@nusoft/nuos-build-catalogue 0.19.1 → 0.20.0

package/dist/cli.js

@@ -460,7 +460,17 @@ async function main() {
             // when the user switched machines and needs to pull the model
             // freshly. Same orchestrator that `init` calls internally.
             const { runLlmSetup } = await import('./setup/run-llm-setup.js');
+            const { ensureIndexBuilt } = await import('./setup/auto-index.js');
             const result = await runLlmSetup({ nonInteractive: false });
+            // After the LLM stack is ready, auto-build the search index when
+            // it isn't already present. Same helper init and install-protocols
+            // use — keeps the three commands aligned on "after this finishes
+            // the project is search-ready".
+            if (result.kind === 'already_ready' ||
+                result.kind === 'pulled_only' ||
+                result.kind === 'installed_and_pulled') {
+                await ensureIndexBuilt({});
+            }
             // Most failure paths emit guidance in-band; we exit non-zero only
             // when a pull actually failed (so CI scripting can branch on it).
             const exitCode = result.kind === 'pull_failed' || result.kind === 'install_failed' ? 1 : 0;

package/dist/commands/init.js

@@ -211,6 +211,7 @@ export async function cmdInit(prompt, options = {}) {
     // `nuos-catalogue setup-llm` later.
     if (!options.noLlm) {
         const { runLlmSetup } = await import('../setup/run-llm-setup.js');
+        const { ensureIndexBuilt } = await import('../setup/auto-index.js');
         await runLlmSetup({
             // The setup module writes its own progress directly to stderr; we
             // don't route through `prompt.print` because the in-place progress
@@ -224,6 +225,17 @@ export async function cmdInit(prompt, options = {}) {
             // so this is safe in unattended runs too.
             nonInteractive: false,
         });
+        // After LLM setup succeeds, auto-build the first search index. On a
+        // fresh project this is ~30s of starter-kit boilerplate; trivial,
+        // and finishing here means `search` works out of the box. When the
+        // LLM stack isn't ready, `ensureIndexBuilt` skips with a hint
+        // pointing back to setup-llm.
+        const indexResult = await ensureIndexBuilt({ cwd });
+        if (indexResult.kind === 'skipped_llm_not_ready') {
+            prompt.print('');
+            prompt.print(`  · Skipping first-index build: ${indexResult.reason}.`);
+            prompt.print(`  · ${indexResult.hint}`);
+        }
     }
     else {
         prompt.print('');
@@ -284,6 +296,16 @@ export async function cmdInstallProtocols(prompt, options = {}) {
     prompt.print('');
     prompt.print('Checking local semantic search (Ollama + qwen3-embedding:0.6b):');
     await reportLlmStatus((msg) => prompt.print(`  ${msg}`));
+    // Auto-build the search index when the LLM is ready but the index
+    // isn't present yet (typical upgrade-path state: pre-0.19 install +
+    // someone just added docs/build/ content). When the index already
+    // exists this is a no-op; when the LLM isn't ready the helper skips
+    // with a hint that's already printed by reportLlmStatus.
+    const { ensureIndexBuilt } = await import('../setup/auto-index.js');
+    const indexResult = await ensureIndexBuilt({ cwd });
+    if (indexResult.kind === 'just_built') {
+        prompt.print(`  ✓ Built search index (${indexResult.indexed} files, ${indexResult.chunks} chunks).`);
+    }
     return { output: '', exitCode: 0 };
 }
 /**

package/dist/setup/auto-index.d.ts

@@ -0,0 +1,54 @@
+/**
+ * Shared helper that runs the first search index build automatically
+ * from `init`, `install-protocols`, and `setup-llm`.
+ *
+ * Gated on the LLM stack being ready (Ollama + the configured embedding
+ * model). If the LLM isn't ready, this helper returns a `skipped_llm`
+ * result with a hint string the caller prints. The hint references
+ * `setup-llm` so the user has a clear path forward.
+ *
+ * Indexing on a fresh project takes ~30s — small enough that auto-
+ * running on first install is friendlier than asking. Subsequent calls
+ * are incremental via the per-file SHA hashes, so re-running on an
+ * existing index is cheap.
+ *
+ * @module setup/auto-index
+ */
+/** Outcome of an auto-index attempt. */
+export type AutoIndexResult = {
+    kind: 'already_built';
+    indexPath: string;
+} | {
+    kind: 'just_built';
+    indexPath: string;
+    indexed: number;
+    chunks: number;
+    durationMs: number;
+} | {
+    kind: 'skipped_llm_not_ready';
+    reason: string;
+    hint: string;
+} | {
+    kind: 'skipped_no_catalogue';
+} | {
+    kind: 'failed';
+    error: string;
+};
+export interface AutoIndexOptions {
+    /** Project root for path resolution. Defaults to `process.cwd()`. */
+    cwd?: string;
+    /** Output sink — defaults to process.stderr. */
+    out?: (text: string) => void;
+    /** Force a full reindex even if the index file already exists. */
+    force?: boolean;
+}
+/**
+ * Build the first search index when conditions allow. Idempotent: returns
+ * `already_built` and prints nothing when the index file exists (unless
+ * `force` is set). Returns `skipped_llm_not_ready` with a hint when the
+ * Ollama probe fails — the caller prints the hint and the user runs
+ * `setup-llm` to fix things.
+ *
+ * Never throws on user-facing failures.
+ */
+export declare function ensureIndexBuilt(opts?: AutoIndexOptions): Promise<AutoIndexResult>;

package/dist/setup/auto-index.js

@@ -0,0 +1,116 @@
+/**
+ * Shared helper that runs the first search index build automatically
+ * from `init`, `install-protocols`, and `setup-llm`.
+ *
+ * Gated on the LLM stack being ready (Ollama + the configured embedding
+ * model). If the LLM isn't ready, this helper returns a `skipped_llm`
+ * result with a hint string the caller prints. The hint references
+ * `setup-llm` so the user has a clear path forward.
+ *
+ * Indexing on a fresh project takes ~30s — small enough that auto-
+ * running on first install is friendlier than asking. Subsequent calls
+ * are incremental via the per-file SHA hashes, so re-running on an
+ * existing index is cheap.
+ *
+ * @module setup/auto-index
+ */
+import { existsSync } from 'node:fs';
+import { resolveBuildRoot, resolveCatalogueRoot, resolveHashPath, resolveIndexPath, } from '../path-resolution.js';
+import { DEFAULT_OLLAMA_HOST, detectModelPresent, detectOllamaApi } from './ollama-detect.js';
+import { DEFAULT_EMBEDDING_MODEL } from './run-llm-setup.js';
+/**
+ * Build the first search index when conditions allow. Idempotent: returns
+ * `already_built` and prints nothing when the index file exists (unless
+ * `force` is set). Returns `skipped_llm_not_ready` with a hint when the
+ * Ollama probe fails — the caller prints the hint and the user runs
+ * `setup-llm` to fix things.
+ *
+ * Never throws on user-facing failures.
+ */
+export async function ensureIndexBuilt(opts = {}) {
+    const cwd = opts.cwd ?? process.cwd();
+    const out = opts.out ?? ((text) => process.stderr.write(text));
+    // Resolve where the index file lives without forcing the LLM stack to
+    // load — path resolution is cheap and offline. When the project has
+    // no `docs/build/` yet (e.g. install-protocols invoked in a non-
+    // scaffolded directory), resolveBuildRoot throws — we treat that as a
+    // silent no-op, since there is nothing meaningful to index.
+    const ctx = { cwd, env: process.env };
+    let buildRoot;
+    let catalogueRoot;
+    let indexPath;
+    let hashPath;
+    try {
+        buildRoot = resolveBuildRoot(undefined, ctx);
+        catalogueRoot = resolveCatalogueRoot(undefined, ctx);
+        indexPath = resolveIndexPath(buildRoot, undefined, ctx);
+        hashPath = resolveHashPath(buildRoot, undefined, ctx);
+    }
+    catch {
+        return { kind: 'skipped_no_catalogue' };
+    }
+    // Fast path: index file already exists and we're not forcing a rebuild.
+    if (existsSync(indexPath) && !opts.force) {
+        return { kind: 'already_built', indexPath };
+    }
+    // Probe the LLM stack — index requires Ollama + the model. If either
+    // is missing, skip with a hint pointing at setup-llm.
+    const apiHost = process.env.NUOS_CATALOGUE_OLLAMA_HOST ?? DEFAULT_OLLAMA_HOST;
+    const modelId = process.env.NUOS_CATALOGUE_OLLAMA_MODEL ?? DEFAULT_EMBEDDING_MODEL;
+    const api = await detectOllamaApi(apiHost);
+    if (!api.reachable) {
+        return {
+            kind: 'skipped_llm_not_ready',
+            reason: `Ollama is not running at ${apiHost}`,
+            hint: 'Run `nuos-catalogue setup-llm` to set up local semantic search, then re-run `nuos-catalogue index`.',
+        };
+    }
+    const model = await detectModelPresent(apiHost, modelId);
+    if (!model.present) {
+        return {
+            kind: 'skipped_llm_not_ready',
+            reason: `${modelId} is not pulled`,
+            hint: 'Run `nuos-catalogue setup-llm` to pull the embedding model (~600 MB), then re-run `nuos-catalogue index`.',
+        };
+    }
+    // LLM is ready. Build the index. We import lazily so the cold-start
+    // path of `install-protocols` (where the index usually already exists)
+    // doesn't pay the embedder-loading cost.
+    out('Building search index for docs/build/ … (first run may take ~30 seconds)\n');
+    try {
+        const { selectEmbedderFromEnv } = await import('../embedder/select.js');
+        const { openStore } = await import('../store/open.js');
+        const { runIndex } = await import('../indexer/upsert.js');
+        const embedder = await selectEmbedderFromEnv();
+        const store = await openStore({ storagePath: indexPath, dimensions: embedder.dimensions });
+        try {
+            const report = await runIndex({
+                catalogueRoot,
+                hashFilePath: hashPath,
+                store,
+                embedder,
+                force: Boolean(opts.force),
+                dryRun: false,
+            });
+            out(`✓ Indexed ${report.indexed} file(s), ${report.chunks} chunks embedded in ` +
+                `${(report.durationMs / 1000).toFixed(1)}s\n`);
+            return {
+                kind: 'just_built',
+                indexPath,
+                indexed: report.indexed,
+                chunks: report.chunks,
+                durationMs: report.durationMs,
+            };
+        }
+        finally {
+            // Unload-after-use commitment — embedder releases the model.
+            await embedder.dispose();
+        }
+    }
+    catch (err) {
+        const message = err instanceof Error ? err.message : String(err);
+        out(`\n✗ Index build failed: ${message}\n`);
+        out('Re-run `nuos-catalogue index` manually to retry.\n');
+        return { kind: 'failed', error: message };
+    }
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nusoft/nuos-build-catalogue",
-  "version": "0.19.1",
+  "version": "0.20.0",
   "description": "NuOS build-catalogue tooling: semantic search (WU 110) + migration runner that lifts markdown artefacts into JSON-backed workflow records (WU 111, Phase G).",
   "type": "module",
   "bin": {