@sanqianx/project-knowledge 5.0.23 → 5.0.25

package/_site/_test/helpers/parallel-retrieval-followup-r2-wb-host.js

@@ -0,0 +1,82 @@
+'use strict';
+// Lane-owned child process that hosts the REAL claude-ai-workbench module
+// (0.13.6, the copy the installed product ships) for the P-MEMORY-INJECTION-02
+// lane. It speaks the module's real HTTP contract and is started by the lane
+// probes.
+//
+// The shape of this child is a READ-ONLY SNAPSHOT of
+// _site/_test/helpers/parallel-memory-injection-wb-host.js (P-MEMORY-INJECTION-02
+// lane) which is itself a snapshot of _site/_test/helpers/parallel-workbench-host.js
+// (Workbench provenance lane). No source file is modified or imported: this
+// lane never depends on another lane's unfrozen file. The SHA-256 of every
+// source is recorded in the lane report.
+//
+// Environment (all set by the parent):
+//   PK_WB_MODULE_ROOT     the claude-ai-workbench package root to host
+//   PK_WB_HOST_PORT       port to listen on (127.0.0.1)
+//   PK_WB_HOST_DATA_DIR   isolated workbench data directory
+//   PK_WB_HOST_WORKSPACES JSON map of workspaceRef -> absolute directory
+//   PK_WB_ENGINE_URL      the knowledge engine URL (a recording proxy)
+//   PK_WB_PROFILE_SOURCE  none | real-readonly
+//   PK_WB_BRIDGE_URL      optional event-bridge console URL (enables capture)
+//   PK_WB_BRIDGE_TOKEN    optional event-bridge runtime token (memory only)
+const path = require('node:path');
+const fs = require('node:fs');
+
+const moduleRoot = String(process.env.PK_WB_MODULE_ROOT || '').trim();
+const port = Number(process.env.PK_WB_HOST_PORT || 0);
+const dataDir = String(process.env.PK_WB_HOST_DATA_DIR || '').trim();
+const workspaces = JSON.parse(String(process.env.PK_WB_HOST_WORKSPACES || '{}'));
+const engineUrl = String(process.env.PK_WB_ENGINE_URL || '').trim();
+const profileSource = String(process.env.PK_WB_PROFILE_SOURCE || 'none');
+const bridgeUrl = String(process.env.PK_WB_BRIDGE_URL || '').trim();
+const bridgeToken = String(process.env.PK_WB_BRIDGE_TOKEN || '').trim();
+
+if (!moduleRoot || !port || !dataDir || !Object.keys(workspaces).length) {
+  console.error('parallel-retrieval-followup-r2-wb-host: PK_WB_MODULE_ROOT, PK_WB_HOST_PORT, PK_WB_HOST_DATA_DIR and PK_WB_HOST_WORKSPACES are required');
+  process.exit(2);
+}
+
+// Same contract as the module's own bin: the runtime resolves its data
+// directory from these.
+process.env.CLAUDE_AI_WORKBENCH_DATA_DIR = dataDir;
+process.env.KB_DATA_DIR = dataDir;
+if (engineUrl) process.env.KB_KNOWLEDGE_ENGINE_URL = engineUrl;
+
+const { createWorkbenchServer } = require(path.join(moduleRoot, 'packages', 'server'));
+
+function buildOptions() {
+  const options = { dataDir, workspaces };
+  if (profileSource === 'real-readonly') {
+    const shell = require('./parallel-retrieval-followup-r2-shell');
+    options.profiles = shell.createReadOnlyProfileStore(moduleRoot, { dataDir });
+  }
+  if (bridgeUrl && bridgeToken) {
+    // The real Workbench capture: the module records the ORIGINAL user input
+    // (not the augmented prompt) and flushes it to the Bridge.
+    const { ConversationCapture } = require(path.join(moduleRoot, 'packages', 'server', 'lib', 'conversation-capture.js'));
+    options.conversationCapture = new ConversationCapture({
+      file: path.join(dataDir, 'capture-outbox.v1.json'),
+      bridgeUrl,
+      token: bridgeToken,
+    });
+  }
+  return options;
+}
+
+const app = createWorkbenchServer(buildOptions());
+app.server.listen(port, '127.0.0.1', () => {
+  process.stdout.write(`WORKBENCH_HOST_READY http://127.0.0.1:${port} engine=${engineUrl || 'none'} capture=${bridgeUrl ? 'on' : 'off'}\n`);
+});
+
+let stopping = false;
+async function shutdown() {
+  if (stopping) return;
+  stopping = true;
+  setTimeout(() => process.exit(1), 10000).unref();
+  try { await app.service.destroy(); } catch { process.exitCode = 1; }
+  app.server.closeAllConnections?.();
+  app.server.close(() => process.exit(process.exitCode || 0));
+}
+process.on('SIGINT', shutdown);
+process.on('SIGTERM', shutdown);

package/_site/_test/helpers/parallel-retrieval-followup-shell.js

@@ -0,0 +1,279 @@
+'use strict';
+// Lane-owned helpers for the P-RETRIEVAL-FOLLOWUP-01 batch (RET-007/RET-008).
+//
+// Portability contract: no developer path is hard-coded. The operator's
+// Workbench profile store defaults to the real user home captured BEFORE
+// fixture.isolateProcess() replaces HOME/USERPROFILE, and can be overridden
+// with PK_RETRIEVAL_WORKBENCH_HOME. Secrets (the glm subscription key) are
+// read in memory only: they are never printed, logged or persisted, and no
+// quota/usage endpoint is ever contacted by anything in this lane.
+//
+// Isolation contract: same as the sibling parallel-knowledge-fixture helper —
+// ambient module URLs cleared, live 5757/5790/8787/8790 endpoints refused,
+// every artefact under the external temp root. The shared fixture helper is
+// required read-only; this file only adds lane-specific composition.
+const fs = require('node:fs');
+const os = require('node:os');
+const path = require('node:path');
+const http = require('node:http');
+const { spawn, execFileSync } = require('node:child_process');
+
+const fixture = require('./parallel-knowledge-fixture');
+
+// Captured before isolateProcess() rewrites the environment.
+const REAL_HOME = process.env.PK_RETRIEVAL_WORKBENCH_HOME
+  || process.env.USERPROFILE
+  || process.env.HOME
+  || os.homedir();
+
+const REPO_ROOT = fixture.REPO_ROOT;
+const LANE_DIR = path.join(REPO_ROOT, 'qa-autopilot', 'artifacts', 'parallel-20261023', 'retrieval-followup');
+const BRIDGE_BIN = path.join(REPO_ROOT, 'node_modules', '@sanqianx', 'ai-coding-event-bridge-console', 'src', 'bin.js');
+const HUB_DIST = path.join(REPO_ROOT, 'node_modules', '@sanqianx', 'vector-hub', 'dist', 'cjs');
+const MODEL_RELATIVE = path.join('Xenova', 'bge-small-zh-v1.5');
+
+function workbenchProfileHome() { return path.join(REAL_HOME, '.agent-terminal'); }
+
+// Resolve one of the operator's real subscription profiles, in memory only.
+// Shape mirrors the Workbench ProfileStore runtime projection actually used
+// for engine runtime-config: { apiKey, baseUrl, mainModel }.
+//
+// The model list is the operator's instruction order, not a filter: the
+// provider is the authority on which id it accepts, so a model that the local
+// catalog does not list is still attempted and the answer is recorded.
+function loadRuntimeProfile(options = {}) {
+  const profileId = options.profileId || 'minimax';
+  const models = (options.models && options.models.length ? options.models : [options.model].filter(Boolean));
+  const home = workbenchProfileHome();
+  const profileFile = path.join(home, 'profiles.v2.json');
+  const credentialFile = path.join(home, 'credentials.v1.json');
+  const detail = { home, profileFile, credentialFile, profileId, models };
+  if (!models.length) return { ok: false, reason: 'no model requested', ...detail };
+  try {
+    const store = JSON.parse(fs.readFileSync(profileFile, 'utf8'));
+    const profiles = Array.isArray(store.profiles) ? store.profiles : [];
+    const profile = profiles.find(item => item.id === profileId);
+    if (!profile) return { ok: false, reason: `${profileId} profile not found`, ...detail };
+    if (profile.enabled === false) return { ok: false, reason: `${profileId} profile is disabled`, ...detail };
+    const credentials = JSON.parse(fs.readFileSync(credentialFile, 'utf8'));
+    const secret = credentials.credentials && credentials.credentials[profile.credentialRef];
+    if (typeof secret !== 'string' || !secret) return { ok: false, reason: `${profileId} credential is not configured`, ...detail };
+    if (!profile.baseUrl) return { ok: false, reason: `${profileId} profile has no baseUrl`, ...detail };
+    const listedModels = (profile.models && profile.models.list || []).map(entry => entry.id);
+    const mainModel = models[0];
+    // The secret never leaves memory; callers must not serialise this object.
+    return {
+      ok: true,
+      ...detail,
+      listedModels,
+      profile: { apiKey: secret, baseUrl: profile.baseUrl, mainModel },
+      identity: {
+        profileId: profile.id,
+        baseUrl: profile.baseUrl,
+        protocol: profile.protocol || null,
+        mainModel,
+        modelListedInProfile: listedModels.includes(mainModel),
+        credentialConfigured: true,
+      },
+    };
+  } catch (error) {
+    return { ok: false, reason: `workbench profile store unreadable: ${error.message}`, ...detail };
+  }
+}
+
+// The lane's offline model cache. A missing cache is a real BLOCKED, never a
+// download and never the shipped mock provider.
+function verifyModelCache(cacheDir) {
+  const modelDir = path.join(cacheDir, MODEL_RELATIVE);
+  const required = ['onnx' + path.sep + 'model.onnx', 'tokenizer.json'];
+  const missing = required.filter(relative => !fs.existsSync(path.join(modelDir, ...relative.split(/[\\/]/))));
+  return { cacheDir, modelDir, ok: missing.length === 0, missing };
+}
+
+// The product's own embedding service wrapped in the interface the real
+// vector hub expects (same contract as the accepted parallel-knowledge-vector
+// probe: batch = passage, single string = query).
+function createProductEmbedder(service) {
+  const state = { calls: 0, dimensions: [], nonZeroVectors: 0 };
+  return {
+    kind: 'product-local-embedding-service',
+    model: service.modelId,
+    modelId: service.modelId,
+    maxTokens: 500,
+    state,
+    async createEmbeddings(inputs) {
+      const isBatch = Array.isArray(inputs);
+      const texts = isBatch ? inputs : [inputs];
+      const output = [];
+      for (const text of texts) {
+        const vector = isBatch ? await service.embedPassage(text) : await service.embedQuery(text);
+        state.calls += 1;
+        state.dimensions.push(vector.length);
+        if (vector.some(value => value !== 0)) state.nonZeroVectors += 1;
+        output.push(vector);
+      }
+      return { status: 'success', output };
+    },
+  };
+}
+
+// A real vectra index on disk: the stored vectors must be real, of the
+// expected dimension and not all zero.
+function inspectStoredVectors(projectDir, expectedDimensions) {
+  const indexFile = path.join(projectDir, 'index.json');
+  if (!fs.existsSync(indexFile)) {
+    return { inspected: false, reason: 'no index.json in the project folder', files: fs.existsSync(projectDir) ? fs.readdirSync(projectDir) : [] };
+  }
+  const raw = fs.readFileSync(indexFile, 'utf8');
+  let parsed = null;
+  try { parsed = JSON.parse(raw); } catch { parsed = null; }
+  if (!parsed || !Array.isArray(parsed.items)) return { inspected: false, reason: 'index.json is not a readable JSON index', bytes: raw.length };
+  const vectors = parsed.items.map(item => item && item.vector).filter(Array.isArray);
+  const dimensions = [...new Set(vectors.map(vector => vector.length))];
+  return {
+    inspected: true,
+    bytes: raw.length,
+    itemCount: parsed.items.length,
+    vectorCount: vectors.length,
+    dimensions,
+    expectedDimensions,
+    dimensionsMatch: dimensions.length === 1 && dimensions[0] === expectedDimensions,
+    allVectorsNonZero: vectors.length > 0 && vectors.every(vector => vector.some(value => value !== 0)),
+  };
+}
+
+async function httpJson(url, options = {}) {
+  const response = await fetch(url, { signal: AbortSignal.timeout(options.timeoutMs || 8000), ...options });
+  const body = await response.json().catch(() => null);
+  return { status: response.status, ok: response.ok, body };
+}
+
+// The real event-bridge console as its own child process on an isolated port
+// and home — the production topology (hooks, console service and consumers
+// are separate processes over HTTP).
+async function spawnBridgeConsole({ home, port }) {
+  fs.mkdirSync(home, { recursive: true });
+  const child = spawn(process.execPath, [BRIDGE_BIN, 'serve', '--port', String(port), '--host', '127.0.0.1', '--home', home], {
+    env: fixture.sanitizedEnv({ BRIDGE_CONSOLE_PICK_FOLDER: '0', AI_CODING_EVENT_BRIDGE_HOME: home }),
+    windowsHide: true,
+    stdio: ['ignore', 'pipe', 'pipe'],
+  });
+  const output = { stdout: '', stderr: '' };
+  child.stdout.on('data', chunk => { output.stdout += chunk; });
+  child.stderr.on('data', chunk => { output.stderr += chunk; });
+  const url = `http://127.0.0.1:${port}`;
+  const deadline = Date.now() + 20000;
+  let lastError = null;
+  while (Date.now() < deadline) {
+    try {
+      const health = await httpJson(`${url}/api/health`, { timeoutMs: 2000 });
+      if (health.ok) return { child, url, home, output, stop: kill(child) };
+      lastError = new Error(`health HTTP ${health.status}`);
+    } catch (error) { lastError = error; }
+    if (child.exitCode !== null) throw new Error(`bridge console exited early (${child.exitCode}): ${output.stderr.slice(0, 400)}`);
+    await new Promise(resolve => setTimeout(resolve, 200));
+  }
+  kill(child)();
+  throw new Error(`bridge console did not become healthy: ${lastError && lastError.message}`);
+}
+
+function kill(child) {
+  return () => new Promise(resolve => {
+    if (child.exitCode !== null) return resolve();
+    child.once('close', () => resolve());
+    child.kill();
+    setTimeout(() => { if (child.exitCode === null) child.kill('SIGKILL'); resolve(); }, 3000).unref();
+  });
+}
+
+// The real pinned vector-hub package served over its real HTTP layer. The
+// product's LocalEmbeddingService is attached through the package's supported
+// hub.setEmbeddings API (the accepted probe pattern); the shell-managed
+// runtime-config credential path is not applicable to the offline local model
+// and is documented as the lane's injection boundary.
+async function startVectorHttpServer({ root, port, embedder, hubDist = HUB_DIST }) {
+  const { HubManager } = require(path.join(hubDist, 'server', 'manager.js'));
+  const { createServer } = require(path.join(hubDist, 'server', 'server.js'));
+  fs.mkdirSync(root, { recursive: true });
+  process.env.VECTOR_HUB_DISABLE_PICKER = '1';
+  const manager = await HubManager.load({ rootPath: root, configPath: path.join(root, 'vector-config.json'), managed: true });
+  manager.hub.setEmbeddings(embedder);
+  const server = http.createServer(createServer(manager));
+  await new Promise(resolve => server.listen(port, '127.0.0.1', resolve));
+  return {
+    manager,
+    server,
+    url: `http://127.0.0.1:${server.address().port}`,
+    root,
+    async stop() {
+      server.closeAllConnections();
+      await new Promise(resolve => server.close(resolve));
+      await manager.dispose?.();
+    },
+  };
+}
+
+// An isolated sink for the hook trigger's best-effort site notification. The
+// trigger falls back to KB_SITE_PORT (default 5757 — the developer's LIVE
+// site) when no runtime endpoint file exists, so every Git child in this lane
+// gets KB_SITE_PORT pointed here. Nothing may ever reach the real 5757.
+async function startNotificationSink() {
+  const hits = [];
+  const server = http.createServer((req, res) => {
+    let body = '';
+    req.on('data', chunk => { body += chunk; });
+    req.on('end', () => {
+      let parsed = null;
+      try { parsed = JSON.parse(body); } catch { parsed = null; }
+      hits.push({ method: req.method, url: req.url, body: parsed });
+      res.writeHead(404, { 'Content-Type': 'application/json' });
+      res.end(JSON.stringify({ error: 'notification sink' }));
+    });
+  });
+  await new Promise(resolve => server.listen(0, '127.0.0.1', resolve));
+  return { server, hits, port: server.address().port, stop: () => new Promise(resolve => { server.closeAllConnections(); server.close(resolve); }) };
+}
+
+// Git with an explicit environment (the hook inherits it: KB_SITE_PORT sinks
+// the notification, isolated vars keep the trigger away from live surfaces).
+function gitIn(repoPath, args, env = {}) {
+  return execFileSync('git', args, { cwd: repoPath, windowsHide: true, encoding: 'utf8', env: fixture.sanitizedEnv(env) }).trim();
+}
+
+async function waitFor(predicate, timeoutMs, label, intervalMs = 500) {
+  const deadline = Date.now() + timeoutMs;
+  let last = null;
+  while (Date.now() < deadline) {
+    last = await predicate();
+    if (last) return last;
+    await new Promise(resolve => setTimeout(resolve, intervalMs));
+  }
+  throw new Error(`Timed out after ${timeoutMs}ms waiting for ${label}`);
+}
+
+function writeEvidence(name, value) {
+  const file = path.join(LANE_DIR, 'evidence', name);
+  fs.mkdirSync(path.dirname(file), { recursive: true });
+  fs.writeFileSync(file, JSON.stringify(value, null, 2), 'utf8');
+  return { file, sha256: fixture.sha256File(file) };
+}
+
+module.exports = {
+  REAL_HOME,
+  REPO_ROOT,
+  LANE_DIR,
+  HUB_DIST,
+  workbenchProfileHome,
+  loadRuntimeProfile,
+  verifyModelCache,
+  createProductEmbedder,
+  inspectStoredVectors,
+  httpJson,
+  spawnBridgeConsole,
+  startVectorHttpServer,
+  startNotificationSink,
+  gitIn,
+  waitFor,
+  writeEvidence,
+  ...fixture,
+};