aiwcli 0.12.6 → 0.12.8

package/dist/templates/cc-native/_cc-native/lib-ts/rlm/embedding-indexer.ts

@@ -1,287 +1,287 @@
-#!/usr/bin/env bun
-/**
- * Embedding Indexer — Builds vector index from existing JSON session indexes.
- *
- * Reads ~/.claude/rlm-index/{project}/*.index.json (built by /rlm:index),
- * embeds each segment via Ollama nomic-embed-text, and stores vectors in
- * SQLite + sqlite-vec at ~/.claude/rlm-vectors.db.
- *
- * Usage:
- *   bun embedding-indexer.ts --batch                    # Index all sessions
- *   bun embedding-indexer.ts --batch --limit=10         # Index first 10 unindexed
- *   bun embedding-indexer.ts --batch --project=aiwcli   # Index matching project only
- *   bun embedding-indexer.ts --stats                    # Show index statistics
- */
-
-import { readdir } from "fs/promises";
-import { readFileSync, existsSync, readdirSync } from "fs";
-import { join } from "path";
-import { homedir } from "os";
-import { z } from "zod";
-import { RLM_INDEX_DIR, type SessionIndex } from "./types.js";
-import { logInfo, logWarn, logError, logDebug } from "./logger.js";
-import { checkOllamaHealth, embed } from "./ollama-client.js";
-import {
-  openVectorDb,
-  insertChunks,
-  markSessionEmbedded,
-  isSessionEmbedded,
-  deleteSessionChunks,
-  getStats,
-  type ChunkRow,
-} from "./vector-store.js";
-import { loadTranscript } from "./transcript-loader.js";
-
-const HOOK_NAME = "rlm_embed_idx";
-const MAX_EMBED_CHARS = 8000;
-
-// Zod schema for SessionIndex validation
-const SessionIndexSchema = z.object({
-  session_id: z.string(),
-  project: z.string(),
-  date: z.string(),
-  source_mtime: z.number(),
-  segments: z.array(z.object({
-    lines: z.tuple([z.number(), z.number()]),
-    topic: z.string(),
-    keywords: z.array(z.string()),
-  })),
-});
-type ValidatedSessionIndex = z.infer<typeof SessionIndexSchema>;
-
-// ---------------------------------------------------------------------------
-// CLI entry
-// ---------------------------------------------------------------------------
-
-const args = process.argv.slice(2);
-const isBatch = args.includes("--batch");
-const isStats = args.includes("--stats");
-const limitArg = args.find((a) => a.startsWith("--limit="));
-const limit = limitArg ? parseInt(limitArg.split("=")[1], 10) : Infinity;
-const projectArg = args.find((a) => a.startsWith("--project="));
-const projectFilter = projectArg ? projectArg.split("=")[1] : null;
-
-if (isStats) {
-  showStats();
-} else if (isBatch) {
-  runBatch().catch((e) => {
-    logError(HOOK_NAME, `Fatal: ${e}`, { stderr: true });
-    process.exitCode = 1;
-  });
-} else {
-  process.stderr.write(
-    "Usage: bun embedding-indexer.ts --batch [--limit=N] [--project=name]\n" +
-    "       bun embedding-indexer.ts --stats\n",
-  );
-  process.exitCode = 1;
-}
-
-// ---------------------------------------------------------------------------
-// Stats
-// ---------------------------------------------------------------------------
-
-function showStats(): void {
-  const db = openVectorDb();
-  const stats = getStats(db);
-  const result = {
-    sessions: stats.session_count,
-    chunks: stats.chunk_count,
-    db_path: db.filename,
-  };
-  process.stdout.write(JSON.stringify(result, null, 2) + "\n");
-  db.close();
-}
-
-// ---------------------------------------------------------------------------
-// Batch runner
-// ---------------------------------------------------------------------------
-
-async function runBatch(): Promise<void> {
-  // Check Ollama health first
-  const health = await checkOllamaHealth();
-  if (!health.ok) {
-    logError(HOOK_NAME, health.error ?? "Unknown Ollama health check error", { stderr: true });
-    process.exitCode = 1;
-    return;
-  }
-
-  // Verify rlm-index exists
-  if (!existsSync(RLM_INDEX_DIR)) {
-    logError(
-      HOOK_NAME,
-      `No JSON indexes found at ${RLM_INDEX_DIR}. Run \`/rlm:index\` first to build keyword indexes, then re-run \`/rlm:embed-index\`.`,
-      { stderr: true },
-    );
-    process.exitCode = 1;
-    return;
-  }
-
-  const db = openVectorDb();
-  let embedded = 0;
-  let skipped = 0;
-  let errors = 0;
-  let total = 0;
-
-  try {
-    // Scan project directories
-    const projectDirs = await readdir(RLM_INDEX_DIR, { withFileTypes: true });
-    const projects = projectDirs
-      .filter((d) => d.isDirectory())
-      .map((d) => d.name)
-      .filter((name) => !projectFilter || name.includes(projectFilter));
-
-    for (const project of projects) {
-      const projectDir = join(RLM_INDEX_DIR, project);
-      const files = await readdir(projectDir);
-      const indexFiles = files.filter((f) => f.endsWith(".index.json"));
-
-      for (const indexFile of indexFiles) {
-        if (embedded >= limit) break;
-        total++;
-
-        const indexPath = join(projectDir, indexFile);
-
-        try {
-          const rawJson = JSON.parse(readFileSync(indexPath, "utf-8"));
-          const parseResult = SessionIndexSchema.safeParse(rawJson);
-          if (!parseResult.success) {
-            errors++;
-            logWarn(HOOK_NAME, `Invalid index format ${indexFile}: ${parseResult.error.message}`);
-            continue;
-          }
-          const indexData = parseResult.data;
-
-          // Skip if already embedded at same mtime
-          if (
-            isSessionEmbedded(db, indexData.session_id, project, indexData.source_mtime)
-          ) {
-            skipped++;
-            continue;
-          }
-
-          // Re-index: delete old chunks if any
-          deleteSessionChunks(db, indexData.session_id, project);
-
-          // Build embedding texts for each segment
-          const texts: string[] = [];
-          const segmentMeta: Array<{
-            index: number;
-            lines: [number, number];
-            topic: string;
-          }> = [];
-
-          for (let i = 0; i < indexData.segments.length; i++) {
-            const seg = indexData.segments[i];
-
-            // Load transcript content for this segment
-            let content = "";
-            try {
-              const loaded = await loadTranscript(
-                join(
-                  // Derive source path from index data
-                  // source_path might be stored in the index, otherwise reconstruct
-                  getSourcePath(indexData, project),
-                ),
-                seg.lines,
-                4000,
-              );
-              content = loaded.content;
-            } catch {
-              // Fall back to topic + keywords as embedding text
-              content = `Topic: ${seg.topic}. Keywords: ${seg.keywords.join(", ")}`;
-            }
-
-            const embedText = `Project: ${project}. Date: ${indexData.date}. Topic: ${seg.topic}.\n${content}`;
-            texts.push(embedText.slice(0, MAX_EMBED_CHARS));
-            segmentMeta.push({ index: i, lines: seg.lines, topic: seg.topic });
-          }
-
-          if (texts.length === 0) {
-            logWarn(HOOK_NAME, `No segments for ${indexData.session_id}, skipping`);
-            skipped++;
-            continue;
-          }
-
-          // Embed all segments
-          const embeddings = await embed(texts);
-
-          // Build chunk rows
-          const chunks: ChunkRow[] = embeddings.map((emb, i) => ({
-            session_id: indexData.session_id,
-            project,
-            date: indexData.date,
-            segment_index: segmentMeta[i].index,
-            line_start: segmentMeta[i].lines[0],
-            line_end: segmentMeta[i].lines[1],
-            topic: segmentMeta[i].topic,
-            chunk_text: texts[i].slice(0, 2000),
-            source_path: getSourcePath(indexData, project),
-            embedding: emb,
-          }));
-
-          insertChunks(db, chunks);
-          markSessionEmbedded(
-            db,
-            indexData.session_id,
-            project,
-            indexData.source_mtime,
-            chunks.length,
-          );
-
-          embedded++;
-          if (embedded % 50 === 0) {
-            logInfo(HOOK_NAME, `Progress: ${embedded} sessions embedded`, { stderr: true });
-          }
-        } catch (e) {
-          errors++;
-          logWarn(HOOK_NAME, `Error embedding ${indexFile}: ${e}`);
-        }
-      }
-
-      if (embedded >= limit) break;
-    }
-  } finally {
-    db.close();
-  }
-
-  const result = { embedded, skipped, errors, total };
-  process.stdout.write(JSON.stringify(result) + "\n");
-  logInfo(HOOK_NAME, `Done: ${JSON.stringify(result)}`, { stderr: true });
-}
-
-// ---------------------------------------------------------------------------
-// Helpers
-// ---------------------------------------------------------------------------
-
-/**
- * Derive the source JSONL path from a ValidatedSessionIndex.
- * The index stores source_mtime but not always the full path.
- * Reconstruct from ~/.claude/projects/{project-slug}/{session_id}.jsonl
- */
-function getSourcePath(index: ValidatedSessionIndex, _project: string): string {
-  // SessionIndex doesn't have a source_path field, but the searcher derives it.
-  // The JSONL files live under ~/.claude/projects/{encoded-project-path}/
-  // We need to find the actual file. Use the index's session_id to locate it.
-  const claudeProjectsDir = join(homedir(), ".claude", "projects");
-
-  // Search for the session file across all project dirs
-  try {
-    const projectDirs = readdirSync(claudeProjectsDir, { withFileTypes: true });
-    for (const dir of projectDirs) {
-      if (!dir.isDirectory()) continue;
-      const candidatePath = join(
-        claudeProjectsDir,
-        dir.name,
-        `${index.session_id}.jsonl`,
-      );
-      if (existsSync(candidatePath)) {
-        return candidatePath;
-      }
-    }
-  } catch {
-    // Fall through
-  }
-
-  // Fallback: best guess
-  return join(claudeProjectsDir, _project, `${index.session_id}.jsonl`);
-}
+#!/usr/bin/env bun
+/**
+ * Embedding Indexer — Builds vector index from existing JSON session indexes.
+ *
+ * Reads ~/.claude/rlm-index/{project}/*.index.json (built by /rlm:index),
+ * embeds each segment via Ollama nomic-embed-text, and stores vectors in
+ * SQLite + sqlite-vec at ~/.claude/rlm-vectors.db.
+ *
+ * Usage:
+ *   bun embedding-indexer.ts --batch                    # Index all sessions
+ *   bun embedding-indexer.ts --batch --limit=10         # Index first 10 unindexed
+ *   bun embedding-indexer.ts --batch --project=aiwcli   # Index matching project only
+ *   bun embedding-indexer.ts --stats                    # Show index statistics
+ */
+
+import { readdir } from "fs/promises";
+import { readFileSync, existsSync, readdirSync } from "fs";
+import { join } from "path";
+import { homedir } from "os";
+import { z } from "zod";
+import { RLM_INDEX_DIR, type SessionIndex } from "./types.js";
+import { logInfo, logWarn, logError, logDebug } from "./logger.js";
+import { checkOllamaHealth, embed } from "./ollama-client.js";
+import {
+  openVectorDb,
+  insertChunks,
+  markSessionEmbedded,
+  isSessionEmbedded,
+  deleteSessionChunks,
+  getStats,
+  type ChunkRow,
+} from "./vector-store.js";
+import { loadTranscript } from "./transcript-loader.js";
+
+const HOOK_NAME = "rlm_embed_idx";
+const MAX_EMBED_CHARS = 8000;
+
+// Zod schema for SessionIndex validation
+const SessionIndexSchema = z.object({
+  session_id: z.string(),
+  project: z.string(),
+  date: z.string(),
+  source_mtime: z.number(),
+  segments: z.array(z.object({
+    lines: z.tuple([z.number(), z.number()]),
+    topic: z.string(),
+    keywords: z.array(z.string()),
+  })),
+});
+type ValidatedSessionIndex = z.infer<typeof SessionIndexSchema>;
+
+// ---------------------------------------------------------------------------
+// CLI entry
+// ---------------------------------------------------------------------------
+
+const args = process.argv.slice(2);
+const isBatch = args.includes("--batch");
+const isStats = args.includes("--stats");
+const limitArg = args.find((a) => a.startsWith("--limit="));
+const limit = limitArg ? parseInt(limitArg.split("=")[1], 10) : Infinity;
+const projectArg = args.find((a) => a.startsWith("--project="));
+const projectFilter = projectArg ? projectArg.split("=")[1] : null;
+
+if (isStats) {
+  showStats();
+} else if (isBatch) {
+  runBatch().catch((e) => {
+    logError(HOOK_NAME, `Fatal: ${e}`, { stderr: true });
+    process.exitCode = 1;
+  });
+} else {
+  process.stderr.write(
+    "Usage: bun embedding-indexer.ts --batch [--limit=N] [--project=name]\n" +
+    "       bun embedding-indexer.ts --stats\n",
+  );
+  process.exitCode = 1;
+}
+
+// ---------------------------------------------------------------------------
+// Stats
+// ---------------------------------------------------------------------------
+
+function showStats(): void {
+  const db = openVectorDb();
+  const stats = getStats(db);
+  const result = {
+    sessions: stats.session_count,
+    chunks: stats.chunk_count,
+    db_path: db.filename,
+  };
+  process.stdout.write(JSON.stringify(result, null, 2) + "\n");
+  db.close();
+}
+
+// ---------------------------------------------------------------------------
+// Batch runner
+// ---------------------------------------------------------------------------
+
+async function runBatch(): Promise<void> {
+  // Check Ollama health first
+  const health = await checkOllamaHealth();
+  if (!health.ok) {
+    logError(HOOK_NAME, health.error ?? "Unknown Ollama health check error", { stderr: true });
+    process.exitCode = 1;
+    return;
+  }
+
+  // Verify rlm-index exists
+  if (!existsSync(RLM_INDEX_DIR)) {
+    logError(
+      HOOK_NAME,
+      `No JSON indexes found at ${RLM_INDEX_DIR}. Run \`/rlm:index\` first to build keyword indexes, then re-run \`/rlm:embed-index\`.`,
+      { stderr: true },
+    );
+    process.exitCode = 1;
+    return;
+  }
+
+  const db = openVectorDb();
+  let embedded = 0;
+  let skipped = 0;
+  let errors = 0;
+  let total = 0;
+
+  try {
+    // Scan project directories
+    const projectDirs = await readdir(RLM_INDEX_DIR, { withFileTypes: true });
+    const projects = projectDirs
+      .filter((d) => d.isDirectory())
+      .map((d) => d.name)
+      .filter((name) => !projectFilter || name.includes(projectFilter));
+
+    for (const project of projects) {
+      const projectDir = join(RLM_INDEX_DIR, project);
+      const files = await readdir(projectDir);
+      const indexFiles = files.filter((f) => f.endsWith(".index.json"));
+
+      for (const indexFile of indexFiles) {
+        if (embedded >= limit) break;
+        total++;
+
+        const indexPath = join(projectDir, indexFile);
+
+        try {
+          const rawJson = JSON.parse(readFileSync(indexPath, "utf-8"));
+          const parseResult = SessionIndexSchema.safeParse(rawJson);
+          if (!parseResult.success) {
+            errors++;
+            logWarn(HOOK_NAME, `Invalid index format ${indexFile}: ${parseResult.error.message}`);
+            continue;
+          }
+          const indexData = parseResult.data;
+
+          // Skip if already embedded at same mtime
+          if (
+            isSessionEmbedded(db, indexData.session_id, project, indexData.source_mtime)
+          ) {
+            skipped++;
+            continue;
+          }
+
+          // Re-index: delete old chunks if any
+          deleteSessionChunks(db, indexData.session_id, project);
+
+          // Build embedding texts for each segment
+          const texts: string[] = [];
+          const segmentMeta: Array<{
+            index: number;
+            lines: [number, number];
+            topic: string;
+          }> = [];
+
+          for (let i = 0; i < indexData.segments.length; i++) {
+            const seg = indexData.segments[i];
+
+            // Load transcript content for this segment
+            let content = "";
+            try {
+              const loaded = await loadTranscript(
+                join(
+                  // Derive source path from index data
+                  // source_path might be stored in the index, otherwise reconstruct
+                  getSourcePath(indexData, project),
+                ),
+                seg.lines,
+                4000,
+              );
+              content = loaded.content;
+            } catch {
+              // Fall back to topic + keywords as embedding text
+              content = `Topic: ${seg.topic}. Keywords: ${seg.keywords.join(", ")}`;
+            }
+
+            const embedText = `Project: ${project}. Date: ${indexData.date}. Topic: ${seg.topic}.\n${content}`;
+            texts.push(embedText.slice(0, MAX_EMBED_CHARS));
+            segmentMeta.push({ index: i, lines: seg.lines, topic: seg.topic });
+          }
+
+          if (texts.length === 0) {
+            logWarn(HOOK_NAME, `No segments for ${indexData.session_id}, skipping`);
+            skipped++;
+            continue;
+          }
+
+          // Embed all segments
+          const embeddings = await embed(texts);
+
+          // Build chunk rows
+          const chunks: ChunkRow[] = embeddings.map((emb, i) => ({
+            session_id: indexData.session_id,
+            project,
+            date: indexData.date,
+            segment_index: segmentMeta[i].index,
+            line_start: segmentMeta[i].lines[0],
+            line_end: segmentMeta[i].lines[1],
+            topic: segmentMeta[i].topic,
+            chunk_text: texts[i].slice(0, 2000),
+            source_path: getSourcePath(indexData, project),
+            embedding: emb,
+          }));
+
+          insertChunks(db, chunks);
+          markSessionEmbedded(
+            db,
+            indexData.session_id,
+            project,
+            indexData.source_mtime,
+            chunks.length,
+          );
+
+          embedded++;
+          if (embedded % 50 === 0) {
+            logInfo(HOOK_NAME, `Progress: ${embedded} sessions embedded`, { stderr: true });
+          }
+        } catch (e) {
+          errors++;
+          logWarn(HOOK_NAME, `Error embedding ${indexFile}: ${e}`);
+        }
+      }
+
+      if (embedded >= limit) break;
+    }
+  } finally {
+    db.close();
+  }
+
+  const result = { embedded, skipped, errors, total };
+  process.stdout.write(JSON.stringify(result) + "\n");
+  logInfo(HOOK_NAME, `Done: ${JSON.stringify(result)}`, { stderr: true });
+}
+
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+
+/**
+ * Derive the source JSONL path from a ValidatedSessionIndex.
+ * The index stores source_mtime but not always the full path.
+ * Reconstruct from ~/.claude/projects/{project-slug}/{session_id}.jsonl
+ */
+function getSourcePath(index: ValidatedSessionIndex, _project: string): string {
+  // SessionIndex doesn't have a source_path field, but the searcher derives it.
+  // The JSONL files live under ~/.claude/projects/{encoded-project-path}/
+  // We need to find the actual file. Use the index's session_id to locate it.
+  const claudeProjectsDir = join(homedir(), ".claude", "projects");
+
+  // Search for the session file across all project dirs
+  try {
+    const projectDirs = readdirSync(claudeProjectsDir, { withFileTypes: true });
+    for (const dir of projectDirs) {
+      if (!dir.isDirectory()) continue;
+      const candidatePath = join(
+        claudeProjectsDir,
+        dir.name,
+        `${index.session_id}.jsonl`,
+      );
+      if (existsSync(candidatePath)) {
+        return candidatePath;
+      }
+    }
+  } catch {
+    // Fall through
+  }
+
+  // Fallback: best guess
+  return join(claudeProjectsDir, _project, `${index.session_id}.jsonl`);
+}