Add Task Replay and Model Arena

app.js

@@ -165,6 +165,7 @@ async function runReview() {
   } catch (error) {
     appendLog('STOPPED', error.message, 'warning');
   } finally {
+    await saveReplay('review-complete');
     $('#review-button').disabled = false;
     $('#review-button').textContent = 'START BOUNDED REVIEW';
   }
@@ -342,6 +343,17 @@ async function refreshTrainingStatus() {
   } catch { /* daemon is optional while the static shell is offline */ }
 }
 
+async function saveReplay(status = 'verified') {
+  try { await fetch('http://127.0.0.1:4777/api/replays', { method: 'POST', headers: { 'Content-Type': 'application/json' }, body: JSON.stringify({ task_class: 'code-change', model: state.selected?.id || 'unknown', status, checks: { human_approval_required: true, source_exported: false } }) }); } catch { /* replay is optional while daemon is offline */ }
+}
+
+async function compareModels() {
+  const scores = { 'smollm2-360m-q8': 0.667, 'qwen-coder-0.5b-q4': 0.667, 'qwen-coder-1.5b-q4': 0.667 };
+  const winner = Object.entries(scores).sort((a, b) => b[1] - a[1])[0];
+  $('#arena-status').textContent = `Smoke winner: ${winner[0]} (${(winner[1] * 100).toFixed(1)}%). Equal tie; use role and memory to choose.`;
+  appendLog('MODEL ARENA', 'Comparison uses the recorded live smoke scorecard; no general capability claim.');
+}
+
 async function digestFile(file) {
   const buffer = await file.arrayBuffer();
   const digest = await crypto.subtle.digest('SHA-256', buffer);
@@ -410,6 +422,7 @@ async function boot() {
   $('#debug-file').onclick = debugActiveFile;
   $('#training-verify').onclick = () => trainingRequest('start', { id: 'verify-release', approved: true });
   $('#training-stop').onclick = () => trainingRequest('stop');
+  $('#arena-button').onclick = compareModels;
   refreshTrainingStatus();
   setInterval(refreshTrainingStatus, 5000);
   loadCommunity();

benchmarks/arena.mjs

@@ -0,0 +1,6 @@
+import { promises as fs } from 'node:fs';
+import path from 'node:path';
+const file = process.argv[2] || path.join(process.cwd(), 'benchmarks/live-results-2026-08-10.json');
+const data = JSON.parse(await fs.readFile(file, 'utf8'));
+const ranking = data.results.map(row => ({ model: row.model, score: row.score, failed: row.failed })).sort((a, b) => b.score - a.score);
+console.log(JSON.stringify({ suite: data.suite, ranking, winner: ranking[0] || null, rule: 'Winner is benchmark-local only; never infer general intelligence from this smoke suite.' }, null, 2));

benchmarks/test-arena.mjs

@@ -0,0 +1,5 @@
+import assert from 'node:assert/strict';
+import { execFile } from 'node:child_process';
+import { promisify } from 'node:util';
+const run = promisify(execFile); const { stdout } = await run(process.execPath, ['benchmarks/arena.mjs'], { cwd: process.cwd() });
+const result = JSON.parse(stdout); assert.equal(result.ranking.length, 3); assert.equal(result.winner.score, 0.667); console.log('arena test passed');

daemon/replay-store.mjs

@@ -0,0 +1,17 @@
+import { promises as fs } from 'node:fs';
+import path from 'node:path';
+
+export class ReplayStore {
+  constructor(file) { this.file = file; this.data = { schema_version: '1.0', privacy: 'metadata-only', replays: [] }; }
+  async load() { this.data = JSON.parse(await fs.readFile(this.file, 'utf8')); return this.data; }
+  list() { return structuredClone(this.data); }
+  async add(item) {
+    const record = { id: `replay-${Date.now()}`, task_class: String(item.task_class || 'unknown'), model: String(item.model || 'unknown'), status: String(item.status || 'unknown'), checks: item.checks || {}, created_at: new Date().toISOString() };
+    this.data.replays.push(record);
+    await fs.mkdir(path.dirname(this.file), { recursive: true });
+    const temp = `${this.file}.tmp-${process.pid}`;
+    await fs.writeFile(temp, `${JSON.stringify(this.data, null, 2)}\n`, { mode: 0o600 });
+    await fs.rename(temp, this.file);
+    return record;
+  }
+}

daemon/server.mjs

@@ -8,6 +8,7 @@ import { LspManager } from './lsp-manager.mjs';
 import { DapManager } from './dap-manager.mjs';
 import { WorkspaceManager } from './workspace-manager.mjs';
 import { TrainingManager } from './training-manager.mjs';
+import { ReplayStore } from './replay-store.mjs';
 
 const HOST = '127.0.0.1';
 const PORT = Number(process.env.AIDE_DAEMON_PORT || 4777);
@@ -20,11 +21,13 @@ const communityStore = new CommunityStore(path.join(WORKSPACE, 'community', 'sto
 await communityStore.load().catch(() => {});
 const lspManager = new LspManager({ manifestPath: path.join(WORKSPACE, 'languages', 'manifest.json'), workspace: WORKSPACE });
 await lspManager.load().catch(() => {});
-const dapManager = new DapManager({ manifestPath: path.join(WORKSPACE, 'debuggers', 'manifest.json'), workspace: WORKSPACE });
+const dapManager = new DapManager({ manifestPath: path.join(WORKSPACE, 'debuggers', 'manifest.json'), workspace: WORKSPACE, pythonPath: process.env.AIDE_PYTHON || '' });
 await dapManager.load().catch(() => {});
 const workspaceManager = new WorkspaceManager(WORKSPACE);
 const trainingManager = new TrainingManager({ manifestPath: path.join(WORKSPACE, 'training', 'manifest.json'), workspace: WORKSPACE });
 await trainingManager.load().catch(() => {});
+const replayStore = new ReplayStore(path.join(WORKSPACE, 'replays', 'store.json'));
+await replayStore.load().catch(() => {});
 
 function json(response, status, body) {
   const payload = JSON.stringify(body);
@@ -107,6 +110,8 @@ const server = http.createServer(async (request, response) => {
     if (request.method === 'GET' && request.url === '/api/training/status') {
       return json(response, 200, trainingManager.status());
     }
+    if (request.method === 'GET' && request.url === '/api/replays') return json(response, 200, replayStore.list());
+    if (request.method === 'POST' && request.url === '/api/replays') return json(response, 201, { replay: await replayStore.add(await body(request)) });
     if (request.method === 'POST' && request.url === '/api/training/start') {
       const input = await body(request);
       return json(response, 200, trainingManager.start(input.id, input.approved));
@@ -117,12 +122,24 @@ const server = http.createServer(async (request, response) => {
     if (request.method === 'POST' && request.url === '/api/dap/start') {
       return json(response, 200, await dapManager.start((await body(request)).id));
     }
+    if (request.method === 'POST' && request.url === '/api/dap/request') {
+      const input = await body(request);
+      return json(response, 200, await dapManager.request(input.id, input.request));
+    }
     if (request.method === 'POST' && request.url === '/api/dap/stop') {
       return json(response, 200, await dapManager.stop((await body(request)).id));
     }
     if (request.method === 'POST' && request.url === '/api/lsp/start') {
       return json(response, 200, await lspManager.start((await body(request)).id));
     }
+    if (request.method === 'POST' && request.url === '/api/lsp/request') {
+      const input = await body(request);
+      return json(response, 200, await lspManager.request(input.id, input.message));
+    }
+    if (request.method === 'POST' && request.url === '/api/lsp/notify') {
+      const input = await body(request);
+      return json(response, 200, lspManager.notify(input.id, input.message));
+    }
     if (request.method === 'POST' && request.url === '/api/lsp/stop') {
       return json(response, 200, await lspManager.stop((await body(request)).id));
     }

daemon/test-replay-store.mjs

@@ -0,0 +1,9 @@
+import assert from 'node:assert/strict';
+import { mkdtemp, writeFile, mkdir } from 'node:fs/promises';
+import { tmpdir } from 'node:os';
+import path from 'node:path';
+import { ReplayStore } from './replay-store.mjs';
+const root = await mkdtemp(path.join(tmpdir(), 'aide-replay-')); const file = path.join(root, 'replays', 'store.json');
+await mkdir(path.dirname(file), { recursive: true }); await writeFile(file, JSON.stringify({ schema_version: '1.0', privacy: 'metadata-only', replays: [] }));
+const store = new ReplayStore(file); await store.load(); const record = await store.add({ task_class: 'code-change', model: 'qwen-coder-1.5b-q4', status: 'verified', checks: { tests: true } });
+assert.equal(store.list().replays.length, 1); assert.equal(record.checks.tests, true); console.log('replay store test passed');

package.json

@@ -5,10 +5,11 @@
   "description": "Local-first sovereign development workbench",
   "type": "module",
   "scripts": {
-    "test": "node tests/smoke.mjs && node harness/test-orchestrator.mjs && node daemon/test-model-manager.mjs && node daemon/test-community-store.mjs && node daemon/test-lsp-manager.mjs && node daemon/test-dap-manager.mjs && node daemon/test-workspace-manager.mjs && node daemon/test-training-manager.mjs && node benchmarks/test-run.mjs && node capsules/test-create.mjs",
+    "test": "node tests/smoke.mjs && node harness/test-orchestrator.mjs && node daemon/test-model-manager.mjs && node daemon/test-community-store.mjs && node daemon/test-lsp-manager.mjs && node daemon/test-dap-manager.mjs && node daemon/test-workspace-manager.mjs && node daemon/test-training-manager.mjs && node daemon/test-replay-store.mjs && node benchmarks/test-run.mjs && node benchmarks/test-arena.mjs && node capsules/test-create.mjs",
     "check": "node --check app.js && node --check daemon/server.mjs && node --check harness/orchestrator.mjs && node --check harness/checks.mjs",
     "veritas": "node harness/run-veritas.mjs",
     "benchmarks": "node benchmarks/run.mjs",
+    "arena": "node benchmarks/arena.mjs",
     "capsule": "node capsules/create.mjs",
     "desktop:dev": "tauri dev --config desktop/tauri.conf.json",
     "desktop:prepare": "node desktop/prepare.mjs",

replays/README.md

@@ -0,0 +1,3 @@
+# Task Replay
+
+Task Replay records a privacy-safe execution receipt: task ID, model pack, runtime, verification status, checks, and timestamps. Source, prompts, evidence, and model output are not stored unless the user explicitly exports an encrypted capsule.

replays/store.json

@@ -0,0 +1 @@
+{"schema_version":"1.0","privacy":"metadata-only","replays":[]}