!currentSet.has(dir)) {
-                        hadRemovedSources = true;
-                        deleteEntriesByStashDir(db, dir);
-                        deleteIndexDirStatesByStashDir(db, dir);
-                    }
-                }
-            }
-        }
-        throwIfAborted(signal);
-        // Memory inference pass (#201). Runs before the walk so any derived-memory
-        // children that get written are picked up by the walker in this same run
-        // and don't have to wait for the next `akm index`. Gated entirely by
-        // `resolveIndexPassLLM("memory", config)` — when the user has no
-        // `akm.llm` block or has set `index.memory.llm = false`, this is a no-op
-        // and existing inferred children are left in place.
-        if (enrich) {
-            try {
-                const inferenceResult = await runMemoryInferencePass(config, allSourceEntries, signal);
-                if (inferenceResult.writtenFacts > 0) {
-                    onProgress({
-                        phase: "llm",
-                        message: `Memory inference wrote ${inferenceResult.writtenFacts} derived memor${inferenceResult.writtenFacts === 1 ? "y" : "ies"} from ${inferenceResult.splitParents} parent memor${inferenceResult.splitParents === 1 ? "y" : "ies"}.`,
-                    });
-                }
-            }
-            catch (err) {
-                warn(`Memory inference pass aborted: ${err instanceof Error ? err.message : String(err)}`);
-            }
-        }
-        else {
-            onProgress({
-                phase: "llm",
-                message: "LLM passes disabled; rerun with --enrich to enable inference and enrichment.",
-            });
-        }
-        // Graph extraction pass (#207). Runs after memory inference so any
-        // atomic-fact children that just got written are visible to the graph
-        // walk. Persists `<stashRoot>/.akm/graph.json` — an indexer artifact,
-        // NOT a user-visible asset, so it is not routed through
-        // writeAssetToSource. The artifact feeds the existing FTS5+boosts
-        // pipeline as a single boost component (see graph-boost.ts); there is
-        // no parallel scoring track. Disabled when either gate (the locked
-        // `llm.features.graph_extraction` feature flag or the per-pass
-        // `index.graph.llm` toggle) is off; the existing graph file is
-        // preserved on disk in that case.
-        if (enrich) {
-            try {
-                const graphResult = await runGraphExtractionPass(config, allSourceEntries, signal);
-                if (graphResult.written) {
-                    onProgress({
-                        phase: "llm",
-                        message: `Graph extraction wrote ${graphResult.totalEntities} entit${graphResult.totalEntities === 1 ? "y" : "ies"} and ${graphResult.totalRelations} relation${graphResult.totalRelations === 1 ? "" : "s"} from ${graphResult.extracted} file${graphResult.extracted === 1 ? "" : "s"}.`,
-                    });
-                }
-            }
-            catch (err) {
-                warn(`Graph extraction pass aborted: ${err instanceof Error ? err.message : String(err)}`);
-            }
-        }
-        throwIfAborted(signal);
-        const tWalkStart = Date.now();
-        // Walk stash dirs and index entries.
-        // doFullDelete=true merges the wipe into the same transaction as the
-        // inserts so readers never see an empty database mid-rebuild.
-        const doFullDelete = options?.full || !isIncremental;
-        const { scannedDirs, skippedDirs, generatedCount, dirsNeedingLlm, warnings } = await indexEntries(db, allSourceEntries, isIncremental, builtAtMs, hadRemovedSources, doFullDelete, onProgress);
-        onProgress({
-            phase: "scan",
-            message: `Scanned ${scannedDirs} ${scannedDirs === 1 ? "directory" : "directories"} and skipped ${skippedDirs}.`,
-        });
-        // Workflow validation noise gate (issue #273): per-spec stderr lines from
-        // `buildMetadataSkipWarning` are suppressed at default verbosity in
-        // `metadata.ts`. Replace them with a single summary line so operators
-        // running a cold-start search against a fresh registry-cloned source
-        // don't get the impression akm is broken. Verbose mode keeps the
-        // per-spec output instead of (not in addition to) the summary.
-        if (!isVerbose()) {
-            const skippedWorkflowCount = warnings.filter(isWorkflowSkipWarning).length;
-            if (skippedWorkflowCount > 0) {
-                const noun = skippedWorkflowCount === 1 ? "workflow spec" : "workflow specs";
-                warn(`${skippedWorkflowCount} ${noun} skipped due to validation errors; ` +
-                    "rerun with --verbose (or AKM_VERBOSE=1) to see details.");
-            }
-        }
-        const tWalkEnd = Date.now();
-        throwIfAborted(signal);
-        // Enhance entries with LLM if configured
-        await enhanceDirsWithLlm(db, config, dirsNeedingLlm, signal, enrich);
-        onProgress({
-            phase: "llm",
-            message: enrich && resolveIndexPassLLM("enrichment", config)
-                ? `LLM enhancement reviewed ${dirsNeedingLlm.length} ${dirsNeedingLlm.length === 1 ? "directory" : "directories"}.`
-                : "LLM enhancement disabled.",
-        });
-        const tLlmEnd = Date.now();
-        throwIfAborted(signal);
-        // Rebuild FTS after all inserts. Use incremental mode when this whole
-        // index run is incremental — only entries touched by `upsertEntry`
-        // since the last rebuild are re-indexed, instead of re-scanning every
-        // row on every `akm index` invocation.
-        rebuildFts(db, { incremental: isIncremental });
-        onProgress({
-            phase: "fts",
-            message: isIncremental ? "Rebuilt full-text search index (dirty rows only)." : "Rebuilt full-text search index.",
-        });
-        const tFtsEnd = Date.now();
-        // Re-link detached usage_events to their new entry_ids via entry_ref.
-        // entry_ref is "type:name" (e.g., "skill:code-review"), entry_key is "stashDir:type:name".
-        // Use substr to extract the "type:name" suffix from entry_key for exact comparison
-        // (avoids LIKE which would require escaping % and _ in user-facing names).
-        try {
-            db.exec(`
-        UPDATE usage_events SET entry_id = (
-          SELECT e.id FROM entries e
-          WHERE substr(e.entry_key, length(e.entry_key) - length(usage_events.entry_ref)) = ':' || usage_events.entry_ref
-          LIMIT 1
-        )
-        WHERE entry_id IS NULL AND entry_ref IS NOT NULL
-      `);
-        }
-        catch {
-            /* ignore if table doesn't exist yet */
-        }
-        // Recompute utility scores from usage_events after FTS rebuild
-        recomputeUtilityScores(db);
-        // Regenerate each wiki's index.md from its pages' frontmatter. Best-effort
-        // — errors are caught inside regenerateAllWikiIndexes and never block the
-        // index run. The primary stash is the only target: additional sources
-        // are read-only caches, and regenerating their indexes would mutate
-        // cache content.
-        try {
-            const { regenerateAllWikiIndexes } = await import("../wiki/wiki.js");
-            regenerateAllWikiIndexes(stashDir);
-        }
-        catch {
-            /* best-effort */
-        }
-        throwIfAborted(signal);
-        // Generate embeddings if semantic search is enabled
-        const embeddingResult = await generateEmbeddingsForDb(db, config, onProgress);
-        const tEmbedEnd = Date.now();
-        // Update metadata
-        setMeta(db, "builtAt", new Date().toISOString());
-        setMeta(db, "stashDir", stashDir);
-        setMeta(db, "stashDirs", JSON.stringify(allSourceDirs));
-        setMeta(db, "hasEmbeddings", embeddingResult.success ? "1" : "0");
-        const totalEntries = getEntryCount(db);
-        // Warn on every index run if using JS fallback with many entries
-        warnIfVecMissing(db);
-        const tEnd = Date.now();
-        const verification = verifyIndexState(db, config, totalEntries, embeddingResult);
-        if (config.semanticSearchMode === "off") {
-            clearSemanticStatus();
-        }
-        else {
-            writeSemanticStatus({
-                status: verification.semanticStatus === "disabled" ? "pending" : verification.semanticStatus,
-                ...(embeddingResult.reason ? { reason: embeddingResult.reason } : {}),
-                ...(embeddingResult.message ? { message: embeddingResult.message } : {}),
-                providerFingerprint: deriveSemanticProviderFingerprint(config.embedding),
-                lastCheckedAt: new Date().toISOString(),
-                entryCount: verification.entryCount,
-                embeddingCount: verification.embeddingCount,
-            });
-        }
-        onProgress({ phase: "verify", message: verification.message });
+        // ────────────────────────────────────────────────────────────────────────
         return {
             stashDir,
             totalEntries,
-            generatedMetadata: generatedCount,
+            generatedMetadata: ctx.generatedCount,
             indexPath: dbPath,
             mode: isIncremental ? "incremental" : "full",
-            directoriesScanned: scannedDirs,
-            directoriesSkipped: skippedDirs,
-            ...(warnings.length > 0 ? { warnings } : {}),
+            directoriesScanned: ctx.scannedDirs,
+            directoriesSkipped: ctx.skippedDirs,
+            ...(ctx.walkWarnings.length > 0 ? { warnings: ctx.walkWarnings } : {}),
             verification,
             timing: {
-                totalMs: tEnd - t0,
-                walkMs: tWalkEnd - tWalkStart,
-                llmMs: tLlmEnd - tWalkEnd,
-                embedMs: tEmbedEnd - tFtsEnd,
-                ftsMs: tFtsEnd - tLlmEnd,
+                totalMs: Date.now() - timing.t0,
+                walkMs: timing.tWalkEnd - timing.tWalkStart,
+                llmMs: timing.tLlmEnd - timing.tWalkEnd,
+                embedMs: timing.tEmbedEnd - timing.tFtsEnd,
+                ftsMs: timing.tFtsEnd - timing.tLlmEnd,
             },
+            ...(cleanResult !== undefined ? { clean: cleanResult } : {}),
         };
     }
     finally {
@@ -500,8 +562,10 @@ async function indexEntries(db, allSourceEntries, isIncremental, builtAtMs, hadR
             if (stash) {
                 for (const entry of stash.entries) {
                     const entryPath = entry.filename ? path.join(dirPath, entry.filename) : null;
-                    if (!entryPath)
-                        continue; // skip unresolvable entries
+                    if (!entryPath) {
+                        warn(`Skipping entry with no resolvable path in ${dirPath}`);
+                        continue;
+                    }
                     if (!shouldIndexStashFile(currentStashDir, entryPath))
                         continue;
                     // Skip if a higher-priority stash root already indexed this asset
@@ -523,7 +587,9 @@ async function indexEntries(db, allSourceEntries, isIncremental, builtAtMs, hadR
                         }
                     }
                 }
-                // Collect dirs needing LLM enhancement during the first walk
+                // Collect dirs needing LLM enhancement during the first walk.
+                // Only dirs with "generated" entries need enrichment (unless reEnrich
+                // forces re-processing of already-enriched entries).
                 if (stash.entries.some((e) => e.quality === "generated")) {
                     dirsNeedingLlm.push({ dirPath, files, currentStashDir, stash });
                 }
@@ -541,7 +607,20 @@ async function indexEntries(db, allSourceEntries, isIncremental, builtAtMs, hadR
                 reason: persistedReason,
             });
             if (persistedRows === 0) {
-                warnVerbose(`[index] zero-row ${dirPath}: ${persistedReason}`);
+                // Warn only when the dir had files that *could* produce entries (.md or
+                // known script extensions). Dirs with only non-indexable types (.json,
+                // .yaml, .conf, .env, .gitkeep) or deduped-only rows are expected and
+                // not actionable at normal log level.
+                const hasIndexableExtension = files.some((f) => {
+                    const ext = path.extname(f).toLowerCase();
+                    return ext === ".md" || SCRIPT_EXTENSIONS.has(ext);
+                });
+                if (persistedReason !== "deduped-zero-row" && hasIndexableExtension) {
+                    warn(`[index] zero-row ${dirPath}: ${persistedReason}`);
+                }
+                else {
+                    warnVerbose(`[index] zero-row ${dirPath}: ${persistedReason}`);
+                }
             }
         }
     });
@@ -640,9 +719,7 @@ function inferZeroRowReason(stash, priorReason, warnings, dirPath, dedupedRows)
         return "empty-generated-set";
     return `zero-row:${priorReason?.kind ?? "unknown"}`;
 }
-async function enhanceDirsWithLlm(db, config, dirsNeedingLlm, signal, enrich = false) {
-    if (!enrich)
-        return;
+async function enhanceDirsWithLlm(db, config, dirsNeedingLlm, onProgress, signal, _enrich = false, reEnrich = false) {
     // Resolve per-pass LLM config via the unified shim. Returns undefined when
     // either no `akm.llm` is configured or the user opted this pass out via
     // `index.enrichment.llm = false`. (#208)
@@ -653,24 +730,142 @@ async function enhanceDirsWithLlm(db, config, dirsNeedingLlm, signal, enrich = f
     // as a single visible warning instead of silently degrading every entry
     // and leaving the user wondering why nothing got enhanced.
     const summary = { attempted: 0, succeeded: 0, failureSamples: [] };
-    for (const { dirPath, files, currentStashDir, stash: originalStash } of dirsNeedingLlm) {
-        throwIfAborted(signal);
-        // Only enhance generated entries; user-provided overrides should not be overwritten
-        const generatedEntries = originalStash.entries.filter((e) => e.quality === "generated");
-        if (generatedEntries.length === 0)
-            continue;
-        const generatedStash = { entries: generatedEntries };
-        const enhanced = await enhanceStashWithLlm(llmConfig, generatedStash, files, summary, signal);
-        // Re-upsert the enhanced entries in a single transaction so a crash
-        // cannot leave half the entries updated and the rest stale.
-        db.transaction(() => {
-            for (const entry of enhanced.entries) {
-                const entryPath = entry.filename ? path.join(dirPath, entry.filename) : files[0] || dirPath;
-                const entryKey = `${currentStashDir}:${entry.type}:${entry.name}`;
-                const searchText = buildSearchText(entry);
-                upsertEntry(db, entryKey, dirPath, entryPath, currentStashDir, attachFileSize(entry, entryPath), searchText);
-            }
-        })();
+    let completedDirs = 0;
+    let completedEntries = 0;
+    const totalDirs = dirsNeedingLlm.length;
+    const totalEntries = dirsNeedingLlm.reduce((sum, { stash }) => {
+        const entriesToEnhance = stash.entries.filter((e) => {
+            if (e.quality !== "generated" && !(reEnrich && e.quality === "enriched"))
+                return false;
+            if (!reEnrich && isEnrichmentComplete(e))
+                return false;
+            return true;
+        });
+        return sum + entriesToEnhance.length;
+    }, 0);
+    // P3 — wall-clock budget for the enrichment pass. Defaults to llm.timeoutMs
+    // (or 10 minutes if not set). Users can extend this via llm.timeoutMs in
+    // config — no separate knob needed.
+    const budgetMs = (llmConfig.timeoutMs ?? 10 * 60 * 1000) * Math.max(totalEntries, 1);
+    const enrichDeadline = AbortSignal.timeout(budgetMs);
+    let deadlineHit = false;
+    const enrichSignal = (() => {
+        if (!signal)
+            return enrichDeadline;
+        // Combine: abort when either fires.
+        const controller = new AbortController();
+        const onAbort = () => controller.abort();
+        signal.addEventListener("abort", onAbort, { once: true });
+        enrichDeadline.addEventListener("abort", () => {
+            deadlineHit = true;
+            controller.abort();
+        }, { once: true });
+        return controller.signal;
+    })();
+    if (totalEntries > 0) {
+        onProgress?.({
+            phase: "llm",
+            message: `LLM enhancement starting for ${totalEntries} entr${totalEntries === 1 ? "y" : "ies"} ` +
+                `across ${totalDirs} director${totalDirs === 1 ? "y" : "ies"} (concurrency ${getDefaultLlmConcurrency(llmConfig)}).`,
+            processed: 0,
+            total: totalEntries,
+        });
+    }
+    let currentDirLabel;
+    let lastProgressAt = Date.now();
+    let heartbeatTimer;
+    if (totalEntries > 0 && onProgress) {
+        heartbeatTimer = setInterval(() => {
+            if (Date.now() - lastProgressAt < 15000)
+                return;
+            onProgress({
+                phase: "llm",
+                message: `Still enriching ${completedEntries}/${totalEntries} entr${totalEntries === 1 ? "y" : "ies"}` +
+                    (currentDirLabel ? `; waiting on ${currentDirLabel}` : "") +
+                    ".",
+                processed: completedEntries,
+                total: totalEntries,
+            });
+            lastProgressAt = Date.now();
+        }, 15000);
+    }
+    try {
+        await concurrentMap(dirsNeedingLlm, async ({ dirPath, files, currentStashDir, stash: originalStash }) => {
+            if (enrichSignal.aborted)
+                return undefined;
+            // Only enhance generated entries (or all when reEnrich=true);
+            // user-provided overrides should not be overwritten.
+            // Skip entries that are already fully enriched (description + tags + searchHints)
+            // unless the caller explicitly requests re-enrichment via reEnrich=true.
+            const entriesToEnhance = originalStash.entries.filter((e) => {
+                if (e.quality !== "generated" && !(reEnrich && e.quality === "enriched"))
+                    return false;
+                if (!reEnrich && isEnrichmentComplete(e)) {
+                    warnVerbose(`[akm] skipping LLM enrichment for "${e.name}" — entry already complete`);
+                    return false;
+                }
+                return true;
+            });
+            if (entriesToEnhance.length === 0)
+                return undefined;
+            currentDirLabel = path.relative(currentStashDir, dirPath) || ".";
+            onProgress?.({
+                phase: "llm",
+                message: `Enhancing ${currentDirLabel} ` +
+                    `(${entriesToEnhance.length} entr${entriesToEnhance.length === 1 ? "y" : "ies"}).`,
+                processed: completedEntries,
+                total: totalEntries,
+            });
+            lastProgressAt = Date.now();
+            const targetStash = { entries: entriesToEnhance };
+            const entryKeys = entriesToEnhance.map((e) => `${currentStashDir}:${e.type}:${e.name}`);
+            const enhanced = await enhanceStashWithLlm(llmConfig, targetStash, files, summary, enrichSignal, db, entryKeys, reEnrich, config, (event) => {
+                completedEntries++;
+                lastProgressAt = Date.now();
+                onProgress?.({
+                    phase: "llm",
+                    message: `Enhanced ${completedEntries}/${totalEntries} entr${totalEntries === 1 ? "y" : "ies"}; ` +
+                        `${completedDirs}/${totalDirs} director${totalDirs === 1 ? "y" : "ies"} complete` +
+                        (event.entryName ? `; current ${event.entryName}` : "") +
+                        (currentDirLabel ? ` in ${currentDirLabel}` : "") +
+                        (event.outcome === "cache-hit" ? " (cache hit)" : ""),
+                    processed: completedEntries,
+                    total: totalEntries,
+                });
+            });
+            // Re-upsert the enhanced entries in a single transaction so a crash
+            // cannot leave half the entries updated and the rest stale.
+            db.transaction(() => {
+                for (const entry of enhanced.entries) {
+                    const entryPath = entry.filename ? path.join(dirPath, entry.filename) : files[0] || dirPath;
+                    const entryKey = `${currentStashDir}:${entry.type}:${entry.name}`;
+                    const searchText = buildSearchText(entry);
+                    upsertEntry(db, entryKey, dirPath, entryPath, currentStashDir, attachFileSize(entry, entryPath), searchText);
+                }
+            })();
+            completedDirs++;
+            lastProgressAt = Date.now();
+            onProgress?.({
+                phase: "llm",
+                message: `Completed ${completedDirs}/${totalDirs} director${totalDirs === 1 ? "y" : "ies"}; ` +
+                    `${completedEntries}/${totalEntries} entr${totalEntries === 1 ? "y" : "ies"} processed.`,
+                processed: completedEntries,
+                total: totalEntries,
+            });
+            return undefined;
+        }, 
+        // Default concurrency of 4 works well for cloud LLM APIs. Local model
+        // servers (LM Studio, Ollama) run one inference at a time — set
+        // `llm.concurrency: 1` in config.json to avoid "Model reloaded" / 500
+        // errors from concurrent request overload.
+        getDefaultLlmConcurrency(llmConfig));
+    }
+    finally {
+        if (heartbeatTimer)
+            clearInterval(heartbeatTimer);
+    }
+    if (deadlineHit) {
+        warn("[akm] LLM enrichment budget exceeded. Re-run `akm index` to continue. Increase llm.timeoutMs for a larger budget.");
     }
     if (summary.attempted > 0 && summary.succeeded === 0) {
         const sample = summary.failureSamples.length ? ` Example: ${summary.failureSamples[0]}` : "";
@@ -742,14 +937,24 @@ async function generateEmbeddingsForDb(db, config, onProgress, signal) {
         throwIfAborted(signal);
         // Wrap all embedding upserts in a single transaction so partial
         // state is rolled back on failure rather than leaving the table half-filled.
+        let storedCount = 0;
+        let skippedCount = 0;
         db.transaction(() => {
             for (let i = 0; i < allEntries.length; i++) {
-                upsertEmbedding(db, allEntries[i].id, embeddings[i]);
+                if (upsertEmbedding(db, allEntries[i].id, embeddings[i])) {
+                    storedCount++;
+                }
+                else {
+                    skippedCount++;
+                }
             }
         })();
+        if (skippedCount > 0) {
+            warn(`[embed] ${skippedCount} embedding${skippedCount === 1 ? "" : "s"} skipped (entry deleted between queue and write)`);
+        }
         onProgress({
             phase: "embeddings",
-            message: `Stored ${embeddings.length} embedding${embeddings.length === 1 ? "" : "s"}.`,
+            message: `Stored ${storedCount} embedding${storedCount === 1 ? "" : "s"}.`,
         });
         setMeta(db, "embeddingFingerprint", currentFingerprint);
         return { success: true };
@@ -769,14 +974,6 @@ async function generateEmbeddingsForDb(db, config, onProgress, signal) {
     }
 }
 // ── Helpers ─────────────────────────────────────────────────────────────────
-function getAllEntriesForEmbedding(db) {
-    return db
-        .prepare(`
-      SELECT e.id, e.search_text AS searchText, e.entry_key AS entryKey, e.file_path AS filePath FROM entries e
-      WHERE NOT EXISTS (SELECT 1 FROM embeddings b WHERE b.id = e.id)
-    `)
-        .all();
-}
 function attachFileSize(entry, entryPath) {
     try {
         return { ...entry, fileSize: fs.statSync(entryPath).size };
@@ -785,28 +982,6 @@ function attachFileSize(entry, entryPath) {
         return entry;
     }
 }
-function upsertWorkflowDocument(db, entryId, doc, content) {
-    const sourceHash = computeSourceHash(content);
-    db.prepare(`INSERT INTO workflow_documents (entry_id, schema_version, document_json, source_path, source_hash, updated_at)
-     VALUES (?, ?, ?, ?, ?, ?)
-     ON CONFLICT(entry_id) DO UPDATE SET
-       schema_version = excluded.schema_version,
-       document_json = excluded.document_json,
-       source_path = excluded.source_path,
-       source_hash = excluded.source_hash,
-       updated_at = excluded.updated_at`).run(entryId, doc.schemaVersion, JSON.stringify(doc), doc.source.path, sourceHash, new Date().toISOString());
-}
-function computeSourceHash(content) {
-    // Cheap, stable identity for the source markdown — used by future
-    // incremental fast-paths that skip re-validation when content is unchanged.
-    // Not security-sensitive; FNV-1a over the bytes is sufficient.
-    let hash = 0x811c9dc5;
-    for (let i = 0; i < content.length; i++) {
-        hash ^= content[i];
-        hash = Math.imul(hash, 0x01000193);
-    }
-    return (hash >>> 0).toString(16);
-}
 function buildIndexSummaryMessage(options) {
     const stashSourceLabel = options.sourcesCount === 1 ? "stash source" : "stash sources";
     const semanticDetail = getSemanticSearchLabel(options.semanticSearchMode, options.embeddingProvider, options.vecAvailable);
@@ -899,11 +1074,12 @@ function resolveIndexedFiles(dirPath, files, stash) {
     }
     return resolved.size > 0 ? [...resolved] : files;
 }
-async function enhanceStashWithLlm(llmConfig, stash, files, summary, signal) {
+async function enhanceStashWithLlm(llmConfig, stash, files, summary, signal, db, entryKeys, reEnrich, akmConfig, onEntryDone) {
     const { enhanceMetadata } = await import("../llm/metadata-enhance");
-    const enhanced = [];
-    for (const entry of stash.entries) {
-        throwIfAborted(signal);
+    const { computeBodyHash, getLlmCacheEntry, upsertLlmCacheEntry } = await import("./db.js");
+    const results = await concurrentMap(stash.entries, async (entry, idx) => {
+        if (signal?.aborted)
+            return entry;
         summary.attempted++;
         try {
             const entryFile = entry.filename
@@ -915,10 +1091,38 @@ async function enhanceStashWithLlm(llmConfig, stash, files, summary, signal) {
                     fileContent = fs.readFileSync(entryFile, "utf8");
                 }
                 catch {
-                    /* ignore unreadable files */
+                    warn(`Could not read file for LLM enrichment: ${entry.filename ?? entry.name}`);
+                }
+            }
+            // Incremental cache: skip LLM call when file body is unchanged and
+            // --re-enrich was not requested. The cache key is the entry_key
+            // (stashDir:type:name) which is stable across index runs.
+            const cacheBody = fileContent ?? `${entry.name}\n${entry.description ?? ""}`;
+            const bodyHash = computeBodyHash(cacheBody);
+            const cacheKey = entryKeys?.[idx] ?? `${entry.type}:${entry.name}`;
+            if (db && !reEnrich) {
+                const cached = getLlmCacheEntry(db, cacheKey, bodyHash);
+                if (cached) {
+                    try {
+                        const parsed = JSON.parse(cached.resultJson);
+                        const updated = { ...entry };
+                        if (parsed.description)
+                            updated.description = parsed.description;
+                        if (parsed.searchHints?.length)
+                            updated.searchHints = parsed.searchHints;
+                        if (parsed.tags?.length)
+                            updated.tags = parsed.tags;
+                        updated.quality = "enriched";
+                        summary.succeeded++;
+                        onEntryDone?.({ entryName: entry.name, outcome: "cache-hit" });
+                        return updated;
+                    }
+                    catch {
+                        warn(`LLM enrichment cache entry corrupt for ${entry.name}; re-running enrichment`);
+                    }
                 }
             }
-            const improvements = await enhanceMetadata(llmConfig, entry, fileContent, signal);
+            const improvements = await enhanceMetadata(llmConfig, entry, fileContent, signal, akmConfig);
             const updated = { ...entry };
             if (improvements.description)
                 updated.description = improvements.description;
@@ -926,19 +1130,39 @@ async function enhanceStashWithLlm(llmConfig, stash, files, summary, signal) {
                 updated.searchHints = improvements.searchHints;
             if (improvements.tags?.length)
                 updated.tags = improvements.tags;
-            enhanced.push(updated);
+            // Mark as enriched so subsequent index runs skip re-enrichment (P2)
+            updated.quality = "enriched";
+            // Persist to cache so the next run can skip the LLM call when the
+            // file body has not changed.
+            if (db) {
+                upsertLlmCacheEntry(db, cacheKey, bodyHash, JSON.stringify({
+                    description: improvements.description,
+                    searchHints: improvements.searchHints,
+                    tags: improvements.tags,
+                }));
+            }
             summary.succeeded++;
+            onEntryDone?.({ entryName: entry.name, outcome: "llm" });
+            return updated;
         }
         catch (err) {
-            enhanced.push(entry);
             const msg = toErrorMessage(err);
             // failureSamples is bounded to 3 items, so a linear scan is cheaper
             // than maintaining a parallel Set for membership checks (#177 review).
             if (summary.failureSamples.length < 3 && !summary.failureSamples.includes(msg)) {
                 summary.failureSamples.push(msg);
             }
+            onEntryDone?.({ entryName: entry.name, outcome: "failed" });
+            return entry;
         }
-    }
+    }, 
+    // Default concurrency of 4 works well for cloud LLM APIs. Set
+    // `llm.concurrency: 1` in config.json for local model servers.
+    getDefaultLlmConcurrency(llmConfig));
+    // concurrentMap returns Array<T | undefined>; filter out undefined slots
+    // (which can only occur if the callback itself returned undefined, which
+    // it never does above — but TypeScript needs the filter for type safety).
+    const enhanced = results.map((r, i) => r ?? stash.entries[i]);
     return { entries: enhanced };
 }
 /**
@@ -1018,13 +1242,13 @@ export async function lookup(ref) {
     const dbPath = getDbPath();
     const db = openExistingDatabase(dbPath);
     try {
-        // entry_key shape: `${stashDir}:${type}:${name}`. Suffix-match on
-        // `:type:name` so we can scope by source dir as a prefix when origin is
-        // supplied. Use parameterised queries throughout — names may include
-        // user-supplied glob characters.
         const escapeLike = (value) => value.replace(/\\/g, "\\\\").replace(/%/g, "\\%").replace(/_/g, "\\_");
-        const suffix = `:${ref.type}:${ref.name}`;
-        const escapedSuffix = escapeLike(suffix);
+        // Canonical names strip .md for markdown assets, but users often pass
+        // refs with .md (e.g. command:release.md). Normalize by trying both.
+        const nameVariants = [ref.name];
+        if (ref.name.endsWith(".md")) {
+            nameVariants.push(ref.name.slice(0, -3));
+        }
         const candidateDirs = (() => {
             if (!ref.origin)
                 return sources.map((s) => s.path);
@@ -1035,20 +1259,24 @@ export async function lookup(ref) {
         })();
         if (candidateDirs.length === 0)
             return null;
-        for (const dir of candidateDirs) {
-            const escapedDir = escapeLike(dir);
-            const row = db
-                .prepare("SELECT entry_key AS entryKey, file_path AS filePath, stash_dir AS stashDir, entry_type AS type FROM entries " +
-                "WHERE entry_key LIKE ? ESCAPE '\\' AND entry_type = ? LIMIT 1")
-                .get(`${escapedDir}${escapedSuffix}`, ref.type);
-            if (row) {
-                return {
-                    entryKey: row.entryKey,
-                    filePath: row.filePath,
-                    stashDir: row.stashDir,
-                    type: row.type,
-                    name: ref.name,
-                };
+        for (const name of nameVariants) {
+            const suffix = `:${ref.type}:${name}`;
+            const escapedSuffix = escapeLike(suffix);
+            for (const dir of candidateDirs) {
+                const escapedDir = escapeLike(dir);
+                const row = db
+                    .prepare("SELECT entry_key AS entryKey, file_path AS filePath, stash_dir AS stashDir, entry_type AS type FROM entries " +
+                    "WHERE entry_key LIKE ? ESCAPE '\\' AND entry_type = ? LIMIT 1")
+                    .get(`${escapedDir}${escapedSuffix}`, ref.type);
+                if (row) {
+                    return {
+                        entryKey: row.entryKey,
+                        filePath: row.filePath,
+                        stashDir: row.stashDir,
+                        type: row.type,
+                        name: ref.name,
+                    };
+                }
             }
         }
         return null;
@@ -1093,18 +1321,25 @@ export function recomputeUtilityScores(db) {
     const emaDecay = EMA_DECAY ** elapsedDays;
     const emaNew = 1 - emaDecay; // complement so weights still sum to 1
     // Single aggregate query instead of N+1 per-entry queries.
-    // Only processes entries that actually have usage events.
+    // Only processes entries that actually have usage events AND still exist
+    // in `entries`. The latter check is critical: usage_events has no FK to
+    // entries, so its entry_id can become stale (entry deleted, re-keyed,
+    // moved between sources). Without the JOIN, writing the derived row to
+    // utility_scores (which DOES have an FK) raises "FOREIGN KEY constraint
+    // failed" and rolls back the whole finalize transaction — failing every
+    // index run.
     const usageRows = db
         .prepare(`
-      SELECT entry_id,
-             SUM(CASE WHEN event_type = 'search' THEN 1 ELSE 0 END) AS search_count,
-             SUM(CASE WHEN event_type = 'show'   THEN 1 ELSE 0 END) AS show_count,
-             SUM(CASE WHEN event_type = 'feedback' AND signal = 'positive' THEN 1 ELSE 0 END) AS positive_feedback_count,
-             SUM(CASE WHEN event_type = 'feedback' AND signal = 'negative' THEN 1 ELSE 0 END) AS negative_feedback_count,
-             MAX(created_at) AS last_used_at
-      FROM usage_events
-      WHERE entry_id IS NOT NULL
-      GROUP BY entry_id
+      SELECT u.entry_id,
+             SUM(CASE WHEN u.event_type = 'search' THEN 1 ELSE 0 END) AS search_count,
+             SUM(CASE WHEN u.event_type = 'show'   THEN 1 ELSE 0 END) AS show_count,
+             SUM(CASE WHEN u.event_type = 'feedback' AND u.signal = 'positive' THEN 1 ELSE 0 END) AS positive_feedback_count,
+             SUM(CASE WHEN u.event_type = 'feedback' AND u.signal = 'negative' THEN 1 ELSE 0 END) AS negative_feedback_count,
+             MAX(u.created_at) AS last_used_at
+      FROM usage_events u
+      JOIN entries e ON e.id = u.entry_id
+      WHERE u.entry_id IS NOT NULL
+      GROUP BY u.entry_id
     `)
         .all();
     if (usageRows.length === 0) {
@@ -1113,23 +1348,26 @@ export function recomputeUtilityScores(db) {
     }
     // Batch-load existing utility scores
     const existingScores = new Map();
-    const scoreRows = db.prepare("SELECT entry_id, utility FROM utility_scores").all();
+    const scoreRows = db.prepare("SELECT entry_id, utility, last_used_at FROM utility_scores").all();
     for (const row of scoreRows) {
-        existingScores.set(row.entry_id, row.utility);
+        existingScores.set(row.entry_id, { utility: row.utility, lastUsedAt: row.last_used_at ?? undefined });
     }
+    const now = new Date().toISOString();
     for (const row of usageRows) {
         const selectRate = row.search_count > 0 ? Math.min(1, row.show_count / row.search_count) : 0;
         const feedbackTotal = row.positive_feedback_count + row.negative_feedback_count;
         const feedbackRate = feedbackTotal > 0 ? Math.max(0, row.positive_feedback_count - row.negative_feedback_count) / feedbackTotal : 0;
         const effectiveRate = Math.max(selectRate, feedbackRate);
-        const prevUtility = existingScores.get(row.entry_id) ?? 0;
+        const existing = existingScores.get(row.entry_id);
+        const prevUtility = existing?.utility ?? 0;
         const utility = prevUtility * emaDecay + effectiveRate * emaNew;
+        const lastUsedAt = effectiveRate > 0.5 ? now : (existing?.lastUsedAt ?? undefined);
         upsertUtilityScore(db, row.entry_id, {
             utility,
             showCount: row.show_count,
             searchCount: row.search_count,
             selectRate,
-            lastUsedAt: row.last_used_at ?? undefined,
+            lastUsedAt,
         });
     }
     setMeta(db, "last_utility_computed_at", new Date().toISOString());