@kolisachint/hoocode-agent 0.4.140 → 0.4.142

package/dist/core/embsearch/embsearch-service.js

@@ -0,0 +1,254 @@
+/**
+ * Orchestrates semantic indexing and search for a repository.
+ *
+ * Lifecycle (all behind --enable-embsearchtools):
+ *  1. `start()` — resolve the embsearch binary, scan the repo (ignore-aware),
+ *     apply the byte threshold. Under threshold → dormant. Over → spawn the
+ *     daemon, verify the backend is not the mock embedder, then index changed
+ *     files in the background in small batches, reporting progress.
+ *  2. `search()` — top-k semantic query, mapping chunk ids back to
+ *     `path:start-end` via the sidecar metadata.
+ *  3. `dispose()` — save + close the daemon.
+ *
+ * Every failure degrades to `unavailable` with a reason; nothing here ever
+ * blocks session startup or affects grep/find.
+ */
+import { readFileSync } from "fs";
+import { ensureTool } from "../../utils/tools-manager.js";
+import { chunkFile } from "./chunker.js";
+import { EmbSearchClient } from "./client.js";
+import { emptyIndexMeta, getEmbsearchStoreDir, getVectorStoreDir, hashContent, loadIndexMeta, saveIndexMeta, } from "./index-meta.js";
+import { scanRepo } from "./repo-scan.js";
+/** Chunks per bulk request. Small enough that a concurrent query is never
+ *  stuck long behind one padded batch inference. */
+const BULK_BATCH_SIZE = 48;
+/** Yield between batches so background indexing doesn't starve the session. */
+const BATCH_YIELD_MS = 15;
+/** The Rust mock backend's model id — semantically meaningless, never index with it. */
+const MOCK_MODEL_ID = "mock-hash-v1";
+export class EmbsearchService {
+    options;
+    client;
+    meta;
+    state = { phase: "idle" };
+    disposed = false;
+    constructor(options) {
+        this.options = options;
+    }
+    getState() {
+        return this.state;
+    }
+    /** Semantic search is usable (index ready, or still building with partial data). */
+    isAvailable() {
+        return this.state.phase === "ready" || this.state.phase === "indexing";
+    }
+    setState(state) {
+        this.state = state;
+        this.options.onProgress?.(state);
+    }
+    async resolveBinary() {
+        if (this.options.binaryPath) {
+            return this.options.binaryPath;
+        }
+        return await ensureTool("embsearch", true);
+    }
+    /**
+     * Scan, threshold-check, and (when needed) index in the background.
+     * Resolves when indexing completes or the feature settles dormant.
+     */
+    async start(signal) {
+        try {
+            await this.run(signal);
+        }
+        catch (e) {
+            const reason = e instanceof Error ? e.message : String(e);
+            this.setState({ phase: "unavailable", reason });
+            await this.closeClient();
+        }
+    }
+    async run(signal) {
+        const binary = await this.resolveBinary();
+        if (!binary) {
+            this.setState({
+                phase: "unavailable",
+                reason: "embsearch binary not found (PATH or embsearchBinaryPath setting)",
+            });
+            return;
+        }
+        const scan = scanRepo(this.options.cwd, signal);
+        if (scan.totalBytes < this.options.thresholdBytes) {
+            this.setState({
+                phase: "skipped",
+                reason: `repo under threshold (${scan.totalBytes} < ${this.options.thresholdBytes} bytes)`,
+            });
+            return;
+        }
+        const storeDir = getEmbsearchStoreDir(this.options.cwd);
+        this.client = new EmbSearchClient({ binaryPath: binary, storePath: getVectorStoreDir(storeDir) });
+        await this.client.ready();
+        const info = await this.client.info();
+        if (info.modelId === MOCK_MODEL_ID) {
+            throw new Error("embsearch binary uses the mock embedder (not semantic); install an onnx build");
+        }
+        // Missing/stale sidecar (format, chunker, or model changed) → clean rebuild.
+        this.meta = loadIndexMeta(storeDir, info.modelId) ?? emptyIndexMeta(this.options.cwd, info.modelId);
+        this.meta.lastUsedMs = Date.now();
+        await this.indexChangedFiles(scan.files, storeDir, signal);
+    }
+    async indexChangedFiles(files, storeDir, signal) {
+        const meta = this.meta;
+        const client = this.client;
+        // Diff scan vs sidecar: cheap mtime+size check first, hash only on delta.
+        const toIndex = [];
+        const seen = new Set();
+        for (const file of files) {
+            seen.add(file.rel);
+            const known = meta.files[file.rel];
+            if (known && known.mtimeMs === file.mtimeMs && known.size === file.size)
+                continue;
+            let content;
+            try {
+                content = readFileSync(file.abs, "utf-8");
+            }
+            catch {
+                continue;
+            }
+            const hash = hashContent(content);
+            if (known && known.hash === hash) {
+                // Touched but unchanged — refresh stat info only.
+                known.mtimeMs = file.mtimeMs;
+                known.size = file.size;
+                continue;
+            }
+            toIndex.push({ file, content, hash });
+        }
+        const toRemove = Object.keys(meta.files).filter((rel) => !seen.has(rel));
+        if (toIndex.length === 0 && toRemove.length === 0) {
+            saveIndexMeta(storeDir, meta);
+            this.setState({ phase: "ready", chunkCount: this.countChunks(meta) });
+            return;
+        }
+        // Chunk changed files; count total upserts for exact progress.
+        const work = [];
+        let totalChunks = 0;
+        for (const { file, content, hash } of toIndex) {
+            const chunks = chunkFile(file.rel, content);
+            work.push({
+                rel: file.rel,
+                fileMeta: {
+                    mtimeMs: file.mtimeMs,
+                    size: file.size,
+                    hash,
+                    chunks: chunks.map((c) => [c.startLine, c.endLine]),
+                },
+                chunks: chunks.map((c) => ({ id: c.id, text: c.text })),
+            });
+            totalChunks += chunks.length;
+        }
+        this.setState({ phase: "indexing", done: 0, total: totalChunks });
+        // Drop vectors of deleted files and superseded chunk tails.
+        for (const rel of toRemove) {
+            for (let i = 0; i < meta.files[rel].chunks.length; i++)
+                await client.remove(`${rel}#${i}`);
+            delete meta.files[rel];
+        }
+        let done = 0;
+        for (const item of work) {
+            if (signal?.aborted || this.disposed)
+                return;
+            const oldChunkCount = meta.files[item.rel]?.chunks.length ?? 0;
+            // Remove old chunks beyond the new count (upsert covers the rest).
+            for (let i = item.chunks.length; i < oldChunkCount; i++)
+                await client.remove(`${item.rel}#${i}`);
+            for (let offset = 0; offset < item.chunks.length; offset += BULK_BATCH_SIZE) {
+                if (signal?.aborted || this.disposed)
+                    return;
+                const batch = item.chunks.slice(offset, offset + BULK_BATCH_SIZE);
+                await client.bulk(batch);
+                done += batch.length;
+                this.setState({ phase: "indexing", done, total: totalChunks });
+                // Yield so queries and the event loop stay responsive.
+                await new Promise((resolve) => setTimeout(resolve, BATCH_YIELD_MS));
+            }
+            meta.files[item.rel] = item.fileMeta;
+        }
+        await client.compact();
+        await client.save();
+        saveIndexMeta(storeDir, meta);
+        this.setState({ phase: "ready", chunkCount: this.countChunks(meta) });
+    }
+    countChunks(meta) {
+        let n = 0;
+        for (const rel of Object.keys(meta.files))
+            n += meta.files[rel].chunks.length;
+        return n;
+    }
+    /** Top-`k` semantic hits as `path` + line range + score. */
+    async search(query, k = 10) {
+        if (!this.client || this.client.isClosed || !this.meta) {
+            throw new Error("semantic index is not available");
+        }
+        const results = await this.client.query(query, k);
+        const hits = [];
+        for (const result of results) {
+            const sep = result.id.lastIndexOf("#");
+            if (sep === -1)
+                continue;
+            const rel = result.id.slice(0, sep);
+            const chunkIndex = Number.parseInt(result.id.slice(sep + 1), 10);
+            const range = this.meta.files[rel]?.chunks[chunkIndex];
+            if (!range)
+                continue;
+            hits.push({ path: rel, startLine: range[0], endLine: range[1], score: result.score });
+        }
+        return hits;
+    }
+    async closeClient() {
+        const client = this.client;
+        this.client = undefined;
+        if (client && !client.isClosed) {
+            try {
+                await client.close();
+            }
+            catch {
+                // already dead
+            }
+        }
+    }
+    /** Persist state and shut the daemon down. Safe to call twice. */
+    async dispose() {
+        if (this.disposed)
+            return;
+        this.disposed = true;
+        if (this.client && !this.client.isClosed) {
+            try {
+                await this.client.save();
+            }
+            catch {
+                // daemon may have exited; nothing to save
+            }
+        }
+        await this.closeClient();
+    }
+}
+// --- Per-cwd service registry ---
+//
+// The semantic_search tool is constructed by the generic tool factory table and
+// only receives `cwd`; the service is created later during session init (flag
+// gated). This registry connects the two without threading a service instance
+// through every layer between main.ts and the tool factories.
+const services = new Map();
+export function registerEmbsearchService(cwd, service) {
+    const old = services.get(cwd);
+    if (old && old !== service) {
+        old.dispose().catch(() => { });
+    }
+    services.set(cwd, service);
+}
+export function getEmbsearchService(cwd) {
+    return services.get(cwd);
+}
+export function unregisterEmbsearchService(cwd) {
+    services.delete(cwd);
+}
+//# sourceMappingURL=embsearch-service.js.map

package/dist/core/embsearch/embsearch-service.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"embsearch-service.js","sourceRoot":"","sources":["../../../src/core/embsearch/embsearch-service.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,OAAO,EAAE,YAAY,EAAE,MAAM,IAAI,CAAC;AAClC,OAAO,EAAE,UAAU,EAAE,MAAM,8BAA8B,CAAC;AAC1D,OAAO,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AACzC,OAAO,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AAC9C,OAAO,EACN,cAAc,EAEd,oBAAoB,EACpB,iBAAiB,EACjB,WAAW,EAEX,aAAa,EACb,aAAa,GACb,MAAM,iBAAiB,CAAC;AACzB,OAAO,EAAqB,QAAQ,EAAE,MAAM,gBAAgB,CAAC;AAE7D;oDACoD;AACpD,MAAM,eAAe,GAAG,EAAE,CAAC;AAC3B,+EAA+E;AAC/E,MAAM,cAAc,GAAG,EAAE,CAAC;AAC1B,0FAAwF;AACxF,MAAM,aAAa,GAAG,cAAc,CAAC;AA0BrC,MAAM,OAAO,gBAAgB;IACX,OAAO,CAA0B;IAC1C,MAAM,CAA8B;IACpC,IAAI,CAAwB;IAC5B,KAAK,GAAmB,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC;IAC1C,QAAQ,GAAG,KAAK,CAAC;IAEzB,YAAY,OAAgC,EAAE;QAC7C,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;IAAA,CACvB;IAED,QAAQ,GAAmB;QAC1B,OAAO,IAAI,CAAC,KAAK,CAAC;IAAA,CAClB;IAED,oFAAoF;IACpF,WAAW,GAAY;QACtB,OAAO,IAAI,CAAC,KAAK,CAAC,KAAK,KAAK,OAAO,IAAI,IAAI,CAAC,KAAK,CAAC,KAAK,KAAK,UAAU,CAAC;IAAA,CACvE;IAEO,QAAQ,CAAC,KAAqB,EAAQ;QAC7C,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC,KAAK,CAAC,CAAC;IAAA,CACjC;IAEO,KAAK,CAAC,aAAa,GAAgC;QAC1D,IAAI,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC;YAC7B,OAAO,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC;QAChC,CAAC;QACD,OAAO,MAAM,UAAU,CAAC,WAAW,EAAE,IAAI,CAAC,CAAC;IAAA,CAC3C;IAED;;;OAGG;IACH,KAAK,CAAC,KAAK,CAAC,MAAoB,EAAiB;QAChD,IAAI,CAAC;YACJ,MAAM,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;QACxB,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACZ,MAAM,MAAM,GAAG,CAAC,YAAY,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;YAC1D,IAAI,CAAC,QAAQ,CAAC,EAAE,KAAK,EAAE,aAAa,EAAE,MAAM,EAAE,CAAC,CAAC;YAChD,MAAM,IAAI,CAAC,WAAW,EAAE,CAAC;QAC1B,CAAC;IAAA,CACD;IAEO,KAAK,CAAC,GAAG,CAAC,MAAoB,EAAiB;QACtD,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,aAAa,EAAE,CAAC;QAC1C,IAAI,CAAC,MAAM,EAAE,CAAC;YACb,IAAI,CAAC,QAAQ,CAAC;gBACb,KAAK,EAAE,aAAa;gBACpB,MAAM,EAAE,kEAAkE;aAC1E,CAAC,CAAC;YACH,OAAO;QACR,CAAC;QAED,MAAM,IAAI,GAAG,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAChD,IAAI,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,OAAO,CAAC,cAAc,EAAE,CAAC;YACnD,IAAI,CAAC,QAAQ,CAAC;gBACb,KAAK,EAAE,SAAS;gBAChB,MAAM,EAAE,yBAAyB,IAAI,CAAC,UAAU,MAAM,IAAI,CAAC,OAAO,CAAC,cAAc,SAAS;aAC1F,CAAC,CAAC;YACH,OAAO;QACR,CAAC;QAED,MAAM,QAAQ,GAAG,oBAAoB,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;QACxD,IAAI,CAAC,MAAM,GAAG,IAAI,eAAe,CAAC,EAAE,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,iBAAiB,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC;QAClG,MAAM,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC;QAE1B,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;QACtC,IAAI,IAAI,CAAC,OAAO,KAAK,aAAa,EAAE,CAAC;YACpC,MAAM,IAAI,KAAK,CAAC,+EAA+E,CAAC,CAAC;QAClG,CAAC;QAED,+EAA6E;QAC7E,IAAI,CAAC,IAAI,GAAG,aAAa,CAAC,QAAQ,EAAE,IAAI,CAAC,OAAO,CAAC,IAAI,cAAc,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,EAAE,IAAI,CAAC,OAAO,CAAC,CAAC;QACpG,IAAI,CAAC,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAElC,MAAM,IAAI,CAAC,iBAAiB,CAAC,IAAI,CAAC,KAAK,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC;IAAA,CAC3D;IAEO,KAAK,CAAC,iBAAiB,CAAC,KAAqB,EAAE,QAAgB,EAAE,MAAoB,EAAiB;QAC7G,MAAM,IAAI,GAAG,IAAI,CAAC,IAAK,CAAC;QACxB,MAAM,MAAM,GAAG,IAAI,CAAC,MAAO,CAAC;QAE5B,0EAA0E;QAC1E,MAAM,OAAO,GAAiE,EAAE,CAAC;QACjF,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;QAC/B,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YAC1B,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YACnB,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YACnC,IAAI,KAAK,IAAI,KAAK,CAAC,OAAO,KAAK,IAAI,CAAC,OAAO,IAAI,KAAK,CAAC,IAAI,KAAK,IAAI,CAAC,IAAI;gBAAE,SAAS;YAClF,IAAI,OAAe,CAAC;YACpB,IAAI,CAAC;gBACJ,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;YAC3C,CAAC;YAAC,MAAM,CAAC;gBACR,SAAS;YACV,CAAC;YACD,MAAM,IAAI,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;YAClC,IAAI,KAAK,IAAI,KAAK,CAAC,IAAI,KAAK,IAAI,EAAE,CAAC;gBAClC,oDAAkD;gBAClD,KAAK,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC;gBAC7B,KAAK,CAAC,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC;gBACvB,SAAS;YACV,CAAC;YACD,OAAO,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;QACvC,CAAC;QACD,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC;QAEzE,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACnD,aAAa,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;YAC9B,IAAI,CAAC,QAAQ,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,UAAU,EAAE,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACtE,OAAO;QACR,CAAC;QAED,+DAA+D;QAC/D,MAAM,IAAI,GAA4F,EAAE,CAAC;QACzG,IAAI,WAAW,GAAG,CAAC,CAAC;QACpB,KAAK,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,OAAO,EAAE,CAAC;YAC/C,MAAM,MAAM,GAAG,SAAS,CAAC,IAAI,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;YAC5C,IAAI,CAAC,IAAI,CAAC;gBACT,GAAG,EAAE,IAAI,CAAC,GAAG;gBACb,QAAQ,EAAE;oBACT,OAAO,EAAE,IAAI,CAAC,OAAO;oBACrB,IAAI,EAAE,IAAI,CAAC,IAAI;oBACf,IAAI;oBACJ,MAAM,EAAE,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC;iBACnD;gBACD,MAAM,EAAE,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;aACvD,CAAC,CAAC;YACH,WAAW,IAAI,MAAM,CAAC,MAAM,CAAC;QAC9B,CAAC;QAED,IAAI,CAAC,QAAQ,CAAC,EAAE,KAAK,EAAE,UAAU,EAAE,IAAI,EAAE,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,CAAC,CAAC;QAElE,4DAA4D;QAC5D,KAAK,MAAM,GAAG,IAAI,QAAQ,EAAE,CAAC;YAC5B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC,EAAE;gBAAE,MAAM,MAAM,CAAC,MAAM,CAAC,GAAG,GAAG,IAAI,CAAC,EAAE,CAAC,CAAC;YAC3F,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QACxB,CAAC;QAED,IAAI,IAAI,GAAG,CAAC,CAAC;QACb,KAAK,MAAM,IAAI,IAAI,IAAI,EAAE,CAAC;YACzB,IAAI,MAAM,EAAE,OAAO,IAAI,IAAI,CAAC,QAAQ;gBAAE,OAAO;YAC7C,MAAM,aAAa,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,IAAI,CAAC,CAAC;YAC/D,mEAAmE;YACnE,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC,GAAG,aAAa,EAAE,CAAC,EAAE;gBAAE,MAAM,MAAM,CAAC,MAAM,CAAC,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC,EAAE,CAAC,CAAC;YAEjG,KAAK,IAAI,MAAM,GAAG,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,IAAI,eAAe,EAAE,CAAC;gBAC7E,IAAI,MAAM,EAAE,OAAO,IAAI,IAAI,CAAC,QAAQ;oBAAE,OAAO;gBAC7C,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,EAAE,MAAM,GAAG,eAAe,CAAC,CAAC;gBAClE,MAAM,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;gBACzB,IAAI,IAAI,KAAK,CAAC,MAAM,CAAC;gBACrB,IAAI,CAAC,QAAQ,CAAC,EAAE,KAAK,EAAE,UAAU,EAAE,IAAI,EAAE,KAAK,EAAE,WAAW,EAAE,CAAC,CAAC;gBAC/D,uDAAuD;gBACvD,MAAM,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC,CAAC;YACrE,CAAC;YACD,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,IAAI,CAAC,QAAQ,CAAC;QACtC,CAAC;QAED,MAAM,MAAM,CAAC,OAAO,EAAE,CAAC;QACvB,MAAM,MAAM,CAAC,IAAI,EAAE,CAAC;QACpB,aAAa,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;QAC9B,IAAI,CAAC,QAAQ,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,UAAU,EAAE,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAAA,CACtE;IAEO,WAAW,CAAC,IAAe,EAAU;QAC5C,IAAI,CAAC,GAAG,CAAC,CAAC;QACV,KAAK,MAAM,GAAG,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC;YAAE,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC;QAC9E,OAAO,CAAC,CAAC;IAAA,CACT;IAED,4DAA4D;IAC5D,KAAK,CAAC,MAAM,CAAC,KAAa,EAAE,CAAC,GAAG,EAAE,EAA0B;QAC3D,IAAI,CAAC,IAAI,CAAC,MAAM,IAAI,IAAI,CAAC,MAAM,CAAC,QAAQ,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;YACxD,MAAM,IAAI,KAAK,CAAC,iCAAiC,CAAC,CAAC;QACpD,CAAC;QACD,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;QAClD,MAAM,IAAI,GAAkB,EAAE,CAAC;QAC/B,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;YAC9B,MAAM,GAAG,GAAG,MAAM,CAAC,EAAE,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC;YACvC,IAAI,GAAG,KAAK,CAAC,CAAC;gBAAE,SAAS;YACzB,MAAM,GAAG,GAAG,MAAM,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;YACpC,MAAM,UAAU,GAAG,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC,KAAK,CAAC,GAAG,GAAG,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YACjE,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,MAAM,CAAC,UAAU,CAAC,CAAC;YACvD,IAAI,CAAC,KAAK;gBAAE,SAAS;YACrB,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,GAAG,EAAE,SAAS,EAAE,KAAK,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;QACvF,CAAC;QACD,OAAO,IAAI,CAAC;IAAA,CACZ;IAEO,KAAK,CAAC,WAAW,GAAkB;QAC1C,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC;QAC3B,IAAI,CAAC,MAAM,GAAG,SAAS,CAAC;QACxB,IAAI,MAAM,IAAI,CAAC,MAAM,CAAC,QAAQ,EAAE,CAAC;YAChC,IAAI,CAAC;gBACJ,MAAM,MAAM,CAAC,KAAK,EAAE,CAAC;YACtB,CAAC;YAAC,MAAM,CAAC;gBACR,eAAe;YAChB,CAAC;QACF,CAAC;IAAA,CACD;IAED,kEAAkE;IAClE,KAAK,CAAC,OAAO,GAAkB;QAC9B,IAAI,IAAI,CAAC,QAAQ;YAAE,OAAO;QAC1B,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC;QACrB,IAAI,IAAI,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,QAAQ,EAAE,CAAC;YAC1C,IAAI,CAAC;gBACJ,MAAM,IAAI,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;YAC1B,CAAC;YAAC,MAAM,CAAC;gBACR,0CAA0C;YAC3C,CAAC;QACF,CAAC;QACD,MAAM,IAAI,CAAC,WAAW,EAAE,CAAC;IAAA,CACzB;CACD;AAED,mCAAmC;AACnC,EAAE;AACF,gFAAgF;AAChF,8EAA8E;AAC9E,8EAA8E;AAC9E,8DAA8D;AAE9D,MAAM,QAAQ,GAAG,IAAI,GAAG,EAA4B,CAAC;AAErD,MAAM,UAAU,wBAAwB,CAAC,GAAW,EAAE,OAAyB,EAAQ;IACtF,MAAM,GAAG,GAAG,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IAC9B,IAAI,GAAG,IAAI,GAAG,KAAK,OAAO,EAAE,CAAC;QAC5B,GAAG,CAAC,OAAO,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAC,CAAC,CAAC,CAAC;IAC/B,CAAC;IACD,QAAQ,CAAC,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;AAAA,CAC3B;AAED,MAAM,UAAU,mBAAmB,CAAC,GAAW,EAAgC;IAC9E,OAAO,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;AAAA,CACzB;AAED,MAAM,UAAU,0BAA0B,CAAC,GAAW,EAAQ;IAC7D,QAAQ,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;AAAA,CACrB","sourcesContent":["/**\n * Orchestrates semantic indexing and search for a repository.\n *\n * Lifecycle (all behind --enable-embsearchtools):\n *  1. `start()` — resolve the embsearch binary, scan the repo (ignore-aware),\n *     apply the byte threshold. Under threshold → dormant. Over → spawn the\n *     daemon, verify the backend is not the mock embedder, then index changed\n *     files in the background in small batches, reporting progress.\n *  2. `search()` — top-k semantic query, mapping chunk ids back to\n *     `path:start-end` via the sidecar metadata.\n *  3. `dispose()` — save + close the daemon.\n *\n * Every failure degrades to `unavailable` with a reason; nothing here ever\n * blocks session startup or affects grep/find.\n */\n\nimport { readFileSync } from \"fs\";\nimport { ensureTool } from \"../../utils/tools-manager.js\";\nimport { chunkFile } from \"./chunker.js\";\nimport { EmbSearchClient } from \"./client.js\";\nimport {\n\temptyIndexMeta,\n\ttype FileMeta,\n\tgetEmbsearchStoreDir,\n\tgetVectorStoreDir,\n\thashContent,\n\ttype IndexMeta,\n\tloadIndexMeta,\n\tsaveIndexMeta,\n} from \"./index-meta.js\";\nimport { type RepoScanFile, scanRepo } from \"./repo-scan.js\";\n\n/** Chunks per bulk request. Small enough that a concurrent query is never\n *  stuck long behind one padded batch inference. */\nconst BULK_BATCH_SIZE = 48;\n/** Yield between batches so background indexing doesn't starve the session. */\nconst BATCH_YIELD_MS = 15;\n/** The Rust mock backend's model id — semantically meaningless, never index with it. */\nconst MOCK_MODEL_ID = \"mock-hash-v1\";\n\nexport type EmbsearchState =\n\t| { phase: \"idle\" }\n\t| { phase: \"skipped\"; reason: string }\n\t| { phase: \"indexing\"; done: number; total: number }\n\t| { phase: \"ready\"; chunkCount: number }\n\t| { phase: \"unavailable\"; reason: string };\n\nexport interface EmbsearchServiceOptions {\n\tcwd: string;\n\t/** Explicit binary path (settings override). Default: \"embsearch\" from PATH. */\n\tbinaryPath?: string;\n\t/** Minimum indexable bytes before indexing kicks in. */\n\tthresholdBytes: number;\n\t/** Progress callback for UI (footer / stderr lines). */\n\tonProgress?: (state: EmbsearchState) => void;\n}\n\nexport interface SemanticHit {\n\tpath: string;\n\tstartLine: number;\n\tendLine: number;\n\tscore: number;\n}\n\nexport class EmbsearchService {\n\tprivate readonly options: EmbsearchServiceOptions;\n\tprivate client: EmbSearchClient | undefined;\n\tprivate meta: IndexMeta | undefined;\n\tprivate state: EmbsearchState = { phase: \"idle\" };\n\tprivate disposed = false;\n\n\tconstructor(options: EmbsearchServiceOptions) {\n\t\tthis.options = options;\n\t}\n\n\tgetState(): EmbsearchState {\n\t\treturn this.state;\n\t}\n\n\t/** Semantic search is usable (index ready, or still building with partial data). */\n\tisAvailable(): boolean {\n\t\treturn this.state.phase === \"ready\" || this.state.phase === \"indexing\";\n\t}\n\n\tprivate setState(state: EmbsearchState): void {\n\t\tthis.state = state;\n\t\tthis.options.onProgress?.(state);\n\t}\n\n\tprivate async resolveBinary(): Promise<string | undefined> {\n\t\tif (this.options.binaryPath) {\n\t\t\treturn this.options.binaryPath;\n\t\t}\n\t\treturn await ensureTool(\"embsearch\", true);\n\t}\n\n\t/**\n\t * Scan, threshold-check, and (when needed) index in the background.\n\t * Resolves when indexing completes or the feature settles dormant.\n\t */\n\tasync start(signal?: AbortSignal): Promise<void> {\n\t\ttry {\n\t\t\tawait this.run(signal);\n\t\t} catch (e) {\n\t\t\tconst reason = e instanceof Error ? e.message : String(e);\n\t\t\tthis.setState({ phase: \"unavailable\", reason });\n\t\t\tawait this.closeClient();\n\t\t}\n\t}\n\n\tprivate async run(signal?: AbortSignal): Promise<void> {\n\t\tconst binary = await this.resolveBinary();\n\t\tif (!binary) {\n\t\t\tthis.setState({\n\t\t\t\tphase: \"unavailable\",\n\t\t\t\treason: \"embsearch binary not found (PATH or embsearchBinaryPath setting)\",\n\t\t\t});\n\t\t\treturn;\n\t\t}\n\n\t\tconst scan = scanRepo(this.options.cwd, signal);\n\t\tif (scan.totalBytes < this.options.thresholdBytes) {\n\t\t\tthis.setState({\n\t\t\t\tphase: \"skipped\",\n\t\t\t\treason: `repo under threshold (${scan.totalBytes} < ${this.options.thresholdBytes} bytes)`,\n\t\t\t});\n\t\t\treturn;\n\t\t}\n\n\t\tconst storeDir = getEmbsearchStoreDir(this.options.cwd);\n\t\tthis.client = new EmbSearchClient({ binaryPath: binary, storePath: getVectorStoreDir(storeDir) });\n\t\tawait this.client.ready();\n\n\t\tconst info = await this.client.info();\n\t\tif (info.modelId === MOCK_MODEL_ID) {\n\t\t\tthrow new Error(\"embsearch binary uses the mock embedder (not semantic); install an onnx build\");\n\t\t}\n\n\t\t// Missing/stale sidecar (format, chunker, or model changed) → clean rebuild.\n\t\tthis.meta = loadIndexMeta(storeDir, info.modelId) ?? emptyIndexMeta(this.options.cwd, info.modelId);\n\t\tthis.meta.lastUsedMs = Date.now();\n\n\t\tawait this.indexChangedFiles(scan.files, storeDir, signal);\n\t}\n\n\tprivate async indexChangedFiles(files: RepoScanFile[], storeDir: string, signal?: AbortSignal): Promise<void> {\n\t\tconst meta = this.meta!;\n\t\tconst client = this.client!;\n\n\t\t// Diff scan vs sidecar: cheap mtime+size check first, hash only on delta.\n\t\tconst toIndex: Array<{ file: RepoScanFile; content: string; hash: string }> = [];\n\t\tconst seen = new Set<string>();\n\t\tfor (const file of files) {\n\t\t\tseen.add(file.rel);\n\t\t\tconst known = meta.files[file.rel];\n\t\t\tif (known && known.mtimeMs === file.mtimeMs && known.size === file.size) continue;\n\t\t\tlet content: string;\n\t\t\ttry {\n\t\t\t\tcontent = readFileSync(file.abs, \"utf-8\");\n\t\t\t} catch {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tconst hash = hashContent(content);\n\t\t\tif (known && known.hash === hash) {\n\t\t\t\t// Touched but unchanged — refresh stat info only.\n\t\t\t\tknown.mtimeMs = file.mtimeMs;\n\t\t\t\tknown.size = file.size;\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\ttoIndex.push({ file, content, hash });\n\t\t}\n\t\tconst toRemove = Object.keys(meta.files).filter((rel) => !seen.has(rel));\n\n\t\tif (toIndex.length === 0 && toRemove.length === 0) {\n\t\t\tsaveIndexMeta(storeDir, meta);\n\t\t\tthis.setState({ phase: \"ready\", chunkCount: this.countChunks(meta) });\n\t\t\treturn;\n\t\t}\n\n\t\t// Chunk changed files; count total upserts for exact progress.\n\t\tconst work: Array<{ rel: string; fileMeta: FileMeta; chunks: Array<{ id: string; text: string }> }> = [];\n\t\tlet totalChunks = 0;\n\t\tfor (const { file, content, hash } of toIndex) {\n\t\t\tconst chunks = chunkFile(file.rel, content);\n\t\t\twork.push({\n\t\t\t\trel: file.rel,\n\t\t\t\tfileMeta: {\n\t\t\t\t\tmtimeMs: file.mtimeMs,\n\t\t\t\t\tsize: file.size,\n\t\t\t\t\thash,\n\t\t\t\t\tchunks: chunks.map((c) => [c.startLine, c.endLine]),\n\t\t\t\t},\n\t\t\t\tchunks: chunks.map((c) => ({ id: c.id, text: c.text })),\n\t\t\t});\n\t\t\ttotalChunks += chunks.length;\n\t\t}\n\n\t\tthis.setState({ phase: \"indexing\", done: 0, total: totalChunks });\n\n\t\t// Drop vectors of deleted files and superseded chunk tails.\n\t\tfor (const rel of toRemove) {\n\t\t\tfor (let i = 0; i < meta.files[rel].chunks.length; i++) await client.remove(`${rel}#${i}`);\n\t\t\tdelete meta.files[rel];\n\t\t}\n\n\t\tlet done = 0;\n\t\tfor (const item of work) {\n\t\t\tif (signal?.aborted || this.disposed) return;\n\t\t\tconst oldChunkCount = meta.files[item.rel]?.chunks.length ?? 0;\n\t\t\t// Remove old chunks beyond the new count (upsert covers the rest).\n\t\t\tfor (let i = item.chunks.length; i < oldChunkCount; i++) await client.remove(`${item.rel}#${i}`);\n\n\t\t\tfor (let offset = 0; offset < item.chunks.length; offset += BULK_BATCH_SIZE) {\n\t\t\t\tif (signal?.aborted || this.disposed) return;\n\t\t\t\tconst batch = item.chunks.slice(offset, offset + BULK_BATCH_SIZE);\n\t\t\t\tawait client.bulk(batch);\n\t\t\t\tdone += batch.length;\n\t\t\t\tthis.setState({ phase: \"indexing\", done, total: totalChunks });\n\t\t\t\t// Yield so queries and the event loop stay responsive.\n\t\t\t\tawait new Promise((resolve) => setTimeout(resolve, BATCH_YIELD_MS));\n\t\t\t}\n\t\t\tmeta.files[item.rel] = item.fileMeta;\n\t\t}\n\n\t\tawait client.compact();\n\t\tawait client.save();\n\t\tsaveIndexMeta(storeDir, meta);\n\t\tthis.setState({ phase: \"ready\", chunkCount: this.countChunks(meta) });\n\t}\n\n\tprivate countChunks(meta: IndexMeta): number {\n\t\tlet n = 0;\n\t\tfor (const rel of Object.keys(meta.files)) n += meta.files[rel].chunks.length;\n\t\treturn n;\n\t}\n\n\t/** Top-`k` semantic hits as `path` + line range + score. */\n\tasync search(query: string, k = 10): Promise<SemanticHit[]> {\n\t\tif (!this.client || this.client.isClosed || !this.meta) {\n\t\t\tthrow new Error(\"semantic index is not available\");\n\t\t}\n\t\tconst results = await this.client.query(query, k);\n\t\tconst hits: SemanticHit[] = [];\n\t\tfor (const result of results) {\n\t\t\tconst sep = result.id.lastIndexOf(\"#\");\n\t\t\tif (sep === -1) continue;\n\t\t\tconst rel = result.id.slice(0, sep);\n\t\t\tconst chunkIndex = Number.parseInt(result.id.slice(sep + 1), 10);\n\t\t\tconst range = this.meta.files[rel]?.chunks[chunkIndex];\n\t\t\tif (!range) continue;\n\t\t\thits.push({ path: rel, startLine: range[0], endLine: range[1], score: result.score });\n\t\t}\n\t\treturn hits;\n\t}\n\n\tprivate async closeClient(): Promise<void> {\n\t\tconst client = this.client;\n\t\tthis.client = undefined;\n\t\tif (client && !client.isClosed) {\n\t\t\ttry {\n\t\t\t\tawait client.close();\n\t\t\t} catch {\n\t\t\t\t// already dead\n\t\t\t}\n\t\t}\n\t}\n\n\t/** Persist state and shut the daemon down. Safe to call twice. */\n\tasync dispose(): Promise<void> {\n\t\tif (this.disposed) return;\n\t\tthis.disposed = true;\n\t\tif (this.client && !this.client.isClosed) {\n\t\t\ttry {\n\t\t\t\tawait this.client.save();\n\t\t\t} catch {\n\t\t\t\t// daemon may have exited; nothing to save\n\t\t\t}\n\t\t}\n\t\tawait this.closeClient();\n\t}\n}\n\n// --- Per-cwd service registry ---\n//\n// The semantic_search tool is constructed by the generic tool factory table and\n// only receives `cwd`; the service is created later during session init (flag\n// gated). This registry connects the two without threading a service instance\n// through every layer between main.ts and the tool factories.\n\nconst services = new Map<string, EmbsearchService>();\n\nexport function registerEmbsearchService(cwd: string, service: EmbsearchService): void {\n\tconst old = services.get(cwd);\n\tif (old && old !== service) {\n\t\told.dispose().catch(() => {});\n\t}\n\tservices.set(cwd, service);\n}\n\nexport function getEmbsearchService(cwd: string): EmbsearchService | undefined {\n\treturn services.get(cwd);\n}\n\nexport function unregisterEmbsearchService(cwd: string): void {\n\tservices.delete(cwd);\n}\n"]}

package/dist/core/embsearch/index-meta.d.ts

@@ -0,0 +1,45 @@
+/**
+ * Sidecar metadata for a semantic index store.
+ *
+ * The Rust store (manifest.json/ids.json/vectors.bin) only knows vector ids;
+ * this sidecar (index-meta.json, written next to the store) tracks the source
+ * side: per-file freshness (mtime+size, content hash recomputed only when they
+ * change) and each file's chunk line-ranges so search hits can be rendered as
+ * `path:start-end`. It also pins the chunker version and embedding model id —
+ * a mismatch on either triggers a clean rebuild instead of an inconsistent
+ * incremental update.
+ */
+export interface FileMeta {
+    mtimeMs: number;
+    size: number;
+    /** SHA-256 of content, hex. Recomputed only when mtime/size differ. */
+    hash: string;
+    /** Per-chunk 1-based inclusive [start, end] line ranges; index = chunk number. */
+    chunks: Array<[number, number]>;
+}
+export interface IndexMeta {
+    formatVersion: number;
+    chunkerVersion: number;
+    modelId: string;
+    /** Absolute repo root this index was built from. */
+    repoRoot: string;
+    /** Last time this index was opened (ms). Enables later GC of dead stores. */
+    lastUsedMs: number;
+    files: Record<string, FileMeta>;
+}
+/** Directory holding the vector store + sidecar for `repoRoot`. */
+export declare function getEmbsearchStoreDir(repoRoot: string): string;
+/** Subdirectory the Rust store lives in (sidecar sits next to it). */
+export declare function getVectorStoreDir(storeDir: string): string;
+export declare function emptyIndexMeta(repoRoot: string, modelId: string): IndexMeta;
+/**
+ * Load the sidecar. Returns undefined when absent, unreadable, or built with a
+ * different format/chunker/model — callers treat all of those as "rebuild".
+ */
+export declare function loadIndexMeta(storeDir: string, modelId: string): IndexMeta | undefined;
+/** Atomically persist the sidecar (temp + rename, matching the Rust store). */
+export declare function saveIndexMeta(storeDir: string, meta: IndexMeta): void;
+export declare function hashContent(content: string): string;
+/** Display helper: `~/.hoocode/embsearch/<hash>` instead of the absolute path. */
+export declare function shortenStoreDir(storeDir: string): string;
+//# sourceMappingURL=index-meta.d.ts.map

package/dist/core/embsearch/index-meta.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index-meta.d.ts","sourceRoot":"","sources":["../../../src/core/embsearch/index-meta.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAYH,MAAM,WAAW,QAAQ;IACxB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,uEAAuE;IACvE,IAAI,EAAE,MAAM,CAAC;IACb,kFAAkF;IAClF,MAAM,EAAE,KAAK,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC;CAChC;AAED,MAAM,WAAW,SAAS;IACzB,aAAa,EAAE,MAAM,CAAC;IACtB,cAAc,EAAE,MAAM,CAAC;IACvB,OAAO,EAAE,MAAM,CAAC;IAChB,oDAAoD;IACpD,QAAQ,EAAE,MAAM,CAAC;IACjB,6EAA6E;IAC7E,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,QAAQ,CAAC,CAAC;CAChC;AAED,mEAAmE;AACnE,wBAAgB,oBAAoB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAI7D;AAED,sEAAsE;AACtE,wBAAgB,iBAAiB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAE1D;AAED,wBAAgB,cAAc,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,SAAS,CAS3E;AAED;;;GAGG;AACH,wBAAgB,aAAa,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,SAAS,GAAG,SAAS,CAiBtF;AAED,+EAA+E;AAC/E,wBAAgB,aAAa,CAAC,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,SAAS,GAAG,IAAI,CAMrE;AAED,wBAAgB,WAAW,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAEnD;AAED,kFAAkF;AAClF,wBAAgB,eAAe,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAGxD","sourcesContent":["/**\n * Sidecar metadata for a semantic index store.\n *\n * The Rust store (manifest.json/ids.json/vectors.bin) only knows vector ids;\n * this sidecar (index-meta.json, written next to the store) tracks the source\n * side: per-file freshness (mtime+size, content hash recomputed only when they\n * change) and each file's chunk line-ranges so search hits can be rendered as\n * `path:start-end`. It also pins the chunker version and embedding model id —\n * a mismatch on either triggers a clean rebuild instead of an inconsistent\n * incremental update.\n */\n\nimport { createHash } from \"crypto\";\nimport { existsSync, mkdirSync, readFileSync, renameSync, writeFileSync } from \"fs\";\nimport { homedir } from \"os\";\nimport path, { join } from \"path\";\nimport { getAgentDir } from \"../../config.js\";\nimport { CHUNKER_VERSION } from \"./chunker.js\";\n\nconst META_FILE = \"index-meta.json\";\nconst META_FORMAT_VERSION = 1;\n\nexport interface FileMeta {\n\tmtimeMs: number;\n\tsize: number;\n\t/** SHA-256 of content, hex. Recomputed only when mtime/size differ. */\n\thash: string;\n\t/** Per-chunk 1-based inclusive [start, end] line ranges; index = chunk number. */\n\tchunks: Array<[number, number]>;\n}\n\nexport interface IndexMeta {\n\tformatVersion: number;\n\tchunkerVersion: number;\n\tmodelId: string;\n\t/** Absolute repo root this index was built from. */\n\trepoRoot: string;\n\t/** Last time this index was opened (ms). Enables later GC of dead stores. */\n\tlastUsedMs: number;\n\tfiles: Record<string, FileMeta>;\n}\n\n/** Directory holding the vector store + sidecar for `repoRoot`. */\nexport function getEmbsearchStoreDir(repoRoot: string): string {\n\tconst resolved = path.resolve(repoRoot);\n\tconst hash = createHash(\"sha256\").update(resolved).digest(\"hex\").slice(0, 16);\n\treturn join(getAgentDir(), \"embsearch\", hash);\n}\n\n/** Subdirectory the Rust store lives in (sidecar sits next to it). */\nexport function getVectorStoreDir(storeDir: string): string {\n\treturn join(storeDir, \"store\");\n}\n\nexport function emptyIndexMeta(repoRoot: string, modelId: string): IndexMeta {\n\treturn {\n\t\tformatVersion: META_FORMAT_VERSION,\n\t\tchunkerVersion: CHUNKER_VERSION,\n\t\tmodelId,\n\t\trepoRoot: path.resolve(repoRoot),\n\t\tlastUsedMs: Date.now(),\n\t\tfiles: {},\n\t};\n}\n\n/**\n * Load the sidecar. Returns undefined when absent, unreadable, or built with a\n * different format/chunker/model — callers treat all of those as \"rebuild\".\n */\nexport function loadIndexMeta(storeDir: string, modelId: string): IndexMeta | undefined {\n\tconst file = join(storeDir, META_FILE);\n\tif (!existsSync(file)) return undefined;\n\tlet meta: IndexMeta;\n\ttry {\n\t\tmeta = JSON.parse(readFileSync(file, \"utf-8\")) as IndexMeta;\n\t} catch {\n\t\treturn undefined;\n\t}\n\tif (\n\t\tmeta.formatVersion !== META_FORMAT_VERSION ||\n\t\tmeta.chunkerVersion !== CHUNKER_VERSION ||\n\t\tmeta.modelId !== modelId\n\t) {\n\t\treturn undefined;\n\t}\n\treturn meta;\n}\n\n/** Atomically persist the sidecar (temp + rename, matching the Rust store). */\nexport function saveIndexMeta(storeDir: string, meta: IndexMeta): void {\n\tmkdirSync(storeDir, { recursive: true });\n\tconst file = join(storeDir, META_FILE);\n\tconst tmp = `${file}.tmp`;\n\twriteFileSync(tmp, JSON.stringify(meta));\n\trenameSync(tmp, file);\n}\n\nexport function hashContent(content: string): string {\n\treturn createHash(\"sha256\").update(content).digest(\"hex\");\n}\n\n/** Display helper: `~/.hoocode/embsearch/<hash>` instead of the absolute path. */\nexport function shortenStoreDir(storeDir: string): string {\n\tconst home = homedir();\n\treturn storeDir.startsWith(home) ? `~${storeDir.slice(home.length)}` : storeDir;\n}\n"]}

package/dist/core/embsearch/index-meta.js

@@ -0,0 +1,78 @@
+/**
+ * Sidecar metadata for a semantic index store.
+ *
+ * The Rust store (manifest.json/ids.json/vectors.bin) only knows vector ids;
+ * this sidecar (index-meta.json, written next to the store) tracks the source
+ * side: per-file freshness (mtime+size, content hash recomputed only when they
+ * change) and each file's chunk line-ranges so search hits can be rendered as
+ * `path:start-end`. It also pins the chunker version and embedding model id —
+ * a mismatch on either triggers a clean rebuild instead of an inconsistent
+ * incremental update.
+ */
+import { createHash } from "crypto";
+import { existsSync, mkdirSync, readFileSync, renameSync, writeFileSync } from "fs";
+import { homedir } from "os";
+import path, { join } from "path";
+import { getAgentDir } from "../../config.js";
+import { CHUNKER_VERSION } from "./chunker.js";
+const META_FILE = "index-meta.json";
+const META_FORMAT_VERSION = 1;
+/** Directory holding the vector store + sidecar for `repoRoot`. */
+export function getEmbsearchStoreDir(repoRoot) {
+    const resolved = path.resolve(repoRoot);
+    const hash = createHash("sha256").update(resolved).digest("hex").slice(0, 16);
+    return join(getAgentDir(), "embsearch", hash);
+}
+/** Subdirectory the Rust store lives in (sidecar sits next to it). */
+export function getVectorStoreDir(storeDir) {
+    return join(storeDir, "store");
+}
+export function emptyIndexMeta(repoRoot, modelId) {
+    return {
+        formatVersion: META_FORMAT_VERSION,
+        chunkerVersion: CHUNKER_VERSION,
+        modelId,
+        repoRoot: path.resolve(repoRoot),
+        lastUsedMs: Date.now(),
+        files: {},
+    };
+}
+/**
+ * Load the sidecar. Returns undefined when absent, unreadable, or built with a
+ * different format/chunker/model — callers treat all of those as "rebuild".
+ */
+export function loadIndexMeta(storeDir, modelId) {
+    const file = join(storeDir, META_FILE);
+    if (!existsSync(file))
+        return undefined;
+    let meta;
+    try {
+        meta = JSON.parse(readFileSync(file, "utf-8"));
+    }
+    catch {
+        return undefined;
+    }
+    if (meta.formatVersion !== META_FORMAT_VERSION ||
+        meta.chunkerVersion !== CHUNKER_VERSION ||
+        meta.modelId !== modelId) {
+        return undefined;
+    }
+    return meta;
+}
+/** Atomically persist the sidecar (temp + rename, matching the Rust store). */
+export function saveIndexMeta(storeDir, meta) {
+    mkdirSync(storeDir, { recursive: true });
+    const file = join(storeDir, META_FILE);
+    const tmp = `${file}.tmp`;
+    writeFileSync(tmp, JSON.stringify(meta));
+    renameSync(tmp, file);
+}
+export function hashContent(content) {
+    return createHash("sha256").update(content).digest("hex");
+}
+/** Display helper: `~/.hoocode/embsearch/<hash>` instead of the absolute path. */
+export function shortenStoreDir(storeDir) {
+    const home = homedir();
+    return storeDir.startsWith(home) ? `~${storeDir.slice(home.length)}` : storeDir;
+}
+//# sourceMappingURL=index-meta.js.map

package/dist/core/embsearch/index-meta.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index-meta.js","sourceRoot":"","sources":["../../../src/core/embsearch/index-meta.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EAAE,UAAU,EAAE,MAAM,QAAQ,CAAC;AACpC,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,YAAY,EAAE,UAAU,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AACpF,OAAO,EAAE,OAAO,EAAE,MAAM,IAAI,CAAC;AAC7B,OAAO,IAAI,EAAE,EAAE,IAAI,EAAE,MAAM,MAAM,CAAC;AAClC,OAAO,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAC;AAC9C,OAAO,EAAE,eAAe,EAAE,MAAM,cAAc,CAAC;AAE/C,MAAM,SAAS,GAAG,iBAAiB,CAAC;AACpC,MAAM,mBAAmB,GAAG,CAAC,CAAC;AAsB9B,mEAAmE;AACnE,MAAM,UAAU,oBAAoB,CAAC,QAAgB,EAAU;IAC9D,MAAM,QAAQ,GAAG,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;IACxC,MAAM,IAAI,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IAC9E,OAAO,IAAI,CAAC,WAAW,EAAE,EAAE,WAAW,EAAE,IAAI,CAAC,CAAC;AAAA,CAC9C;AAED,sEAAsE;AACtE,MAAM,UAAU,iBAAiB,CAAC,QAAgB,EAAU;IAC3D,OAAO,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;AAAA,CAC/B;AAED,MAAM,UAAU,cAAc,CAAC,QAAgB,EAAE,OAAe,EAAa;IAC5E,OAAO;QACN,aAAa,EAAE,mBAAmB;QAClC,cAAc,EAAE,eAAe;QAC/B,OAAO;QACP,QAAQ,EAAE,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC;QAChC,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE;QACtB,KAAK,EAAE,EAAE;KACT,CAAC;AAAA,CACF;AAED;;;GAGG;AACH,MAAM,UAAU,aAAa,CAAC,QAAgB,EAAE,OAAe,EAAyB;IACvF,MAAM,IAAI,GAAG,IAAI,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;IACvC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;QAAE,OAAO,SAAS,CAAC;IACxC,IAAI,IAAe,CAAC;IACpB,IAAI,CAAC;QACJ,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,EAAE,OAAO,CAAC,CAAc,CAAC;IAC7D,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,IACC,IAAI,CAAC,aAAa,KAAK,mBAAmB;QAC1C,IAAI,CAAC,cAAc,KAAK,eAAe;QACvC,IAAI,CAAC,OAAO,KAAK,OAAO,EACvB,CAAC;QACF,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,OAAO,IAAI,CAAC;AAAA,CACZ;AAED,+EAA+E;AAC/E,MAAM,UAAU,aAAa,CAAC,QAAgB,EAAE,IAAe,EAAQ;IACtE,SAAS,CAAC,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACzC,MAAM,IAAI,GAAG,IAAI,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;IACvC,MAAM,GAAG,GAAG,GAAG,IAAI,MAAM,CAAC;IAC1B,aAAa,CAAC,GAAG,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC;IACzC,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;AAAA,CACtB;AAED,MAAM,UAAU,WAAW,CAAC,OAAe,EAAU;IACpD,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;AAAA,CAC1D;AAED,kFAAkF;AAClF,MAAM,UAAU,eAAe,CAAC,QAAgB,EAAU;IACzD,MAAM,IAAI,GAAG,OAAO,EAAE,CAAC;IACvB,OAAO,QAAQ,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,QAAQ,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC;AAAA,CAChF","sourcesContent":["/**\n * Sidecar metadata for a semantic index store.\n *\n * The Rust store (manifest.json/ids.json/vectors.bin) only knows vector ids;\n * this sidecar (index-meta.json, written next to the store) tracks the source\n * side: per-file freshness (mtime+size, content hash recomputed only when they\n * change) and each file's chunk line-ranges so search hits can be rendered as\n * `path:start-end`. It also pins the chunker version and embedding model id —\n * a mismatch on either triggers a clean rebuild instead of an inconsistent\n * incremental update.\n */\n\nimport { createHash } from \"crypto\";\nimport { existsSync, mkdirSync, readFileSync, renameSync, writeFileSync } from \"fs\";\nimport { homedir } from \"os\";\nimport path, { join } from \"path\";\nimport { getAgentDir } from \"../../config.js\";\nimport { CHUNKER_VERSION } from \"./chunker.js\";\n\nconst META_FILE = \"index-meta.json\";\nconst META_FORMAT_VERSION = 1;\n\nexport interface FileMeta {\n\tmtimeMs: number;\n\tsize: number;\n\t/** SHA-256 of content, hex. Recomputed only when mtime/size differ. */\n\thash: string;\n\t/** Per-chunk 1-based inclusive [start, end] line ranges; index = chunk number. */\n\tchunks: Array<[number, number]>;\n}\n\nexport interface IndexMeta {\n\tformatVersion: number;\n\tchunkerVersion: number;\n\tmodelId: string;\n\t/** Absolute repo root this index was built from. */\n\trepoRoot: string;\n\t/** Last time this index was opened (ms). Enables later GC of dead stores. */\n\tlastUsedMs: number;\n\tfiles: Record<string, FileMeta>;\n}\n\n/** Directory holding the vector store + sidecar for `repoRoot`. */\nexport function getEmbsearchStoreDir(repoRoot: string): string {\n\tconst resolved = path.resolve(repoRoot);\n\tconst hash = createHash(\"sha256\").update(resolved).digest(\"hex\").slice(0, 16);\n\treturn join(getAgentDir(), \"embsearch\", hash);\n}\n\n/** Subdirectory the Rust store lives in (sidecar sits next to it). */\nexport function getVectorStoreDir(storeDir: string): string {\n\treturn join(storeDir, \"store\");\n}\n\nexport function emptyIndexMeta(repoRoot: string, modelId: string): IndexMeta {\n\treturn {\n\t\tformatVersion: META_FORMAT_VERSION,\n\t\tchunkerVersion: CHUNKER_VERSION,\n\t\tmodelId,\n\t\trepoRoot: path.resolve(repoRoot),\n\t\tlastUsedMs: Date.now(),\n\t\tfiles: {},\n\t};\n}\n\n/**\n * Load the sidecar. Returns undefined when absent, unreadable, or built with a\n * different format/chunker/model — callers treat all of those as \"rebuild\".\n */\nexport function loadIndexMeta(storeDir: string, modelId: string): IndexMeta | undefined {\n\tconst file = join(storeDir, META_FILE);\n\tif (!existsSync(file)) return undefined;\n\tlet meta: IndexMeta;\n\ttry {\n\t\tmeta = JSON.parse(readFileSync(file, \"utf-8\")) as IndexMeta;\n\t} catch {\n\t\treturn undefined;\n\t}\n\tif (\n\t\tmeta.formatVersion !== META_FORMAT_VERSION ||\n\t\tmeta.chunkerVersion !== CHUNKER_VERSION ||\n\t\tmeta.modelId !== modelId\n\t) {\n\t\treturn undefined;\n\t}\n\treturn meta;\n}\n\n/** Atomically persist the sidecar (temp + rename, matching the Rust store). */\nexport function saveIndexMeta(storeDir: string, meta: IndexMeta): void {\n\tmkdirSync(storeDir, { recursive: true });\n\tconst file = join(storeDir, META_FILE);\n\tconst tmp = `${file}.tmp`;\n\twriteFileSync(tmp, JSON.stringify(meta));\n\trenameSync(tmp, file);\n}\n\nexport function hashContent(content: string): string {\n\treturn createHash(\"sha256\").update(content).digest(\"hex\");\n}\n\n/** Display helper: `~/.hoocode/embsearch/<hash>` instead of the absolute path. */\nexport function shortenStoreDir(storeDir: string): string {\n\tconst home = homedir();\n\treturn storeDir.startsWith(home) ? `~${storeDir.slice(home.length)}` : storeDir;\n}\n"]}

package/dist/core/embsearch/repo-scan.d.ts

@@ -0,0 +1,26 @@
+/**
+ * Repository scan for semantic indexing: enumerates indexable source files
+ * using the same ignore-aware walk the native find/grep fallbacks use, and
+ * totals their bytes so the caller can apply the size threshold.
+ */
+export interface RepoScanFile {
+    /** Absolute path. */
+    abs: string;
+    /** POSIX path relative to the scan root. */
+    rel: string;
+    /** File size in bytes. */
+    size: number;
+    /** mtime in ms. */
+    mtimeMs: number;
+}
+export interface RepoScanResult {
+    files: RepoScanFile[];
+    totalBytes: number;
+}
+/**
+ * Enumerate indexable files under `root`: respects hierarchical `.gitignore`,
+ * always skips `.git`/`node_modules`, drops binary-ish extensions and files
+ * over 1MB. Returns files plus their byte total for threshold checks.
+ */
+export declare function scanRepo(root: string, signal?: AbortSignal): RepoScanResult;
+//# sourceMappingURL=repo-scan.d.ts.map

package/dist/core/embsearch/repo-scan.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"repo-scan.d.ts","sourceRoot":"","sources":["../../../src/core/embsearch/repo-scan.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AA0DH,MAAM,WAAW,YAAY;IAC5B,qBAAqB;IACrB,GAAG,EAAE,MAAM,CAAC;IACZ,4CAA4C;IAC5C,GAAG,EAAE,MAAM,CAAC;IACZ,0BAA0B;IAC1B,IAAI,EAAE,MAAM,CAAC;IACb,mBAAmB;IACnB,OAAO,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,cAAc;IAC9B,KAAK,EAAE,YAAY,EAAE,CAAC;IACtB,UAAU,EAAE,MAAM,CAAC;CACnB;AAUD;;;;GAIG;AACH,wBAAgB,QAAQ,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,cAAc,CAsB3E","sourcesContent":["/**\n * Repository scan for semantic indexing: enumerates indexable source files\n * using the same ignore-aware walk the native find/grep fallbacks use, and\n * totals their bytes so the caller can apply the size threshold.\n */\n\nimport { statSync } from \"fs\";\nimport path from \"path\";\nimport { collectEntries } from \"../tools/native-search.js\";\n\n/** Files larger than this are never indexed (vendored bundles, lockfiles, data dumps). */\nconst MAX_FILE_BYTES = 1024 * 1024;\n\n/** Files that are part of the ignore mechanism, not source. */\nconst SKIP_NAMES = new Set([\".gitignore\", \".gitattributes\", \".gitkeep\"]);\n\n/** Extensions that are never worth embedding. */\nconst SKIP_EXTENSIONS = new Set([\n\t\".png\",\n\t\".jpg\",\n\t\".jpeg\",\n\t\".gif\",\n\t\".webp\",\n\t\".ico\",\n\t\".bmp\",\n\t\".svg\",\n\t\".pdf\",\n\t\".zip\",\n\t\".gz\",\n\t\".tar\",\n\t\".bz2\",\n\t\".xz\",\n\t\".7z\",\n\t\".jar\",\n\t\".war\",\n\t\".class\",\n\t\".exe\",\n\t\".dll\",\n\t\".so\",\n\t\".dylib\",\n\t\".bin\",\n\t\".dat\",\n\t\".onnx\",\n\t\".pt\",\n\t\".safetensors\",\n\t\".woff\",\n\t\".woff2\",\n\t\".ttf\",\n\t\".otf\",\n\t\".eot\",\n\t\".mp3\",\n\t\".mp4\",\n\t\".wav\",\n\t\".avi\",\n\t\".mov\",\n\t\".webm\",\n\t\".lock\",\n\t\".min.js\",\n\t\".min.css\",\n\t\".map\",\n]);\n\nexport interface RepoScanFile {\n\t/** Absolute path. */\n\tabs: string;\n\t/** POSIX path relative to the scan root. */\n\trel: string;\n\t/** File size in bytes. */\n\tsize: number;\n\t/** mtime in ms. */\n\tmtimeMs: number;\n}\n\nexport interface RepoScanResult {\n\tfiles: RepoScanFile[];\n\ttotalBytes: number;\n}\n\nfunction hasSkippedExtension(rel: string): boolean {\n\tconst lower = rel.toLowerCase();\n\tfor (const ext of SKIP_EXTENSIONS) {\n\t\tif (lower.endsWith(ext)) return true;\n\t}\n\treturn false;\n}\n\n/**\n * Enumerate indexable files under `root`: respects hierarchical `.gitignore`,\n * always skips `.git`/`node_modules`, drops binary-ish extensions and files\n * over 1MB. Returns files plus their byte total for threshold checks.\n */\nexport function scanRepo(root: string, signal?: AbortSignal): RepoScanResult {\n\tconst entries = collectEntries(root, {\n\t\tsignal,\n\t\talwaysSkipDirs: new Set([\".git\", \"node_modules\"]),\n\t});\n\tconst files: RepoScanFile[] = [];\n\tlet totalBytes = 0;\n\tfor (const entry of entries) {\n\t\tif (entry.type !== \"f\") continue;\n\t\tif (SKIP_NAMES.has(path.basename(entry.rel))) continue;\n\t\tif (hasSkippedExtension(entry.rel)) continue;\n\t\tlet stat: ReturnType<typeof statSync>;\n\t\ttry {\n\t\t\tstat = statSync(entry.abs);\n\t\t} catch {\n\t\t\tcontinue;\n\t\t}\n\t\tif (stat.size === 0 || stat.size > MAX_FILE_BYTES) continue;\n\t\tfiles.push({ abs: entry.abs, rel: entry.rel, size: stat.size, mtimeMs: stat.mtimeMs });\n\t\ttotalBytes += stat.size;\n\t}\n\treturn { files, totalBytes };\n}\n"]}

package/dist/core/embsearch/repo-scan.js

@@ -0,0 +1,99 @@
+/**
+ * Repository scan for semantic indexing: enumerates indexable source files
+ * using the same ignore-aware walk the native find/grep fallbacks use, and
+ * totals their bytes so the caller can apply the size threshold.
+ */
+import { statSync } from "fs";
+import path from "path";
+import { collectEntries } from "../tools/native-search.js";
+/** Files larger than this are never indexed (vendored bundles, lockfiles, data dumps). */
+const MAX_FILE_BYTES = 1024 * 1024;
+/** Files that are part of the ignore mechanism, not source. */
+const SKIP_NAMES = new Set([".gitignore", ".gitattributes", ".gitkeep"]);
+/** Extensions that are never worth embedding. */
+const SKIP_EXTENSIONS = new Set([
+    ".png",
+    ".jpg",
+    ".jpeg",
+    ".gif",
+    ".webp",
+    ".ico",
+    ".bmp",
+    ".svg",
+    ".pdf",
+    ".zip",
+    ".gz",
+    ".tar",
+    ".bz2",
+    ".xz",
+    ".7z",
+    ".jar",
+    ".war",
+    ".class",
+    ".exe",
+    ".dll",
+    ".so",
+    ".dylib",
+    ".bin",
+    ".dat",
+    ".onnx",
+    ".pt",
+    ".safetensors",
+    ".woff",
+    ".woff2",
+    ".ttf",
+    ".otf",
+    ".eot",
+    ".mp3",
+    ".mp4",
+    ".wav",
+    ".avi",
+    ".mov",
+    ".webm",
+    ".lock",
+    ".min.js",
+    ".min.css",
+    ".map",
+]);
+function hasSkippedExtension(rel) {
+    const lower = rel.toLowerCase();
+    for (const ext of SKIP_EXTENSIONS) {
+        if (lower.endsWith(ext))
+            return true;
+    }
+    return false;
+}
+/**
+ * Enumerate indexable files under `root`: respects hierarchical `.gitignore`,
+ * always skips `.git`/`node_modules`, drops binary-ish extensions and files
+ * over 1MB. Returns files plus their byte total for threshold checks.
+ */
+export function scanRepo(root, signal) {
+    const entries = collectEntries(root, {
+        signal,
+        alwaysSkipDirs: new Set([".git", "node_modules"]),
+    });
+    const files = [];
+    let totalBytes = 0;
+    for (const entry of entries) {
+        if (entry.type !== "f")
+            continue;
+        if (SKIP_NAMES.has(path.basename(entry.rel)))
+            continue;
+        if (hasSkippedExtension(entry.rel))
+            continue;
+        let stat;
+        try {
+            stat = statSync(entry.abs);
+        }
+        catch {
+            continue;
+        }
+        if (stat.size === 0 || stat.size > MAX_FILE_BYTES)
+            continue;
+        files.push({ abs: entry.abs, rel: entry.rel, size: stat.size, mtimeMs: stat.mtimeMs });
+        totalBytes += stat.size;
+    }
+    return { files, totalBytes };
+}
+//# sourceMappingURL=repo-scan.js.map

package/dist/core/embsearch/repo-scan.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"repo-scan.js","sourceRoot":"","sources":["../../../src/core/embsearch/repo-scan.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAE,QAAQ,EAAE,MAAM,IAAI,CAAC;AAC9B,OAAO,IAAI,MAAM,MAAM,CAAC;AACxB,OAAO,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAE3D,0FAA0F;AAC1F,MAAM,cAAc,GAAG,IAAI,GAAG,IAAI,CAAC;AAEnC,+DAA+D;AAC/D,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC,CAAC,YAAY,EAAE,gBAAgB,EAAE,UAAU,CAAC,CAAC,CAAC;AAEzE,iDAAiD;AACjD,MAAM,eAAe,GAAG,IAAI,GAAG,CAAC;IAC/B,MAAM;IACN,MAAM;IACN,OAAO;IACP,MAAM;IACN,OAAO;IACP,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,KAAK;IACL,MAAM;IACN,MAAM;IACN,KAAK;IACL,KAAK;IACL,MAAM;IACN,MAAM;IACN,QAAQ;IACR,MAAM;IACN,MAAM;IACN,KAAK;IACL,QAAQ;IACR,MAAM;IACN,MAAM;IACN,OAAO;IACP,KAAK;IACL,cAAc;IACd,OAAO;IACP,QAAQ;IACR,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,MAAM;IACN,OAAO;IACP,OAAO;IACP,SAAS;IACT,UAAU;IACV,MAAM;CACN,CAAC,CAAC;AAkBH,SAAS,mBAAmB,CAAC,GAAW,EAAW;IAClD,MAAM,KAAK,GAAG,GAAG,CAAC,WAAW,EAAE,CAAC;IAChC,KAAK,MAAM,GAAG,IAAI,eAAe,EAAE,CAAC;QACnC,IAAI,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC;YAAE,OAAO,IAAI,CAAC;IACtC,CAAC;IACD,OAAO,KAAK,CAAC;AAAA,CACb;AAED;;;;GAIG;AACH,MAAM,UAAU,QAAQ,CAAC,IAAY,EAAE,MAAoB,EAAkB;IAC5E,MAAM,OAAO,GAAG,cAAc,CAAC,IAAI,EAAE;QACpC,MAAM;QACN,cAAc,EAAE,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;KACjD,CAAC,CAAC;IACH,MAAM,KAAK,GAAmB,EAAE,CAAC;IACjC,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC7B,IAAI,KAAK,CAAC,IAAI,KAAK,GAAG;YAAE,SAAS;QACjC,IAAI,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;YAAE,SAAS;QACvD,IAAI,mBAAmB,CAAC,KAAK,CAAC,GAAG,CAAC;YAAE,SAAS;QAC7C,IAAI,IAAiC,CAAC;QACtC,IAAI,CAAC;YACJ,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAC5B,CAAC;QAAC,MAAM,CAAC;YACR,SAAS;QACV,CAAC;QACD,IAAI,IAAI,CAAC,IAAI,KAAK,CAAC,IAAI,IAAI,CAAC,IAAI,GAAG,cAAc;YAAE,SAAS;QAC5D,KAAK,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,KAAK,CAAC,GAAG,EAAE,GAAG,EAAE,KAAK,CAAC,GAAG,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;QACvF,UAAU,IAAI,IAAI,CAAC,IAAI,CAAC;IACzB,CAAC;IACD,OAAO,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC;AAAA,CAC7B","sourcesContent":["/**\n * Repository scan for semantic indexing: enumerates indexable source files\n * using the same ignore-aware walk the native find/grep fallbacks use, and\n * totals their bytes so the caller can apply the size threshold.\n */\n\nimport { statSync } from \"fs\";\nimport path from \"path\";\nimport { collectEntries } from \"../tools/native-search.js\";\n\n/** Files larger than this are never indexed (vendored bundles, lockfiles, data dumps). */\nconst MAX_FILE_BYTES = 1024 * 1024;\n\n/** Files that are part of the ignore mechanism, not source. */\nconst SKIP_NAMES = new Set([\".gitignore\", \".gitattributes\", \".gitkeep\"]);\n\n/** Extensions that are never worth embedding. */\nconst SKIP_EXTENSIONS = new Set([\n\t\".png\",\n\t\".jpg\",\n\t\".jpeg\",\n\t\".gif\",\n\t\".webp\",\n\t\".ico\",\n\t\".bmp\",\n\t\".svg\",\n\t\".pdf\",\n\t\".zip\",\n\t\".gz\",\n\t\".tar\",\n\t\".bz2\",\n\t\".xz\",\n\t\".7z\",\n\t\".jar\",\n\t\".war\",\n\t\".class\",\n\t\".exe\",\n\t\".dll\",\n\t\".so\",\n\t\".dylib\",\n\t\".bin\",\n\t\".dat\",\n\t\".onnx\",\n\t\".pt\",\n\t\".safetensors\",\n\t\".woff\",\n\t\".woff2\",\n\t\".ttf\",\n\t\".otf\",\n\t\".eot\",\n\t\".mp3\",\n\t\".mp4\",\n\t\".wav\",\n\t\".avi\",\n\t\".mov\",\n\t\".webm\",\n\t\".lock\",\n\t\".min.js\",\n\t\".min.css\",\n\t\".map\",\n]);\n\nexport interface RepoScanFile {\n\t/** Absolute path. */\n\tabs: string;\n\t/** POSIX path relative to the scan root. */\n\trel: string;\n\t/** File size in bytes. */\n\tsize: number;\n\t/** mtime in ms. */\n\tmtimeMs: number;\n}\n\nexport interface RepoScanResult {\n\tfiles: RepoScanFile[];\n\ttotalBytes: number;\n}\n\nfunction hasSkippedExtension(rel: string): boolean {\n\tconst lower = rel.toLowerCase();\n\tfor (const ext of SKIP_EXTENSIONS) {\n\t\tif (lower.endsWith(ext)) return true;\n\t}\n\treturn false;\n}\n\n/**\n * Enumerate indexable files under `root`: respects hierarchical `.gitignore`,\n * always skips `.git`/`node_modules`, drops binary-ish extensions and files\n * over 1MB. Returns files plus their byte total for threshold checks.\n */\nexport function scanRepo(root: string, signal?: AbortSignal): RepoScanResult {\n\tconst entries = collectEntries(root, {\n\t\tsignal,\n\t\talwaysSkipDirs: new Set([\".git\", \"node_modules\"]),\n\t});\n\tconst files: RepoScanFile[] = [];\n\tlet totalBytes = 0;\n\tfor (const entry of entries) {\n\t\tif (entry.type !== \"f\") continue;\n\t\tif (SKIP_NAMES.has(path.basename(entry.rel))) continue;\n\t\tif (hasSkippedExtension(entry.rel)) continue;\n\t\tlet stat: ReturnType<typeof statSync>;\n\t\ttry {\n\t\t\tstat = statSync(entry.abs);\n\t\t} catch {\n\t\t\tcontinue;\n\t\t}\n\t\tif (stat.size === 0 || stat.size > MAX_FILE_BYTES) continue;\n\t\tfiles.push({ abs: entry.abs, rel: entry.rel, size: stat.size, mtimeMs: stat.mtimeMs });\n\t\ttotalBytes += stat.size;\n\t}\n\treturn { files, totalBytes };\n}\n"]}

package/dist/core/sdk.d.ts

@@ -70,6 +70,14 @@ export interface CreateAgentSessionOptions {
      * losslessly extract/edit structured/binary documents.
      */
     enableFileTools?: boolean;
+    /**
+     * Enable the built-in `semantic_search` tool (local embedding index via the
+     * `embsearch` binary). Defined but inactive by default. Ignored when an
+     * explicit `tools` allowlist is provided (list it there instead). The tool
+     * only returns results once the per-session index service reports available;
+     * otherwise it errors with a pointer back to grep/find.
+     */
+    enableEmbsearchTools?: boolean;
     /** Custom tools to register (in addition to built-in tools). */
     customTools?: ToolDefinition[];
     /**