@castlemilk/omega 0.6.14 → 0.6.16

package/dist/server/dist/app.js

@@ -7,7 +7,7 @@ import { providerRoutes } from './routes/providers.js';
 import { routerRoutes } from './routes/router.js';
 import { metricsRoutes } from './routes/metrics.js';
 import { promptVersionRoutes } from './routes/prompt-versions.js';
-import { reportRoutes } from './routes/reports.js';
+import { benchmarkRoutes } from './routes/benchmarks.js';
 export const app = express();
 app.use(cors());
 app.use(express.json());
@@ -17,7 +17,7 @@ app.use('/providers', providerRoutes(prisma));
 app.use('/router', routerRoutes(prisma));
 app.use('/metrics', metricsRoutes(prisma));
 app.use('/prompt-versions', promptVersionRoutes(prisma));
-app.use('/reports', reportRoutes());
+app.use('/benchmarks', benchmarkRoutes(prisma));
 app.use((err, _req, res, _next) => {
     console.error(err);
     const message = err instanceof Error ? err.message : 'Internal error';

package/dist/server/dist/app.js.map

@@ -1 +1 @@
-{"version":3,"file":"app.js","sourceRoot":"","sources":["../src/app.ts"],"names":[],"mappings":"AAAA,OAAO,OAAO,MAAM,SAAS,CAAC;AAC9B,OAAO,IAAI,MAAM,MAAM,CAAC;AACxB,OAAO,EAAE,MAAM,EAAE,MAAM,WAAW,CAAC;AACnC,OAAO,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AACrD,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AACvD,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,mBAAmB,EAAE,MAAM,6BAA6B,CAAC;AAClE,OAAO,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AAEnD,MAAM,CAAC,MAAM,GAAG,GAAoB,OAAO,EAAE,CAAC;AAE9C,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC;AAChB,GAAG,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;AAExB,GAAG,CAAC,GAAG,CAAC,WAAW,EAAE,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC;AAC5C,GAAG,CAAC,GAAG,CAAC,QAAQ,EAAE,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC;AACtC,GAAG,CAAC,GAAG,CAAC,YAAY,EAAE,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC;AAC9C,GAAG,CAAC,GAAG,CAAC,SAAS,EAAE,YAAY,CAAC,MAAM,CAAC,CAAC,CAAC;AACzC,GAAG,CAAC,GAAG,CAAC,UAAU,EAAE,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC;AAC3C,GAAG,CAAC,GAAG,CAAC,kBAAkB,EAAE,mBAAmB,CAAC,MAAM,CAAC,CAAC,CAAC;AACzD,GAAG,CAAC,GAAG,CAAC,UAAU,EAAE,YAAY,EAAE,CAAC,CAAC;AAEpC,GAAG,CAAC,GAAG,CAAC,CAAC,GAAY,EAAE,IAAqB,EAAE,GAAqB,EAAE,KAA2B,EAAE,EAAE;IAClG,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IACnB,MAAM,OAAO,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,gBAAgB,CAAC;IACtE,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC,CAAC;AAC3C,CAAC,CAAC,CAAC"}
+{"version":3,"file":"app.js","sourceRoot":"","sources":["../src/app.ts"],"names":[],"mappings":"AAAA,OAAO,OAAO,MAAM,SAAS,CAAC;AAC9B,OAAO,IAAI,MAAM,MAAM,CAAC;AACxB,OAAO,EAAE,MAAM,EAAE,MAAM,WAAW,CAAC;AACnC,OAAO,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AACrD,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AACvD,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,mBAAmB,EAAE,MAAM,6BAA6B,CAAC;AAClE,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AAEzD,MAAM,CAAC,MAAM,GAAG,GAAoB,OAAO,EAAE,CAAC;AAE9C,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC;AAChB,GAAG,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;AAExB,GAAG,CAAC,GAAG,CAAC,WAAW,EAAE,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC;AAC5C,GAAG,CAAC,GAAG,CAAC,QAAQ,EAAE,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC;AACtC,GAAG,CAAC,GAAG,CAAC,YAAY,EAAE,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC;AAC9C,GAAG,CAAC,GAAG,CAAC,SAAS,EAAE,YAAY,CAAC,MAAM,CAAC,CAAC,CAAC;AACzC,GAAG,CAAC,GAAG,CAAC,UAAU,EAAE,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC;AAC3C,GAAG,CAAC,GAAG,CAAC,kBAAkB,EAAE,mBAAmB,CAAC,MAAM,CAAC,CAAC,CAAC;AACzD,GAAG,CAAC,GAAG,CAAC,aAAa,EAAE,eAAe,CAAC,MAAM,CAAC,CAAC,CAAC;AAEhD,GAAG,CAAC,GAAG,CAAC,CAAC,GAAY,EAAE,IAAqB,EAAE,GAAqB,EAAE,KAA2B,EAAE,EAAE;IAClG,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IACnB,MAAM,OAAO,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,gBAAgB,CAAC;IACtE,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC,CAAC;AAC3C,CAAC,CAAC,CAAC"}

package/dist/server/dist/routes/benchmarks.d.ts

@@ -0,0 +1,4 @@
+import { Router } from 'express';
+import type { PrismaClient } from '@omega/db';
+export declare function benchmarkRoutes(_prisma: PrismaClient): Router;
+//# sourceMappingURL=benchmarks.d.ts.map

package/dist/server/dist/routes/benchmarks.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"benchmarks.d.ts","sourceRoot":"","sources":["../../src/routes/benchmarks.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,WAAW,CAAC;AAyD9C,wBAAgB,eAAe,CAAC,OAAO,EAAE,YAAY,GAAG,MAAM,CA6F7D"}

package/dist/server/dist/routes/benchmarks.js

@@ -0,0 +1,131 @@
+import { Router } from 'express';
+import fs from 'node:fs/promises';
+import path from 'node:path';
+import { fileURLToPath } from 'node:url';
+import { spawn } from 'node:child_process';
+import { z } from 'zod';
+import { asyncHandler } from '../lib/async-handler.js';
+const __filename = fileURLToPath(import.meta.url);
+const root = path.resolve(__filename, '..', '..', '..', '..');
+const reportsDir = path.join(root, '.omega', 'reports');
+const statusFile = path.join(root, '.omega', 'bench-run-status.json');
+const runSchema = z.object({
+    suite: z.enum(['synthetic', 'deep-swe']).optional(),
+    nTasks: z.number().int().positive().optional(),
+    provider: z.string().optional(),
+    model: z.string().optional(),
+    timeout: z.number().int().positive().optional(),
+});
+async function readStatus() {
+    try {
+        const raw = await fs.readFile(statusFile, 'utf-8');
+        return JSON.parse(raw);
+    }
+    catch {
+        return { running: false };
+    }
+}
+async function writeStatus(status) {
+    await fs.mkdir(path.dirname(statusFile), { recursive: true });
+    await fs.writeFile(statusFile, JSON.stringify(status, null, 2), 'utf-8');
+}
+function isReportFile(name, prefix) {
+    return name.startsWith(prefix) && name.endsWith('.json');
+}
+async function listReports(prefix) {
+    try {
+        const files = await fs.readdir(reportsDir);
+        return files
+            .filter((f) => isReportFile(f, prefix))
+            .sort()
+            .reverse();
+    }
+    catch {
+        return [];
+    }
+}
+export function benchmarkRoutes(_prisma) {
+    const r = Router();
+    r.get('/reports', asyncHandler(async (_req, res) => {
+        const [benchmark, ab] = await Promise.all([
+            listReports('benchmark-'),
+            listReports('ab-'),
+        ]);
+        res.json({ benchmark, ab });
+    }));
+    r.get('/reports/:file', asyncHandler(async (req, res) => {
+        const file = req.params.file;
+        if (!isReportFile(file, 'benchmark-') && !isReportFile(file, 'ab-')) {
+            res.status(400).json({ error: 'Invalid report file' });
+            return;
+        }
+        const filePath = path.join(reportsDir, file);
+        const resolved = path.resolve(filePath);
+        const resolvedReportsDir = path.resolve(reportsDir);
+        if (!resolved.startsWith(resolvedReportsDir + path.sep) && resolved !== resolvedReportsDir) {
+            res.status(400).json({ error: 'Invalid report path' });
+            return;
+        }
+        try {
+            const raw = await fs.readFile(filePath, 'utf-8');
+            const data = JSON.parse(raw);
+            res.json(data);
+        }
+        catch {
+            res.status(404).json({ error: 'Report not found' });
+        }
+    }));
+    r.post('/run', asyncHandler(async (req, res) => {
+        const body = runSchema.parse(req.body);
+        const status = await readStatus();
+        if (status.running) {
+            res.status(409).json({ error: 'Benchmark run already in progress', pid: status.pid });
+            return;
+        }
+        const args = ['apps/cli/dist/index.js', 'bench', 'run'];
+        if (body.suite) {
+            args.push('--suite', body.suite);
+        }
+        if (body.nTasks) {
+            args.push('--n-tasks', String(body.nTasks));
+        }
+        if (body.timeout) {
+            args.push('--timeout', String(body.timeout));
+        }
+        if (body.provider) {
+            args.push('--provider', body.provider);
+        }
+        if (body.model) {
+            args.push('--model', body.model);
+        }
+        const child = spawn('node', args, {
+            cwd: root,
+            detached: true,
+            stdio: ['ignore', 'pipe', 'pipe'],
+        });
+        let output = '';
+        child.stdout.on('data', (chunk) => {
+            output += chunk.toString();
+        });
+        child.stderr.on('data', (chunk) => {
+            output += chunk.toString();
+        });
+        child.on('exit', () => {
+            void (async () => {
+                const current = await readStatus();
+                if (current.pid === child.pid) {
+                    await writeStatus({ running: false, output });
+                }
+            })();
+        });
+        child.unref();
+        await writeStatus({ running: true, pid: child.pid, output: '' });
+        res.status(202).json({ pid: child.pid, status: 'started' });
+    }));
+    r.get('/run-status', asyncHandler(async (_req, res) => {
+        const status = await readStatus();
+        res.json(status);
+    }));
+    return r;
+}
+//# sourceMappingURL=benchmarks.js.map

package/dist/server/dist/routes/benchmarks.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"benchmarks.js","sourceRoot":"","sources":["../../src/routes/benchmarks.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AAEjC,OAAO,EAAE,MAAM,kBAAkB,CAAC;AAClC,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AACzC,OAAO,EAAE,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAC3C,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,YAAY,EAAE,MAAM,yBAAyB,CAAC;AAEvD,MAAM,UAAU,GAAG,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;AAClD,MAAM,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC;AAC9D,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,QAAQ,EAAE,SAAS,CAAC,CAAC;AACxD,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,QAAQ,EAAE,uBAAuB,CAAC,CAAC;AAEtE,MAAM,SAAS,GAAG,CAAC,CAAC,MAAM,CAAC;IACzB,KAAK,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,WAAW,EAAE,UAAU,CAAC,CAAC,CAAC,QAAQ,EAAE;IACnD,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC9C,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC/B,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC5B,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;CAChD,CAAC,CAAC;AAQH,KAAK,UAAU,UAAU;IACvB,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,MAAM,EAAE,CAAC,QAAQ,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC;QACnD,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAc,CAAC;IACtC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;IAC5B,CAAC;AACH,CAAC;AAED,KAAK,UAAU,WAAW,CAAC,MAAiB;IAC1C,MAAM,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC9D,MAAM,EAAE,CAAC,SAAS,CAAC,UAAU,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;AAC3E,CAAC;AAED,SAAS,YAAY,CAAC,IAAY,EAAE,MAAc;IAChD,OAAO,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC,IAAI,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;AAC3D,CAAC;AAED,KAAK,UAAU,WAAW,CAAC,MAAc;IACvC,IAAI,CAAC;QACH,MAAM,KAAK,GAAG,MAAM,EAAE,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC;QAC3C,OAAO,KAAK;aACT,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,YAAY,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC;aACtC,IAAI,EAAE;aACN,OAAO,EAAE,CAAC;IACf,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,OAAqB;IACnD,MAAM,CAAC,GAAG,MAAM,EAAE,CAAC;IAEnB,CAAC,CAAC,GAAG,CAAC,UAAU,EAAE,YAAY,CAAC,KAAK,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QACjD,MAAM,CAAC,SAAS,EAAE,EAAE,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;YACxC,WAAW,CAAC,YAAY,CAAC;YACzB,WAAW,CAAC,KAAK,CAAC;SACnB,CAAC,CAAC;QACH,GAAG,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,EAAE,EAAE,CAAC,CAAC;IAC9B,CAAC,CAAC,CAAC,CAAC;IAEJ,CAAC,CAAC,GAAG,CAAC,gBAAgB,EAAE,YAAY,CAAC,KAAK,EAAE,GAAG,EAAE,GAAG,EAAE,EAAE;QACtD,MAAM,IAAI,GAAG,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC;QAC7B,IAAI,CAAC,YAAY,CAAC,IAAI,EAAE,YAAY,CAAC,IAAI,CAAC,YAAY,CAAC,IAAI,EAAE,KAAK,CAAC,EAAE,CAAC;YACpE,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,qBAAqB,EAAE,CAAC,CAAC;YACvD,OAAO;QACT,CAAC;QACD,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,IAAI,CAAC,CAAC;QAC7C,MAAM,QAAQ,GAAG,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;QACxC,MAAM,kBAAkB,GAAG,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC;QACpD,IAAI,CAAC,QAAQ,CAAC,UAAU,CAAC,kBAAkB,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,QAAQ,KAAK,kBAAkB,EAAE,CAAC;YAC3F,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,qBAAqB,EAAE,CAAC,CAAC;YACvD,OAAO;QACT,CAAC;QACD,IAAI,CAAC;YACH,MAAM,GAAG,GAAG,MAAM,EAAE,CAAC,QAAQ,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;YACjD,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAY,CAAC;YACxC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACjB,CAAC;QAAC,MAAM,CAAC;YACP,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,kBAAkB,EAAE,CAAC,CAAC;QACtD,CAAC;IACH,CAAC,CAAC,CAAC,CAAC;IAEJ,CAAC,CAAC,IAAI,CAAC,MAAM,EAAE,YAAY,CAAC,KAAK,EAAE,GAAG,EAAE,GAAG,EAAE,EAAE;QAC7C,MAAM,IAAI,GAAG,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACvC,MAAM,MAAM,GAAG,MAAM,UAAU,EAAE,CAAC;QAClC,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;YACnB,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,mCAAmC,EAAE,GAAG,EAAE,MAAM,CAAC,GAAG,EAAE,CAAC,CAAC;YACtF,OAAO;QACT,CAAC;QAED,MAAM,IAAI,GAAG,CAAC,wBAAwB,EAAE,OAAO,EAAE,KAAK,CAAC,CAAC;QACxD,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;YACf,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC;QACnC,CAAC;QACD,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;YAChB,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC;QAC9C,CAAC;QACD,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;YACjB,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC;QAC/C,CAAC;QACD,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;YAClB,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC;QACzC,CAAC;QACD,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;YACf,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC;QACnC,CAAC;QAED,MAAM,KAAK,GAAG,KAAK,CAAC,MAAM,EAAE,IAAI,EAAE;YAChC,GAAG,EAAE,IAAI;YACT,QAAQ,EAAE,IAAI;YACd,KAAK,EAAE,CAAC,QAAQ,EAAE,MAAM,EAAE,MAAM,CAAC;SAClC,CAAC,CAAC;QAEH,IAAI,MAAM,GAAG,EAAE,CAAC;QAChB,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,KAAa,EAAE,EAAE;YACxC,MAAM,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;QAC7B,CAAC,CAAC,CAAC;QACH,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,KAAa,EAAE,EAAE;YACxC,MAAM,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;QAC7B,CAAC,CAAC,CAAC;QAEH,KAAK,CAAC,EAAE,CAAC,MAAM,EAAE,GAAG,EAAE;YACpB,KAAK,CAAC,KAAK,IAAI,EAAE;gBACf,MAAM,OAAO,GAAG,MAAM,UAAU,EAAE,CAAC;gBACnC,IAAI,OAAO,CAAC,GAAG,KAAK,KAAK,CAAC,GAAG,EAAE,CAAC;oBAC9B,MAAM,WAAW,CAAC,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;gBAChD,CAAC;YACH,CAAC,CAAC,EAAE,CAAC;QACP,CAAC,CAAC,CAAC;QAEH,KAAK,CAAC,KAAK,EAAE,CAAC;QAEd,MAAM,WAAW,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,GAAG,EAAE,KAAK,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC,CAAC;QACjE,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,KAAK,CAAC,GAAG,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC9D,CAAC,CAAC,CAAC,CAAC;IAEJ,CAAC,CAAC,GAAG,CAAC,aAAa,EAAE,YAAY,CAAC,KAAK,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QACpD,MAAM,MAAM,GAAG,MAAM,UAAU,EAAE,CAAC;QAClC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IACnB,CAAC,CAAC,CAAC,CAAC;IAEJ,OAAO,CAAC,CAAC;AACX,CAAC"}

package/dist/server/dist/routes/prompt-versions.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"prompt-versions.d.ts","sourceRoot":"","sources":["../../src/routes/prompt-versions.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,WAAW,CAAC;AAc9C,wBAAgB,mBAAmB,CAAC,MAAM,EAAE,YAAY,GAAG,MAAM,CA8BhE"}
+{"version":3,"file":"prompt-versions.d.ts","sourceRoot":"","sources":["../../src/routes/prompt-versions.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,WAAW,CAAC;AAa9C,wBAAgB,mBAAmB,CAAC,MAAM,EAAE,YAAY,GAAG,MAAM,CA0BhE"}

package/dist/server/dist/routes/prompt-versions.js

@@ -1,38 +1,34 @@
 import { Router } from 'express';
 import { z } from 'zod';
 import { asyncHandler } from '../lib/async-handler.js';
-import { createPromptVersionRepository } from '@omega/db';
 const createSchema = z.object({
     name: z.string().min(1),
     sourcePath: z.string().min(1),
-    systemPrompt: z.string(),
-    textToolsPrompt: z.string(),
+    systemPrompt: z.string().min(1),
+    textToolsPrompt: z.string().min(1),
     hash: z.string().min(1),
     metadata: z.record(z.unknown()).optional(),
 });
 export function promptVersionRoutes(prisma) {
-    const repo = createPromptVersionRepository(prisma);
     const r = Router();
     r.get('/', asyncHandler(async (_req, res) => {
-        const versions = await repo.list();
+        const versions = await prisma.promptVersion.findMany({
+            orderBy: { createdAt: 'desc' },
+        });
         res.json(versions);
     }));
-    r.get('/:id', asyncHandler(async (req, res) => {
-        const version = await repo.findById(req.params.id);
-        if (!version) {
-            res.status(404).json({ error: 'Prompt version not found' });
-            return;
-        }
-        res.json(version);
-    }));
     r.post('/', asyncHandler(async (req, res) => {
         const body = createSchema.parse(req.body);
-        const existing = await repo.findByHash(body.hash);
-        if (existing) {
-            res.status(200).json(existing);
-            return;
-        }
-        const version = await repo.create(body);
+        const version = await prisma.promptVersion.create({
+            data: {
+                name: body.name,
+                sourcePath: body.sourcePath,
+                systemPrompt: body.systemPrompt,
+                textToolsPrompt: body.textToolsPrompt,
+                hash: body.hash,
+                metadata: body.metadata ? JSON.stringify(body.metadata) : null,
+            },
+        });
         res.status(201).json(version);
     }));
     return r;

package/dist/server/dist/routes/prompt-versions.js.map

@@ -1 +1 @@
-{"version":3,"file":"prompt-versions.js","sourceRoot":"","sources":["../../src/routes/prompt-versions.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AAEjC,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,YAAY,EAAE,MAAM,yBAAyB,CAAC;AACvD,OAAO,EAAE,6BAA6B,EAAE,MAAM,WAAW,CAAC;AAE1D,MAAM,YAAY,GAAG,CAAC,CAAC,MAAM,CAAC;IAC5B,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACvB,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC7B,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE;IACxB,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE;IAC3B,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACvB,QAAQ,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,QAAQ,EAAE;CAC3C,CAAC,CAAC;AAEH,MAAM,UAAU,mBAAmB,CAAC,MAAoB;IACtD,MAAM,IAAI,GAAG,6BAA6B,CAAC,MAAM,CAAC,CAAC;IACnD,MAAM,CAAC,GAAG,MAAM,EAAE,CAAC;IAEnB,CAAC,CAAC,GAAG,CAAC,GAAG,EAAE,YAAY,CAAC,KAAK,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAC1C,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,IAAI,EAAE,CAAC;QACnC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACrB,CAAC,CAAC,CAAC,CAAC;IAEJ,CAAC,CAAC,GAAG,CAAC,MAAM,EAAE,YAAY,CAAC,KAAK,EAAE,GAAG,EAAE,GAAG,EAAE,EAAE;QAC5C,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;QACnD,IAAI,CAAC,OAAO,EAAE,CAAC;YACb,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,0BAA0B,EAAE,CAAC,CAAC;YAC5D,OAAO;QACT,CAAC;QACD,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IACpB,CAAC,CAAC,CAAC,CAAC;IAEJ,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,YAAY,CAAC,KAAK,EAAE,GAAG,EAAE,GAAG,EAAE,EAAE;QAC1C,MAAM,IAAI,GAAG,YAAY,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QAC1C,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAClD,IAAI,QAAQ,EAAE,CAAC;YACb,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YAC/B,OAAO;QACT,CAAC;QACD,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;QACxC,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IAChC,CAAC,CAAC,CAAC,CAAC;IAEJ,OAAO,CAAC,CAAC;AACX,CAAC"}
+{"version":3,"file":"prompt-versions.js","sourceRoot":"","sources":["../../src/routes/prompt-versions.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AAEjC,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,YAAY,EAAE,MAAM,yBAAyB,CAAC;AAEvD,MAAM,YAAY,GAAG,CAAC,CAAC,MAAM,CAAC;IAC5B,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACvB,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC7B,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC/B,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAClC,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACvB,QAAQ,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,QAAQ,EAAE;CAC3C,CAAC,CAAC;AAEH,MAAM,UAAU,mBAAmB,CAAC,MAAoB;IACtD,MAAM,CAAC,GAAG,MAAM,EAAE,CAAC;IAEnB,CAAC,CAAC,GAAG,CAAC,GAAG,EAAE,YAAY,CAAC,KAAK,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAC1C,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,aAAa,CAAC,QAAQ,CAAC;YACnD,OAAO,EAAE,EAAE,SAAS,EAAE,MAAM,EAAE;SAC/B,CAAC,CAAC;QACH,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACrB,CAAC,CAAC,CAAC,CAAC;IAEJ,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,YAAY,CAAC,KAAK,EAAE,GAAG,EAAE,GAAG,EAAE,EAAE;QAC1C,MAAM,IAAI,GAAG,YAAY,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QAC1C,MAAM,OAAO,GAAG,MAAM,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC;YAChD,IAAI,EAAE;gBACJ,IAAI,EAAE,IAAI,CAAC,IAAI;gBACf,UAAU,EAAE,IAAI,CAAC,UAAU;gBAC3B,YAAY,EAAE,IAAI,CAAC,YAAY;gBAC/B,eAAe,EAAE,IAAI,CAAC,eAAe;gBACrC,IAAI,EAAE,IAAI,CAAC,IAAI;gBACf,QAAQ,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI;aAC/D;SACF,CAAC,CAAC;QACH,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IAChC,CAAC,CAAC,CAAC,CAAC;IAEJ,OAAO,CAAC,CAAC;AACX,CAAC"}

package/dist/server/node_modules/@omega/agent/dist/index.d.ts

@@ -2,7 +2,7 @@ export { runAgentTask, type AgentResult } from './executor.js';
 export { createPlan, type PlanStep, type PlannerResult } from './planner.js';
 export { validateProject, type ValidationSummary } from './validator.js';
 export { publishOmega, type PublishResult } from './publisher.js';
-export * from './prompt-versioning.js';
+export { loadCurrentPrompts, saveCurrentPrompts, readPromptsSource, hashPrompts, envForPrompts, type PromptVersionInput, } from './prompt-versioning.js';
 export * from './git.js';
 export * from './tools.js';
 //# sourceMappingURL=index.d.ts.map

package/dist/server/node_modules/@omega/agent/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,KAAK,WAAW,EAAE,MAAM,eAAe,CAAC;AAC/D,OAAO,EAAE,UAAU,EAAE,KAAK,QAAQ,EAAE,KAAK,aAAa,EAAE,MAAM,cAAc,CAAC;AAC7E,OAAO,EAAE,eAAe,EAAE,KAAK,iBAAiB,EAAE,MAAM,gBAAgB,CAAC;AACzE,OAAO,EAAE,YAAY,EAAE,KAAK,aAAa,EAAE,MAAM,gBAAgB,CAAC;AAClE,cAAc,wBAAwB,CAAC;AACvC,cAAc,UAAU,CAAC;AACzB,cAAc,YAAY,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,KAAK,WAAW,EAAE,MAAM,eAAe,CAAC;AAC/D,OAAO,EAAE,UAAU,EAAE,KAAK,QAAQ,EAAE,KAAK,aAAa,EAAE,MAAM,cAAc,CAAC;AAC7E,OAAO,EAAE,eAAe,EAAE,KAAK,iBAAiB,EAAE,MAAM,gBAAgB,CAAC;AACzE,OAAO,EAAE,YAAY,EAAE,KAAK,aAAa,EAAE,MAAM,gBAAgB,CAAC;AAClE,OAAO,EACL,kBAAkB,EAClB,kBAAkB,EAClB,iBAAiB,EACjB,WAAW,EACX,aAAa,EACb,KAAK,kBAAkB,GACxB,MAAM,wBAAwB,CAAC;AAChC,cAAc,UAAU,CAAC;AACzB,cAAc,YAAY,CAAC"}

package/dist/server/node_modules/@omega/agent/dist/index.js

@@ -2,7 +2,7 @@ export { runAgentTask } from './executor.js';
 export { createPlan } from './planner.js';
 export { validateProject } from './validator.js';
 export { publishOmega } from './publisher.js';
-export * from './prompt-versioning.js';
+export { loadCurrentPrompts, saveCurrentPrompts, readPromptsSource, hashPrompts, envForPrompts, } from './prompt-versioning.js';
 export * from './git.js';
 export * from './tools.js';
 //# sourceMappingURL=index.js.map

package/dist/server/node_modules/@omega/agent/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAoB,MAAM,eAAe,CAAC;AAC/D,OAAO,EAAE,UAAU,EAAqC,MAAM,cAAc,CAAC;AAC7E,OAAO,EAAE,eAAe,EAA0B,MAAM,gBAAgB,CAAC;AACzE,OAAO,EAAE,YAAY,EAAsB,MAAM,gBAAgB,CAAC;AAClE,cAAc,wBAAwB,CAAC;AACvC,cAAc,UAAU,CAAC;AACzB,cAAc,YAAY,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAoB,MAAM,eAAe,CAAC;AAC/D,OAAO,EAAE,UAAU,EAAqC,MAAM,cAAc,CAAC;AAC7E,OAAO,EAAE,eAAe,EAA0B,MAAM,gBAAgB,CAAC;AACzE,OAAO,EAAE,YAAY,EAAsB,MAAM,gBAAgB,CAAC;AAClE,OAAO,EACL,kBAAkB,EAClB,kBAAkB,EAClB,iBAAiB,EACjB,WAAW,EACX,aAAa,GAEd,MAAM,wBAAwB,CAAC;AAChC,cAAc,UAAU,CAAC;AACzB,cAAc,YAAY,CAAC"}

package/dist/server/node_modules/@omega/agent/dist/prompts.d.ts

@@ -1,5 +1,5 @@
 export declare const AGENT_SYSTEM_PROMPT: string;
-export declare const FORCE_ACTION_PROMPT = "You have been thinking without taking action. Stop describing plans and execute the next concrete step using a tool. Use edit_file or run_command.";
+export declare const FORCE_ACTION_PROMPT = "You have been thinking without taking action. Stop describing plans and execute the next concrete step using a tool. Use edit_file, publish, or run_command.";
 export declare const TEXT_TOOLS_SYSTEM_PROMPT: string;
 export declare function buildSystemPrompt(context?: string): string;
 export declare function buildTextToolsSystemPrompt(context?: string): string;

package/dist/server/node_modules/@omega/agent/dist/prompts.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"prompts.d.ts","sourceRoot":"","sources":["../src/prompts.ts"],"names":[],"mappings":"AAiBA,eAAO,MAAM,mBAAmB,QAgCsF,CAAC;AAEvH,eAAO,MAAM,mBAAmB,uJAAuJ,CAAC;AAExL,eAAO,MAAM,wBAAwB,QAkCjC,CAAC;AAEL,wBAAgB,iBAAiB,CAAC,OAAO,CAAC,EAAE,MAAM,GAAG,MAAM,CAG1D;AAED,wBAAgB,0BAA0B,CAAC,OAAO,CAAC,EAAE,MAAM,GAAG,MAAM,CAGnE;AAED,wBAAgB,eAAe,CAAC,KAAK,EAAE,MAAM,EAAE,WAAW,CAAC,EAAE,MAAM,GAAG,MAAM,CAK3E;AAED,wBAAgB,qBAAqB,CACnC,IAAI,EAAE;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,CAAA;CAAE,EAC7C,OAAO,EAAE;IAAE,UAAU,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,EAAE,GAChD,MAAM,CAOR"}
+{"version":3,"file":"prompts.d.ts","sourceRoot":"","sources":["../src/prompts.ts"],"names":[],"mappings":"AAiBA,eAAO,MAAM,mBAAmB,QAyC+D,CAAC;AAEhG,eAAO,MAAM,mBAAmB,iKAAiK,CAAC;AAElM,eAAO,MAAM,wBAAwB,QAmCjC,CAAC;AAEL,wBAAgB,iBAAiB,CAAC,OAAO,CAAC,EAAE,MAAM,GAAG,MAAM,CAG1D;AAED,wBAAgB,0BAA0B,CAAC,OAAO,CAAC,EAAE,MAAM,GAAG,MAAM,CAGnE;AAED,wBAAgB,eAAe,CAAC,KAAK,EAAE,MAAM,EAAE,WAAW,CAAC,EAAE,MAAM,GAAG,MAAM,CAK3E;AAED,wBAAgB,qBAAqB,CACnC,IAAI,EAAE;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,CAAA;CAAE,EAC7C,OAAO,EAAE;IAAE,UAAU,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,EAAE,GAChD,MAAM,CAOR"}

package/dist/server/node_modules/@omega/agent/dist/prompts.js

@@ -28,26 +28,35 @@ Available tools:
 - run_command: Run a single simple command. No pipes (|), &&, ;, redirects, or globs. Prefer pnpm/npm/node. Arguments: { "command": "pnpm lint" }
 - think: Record a reasoning step. Arguments: { "thought": "..." }
 - finish: Mark the task complete. Arguments: { "summary": "what was done", "success": true }. Use summary, not message.
-- publish: Run build/test/publish validation. Only after your edits pass validation. Arguments: { "version": "optional" }
+- publish: Run lint/test/build validation in one step. This is the preferred validation tool. Arguments: { "version": "optional" }
 
 Rules:
 1. Read the task, then use think to plan.
 2. Use edit_file for small changes; write_file only when creating a file or rewriting most of it.
 3. ALWAYS read a file with read_file before using edit_file on it. Do NOT read a file before write_file when creating a new file.
 4. Before creating JavaScript files, read package.json to check whether "type" is "module" and use ESM (export) or CommonJS (module.exports) accordingly.
-5. You MUST run validation (pnpm lint, pnpm test, or the project's validation command) after every edit using publish or run_command.
-6. Do NOT call finish until validation passes. If validation fails, fix the issue and re-run validation.
-7. Do not expose secrets or run destructive commands.
-8. Finish only when the task is done. Always include summary and success.
-9. Do not repeat the same tool call with the same arguments. If something fails, change your approach.
-10. Do not ask the user for clarification. Do not say "No task was provided". Use the Task and Description above to proceed.
+5. Immediately after making any edit, call publish to run validation. Do not read more files, do not think, and do not call finish before validating.
+6. If validation fails, fix the issue and call publish again. Repeat until validation passes.
+7. Do NOT call finish until validation passes. If you call finish without validation, the task will fail.
+8. Do not expose secrets or run destructive commands.
+9. Finish only when the task is done. Always include summary and success.
+10. Do not repeat the same tool call with the same arguments. If something fails, change your approach.
+11. Do not ask the user for clarification. Do not say "No task was provided". Use the Task and Description above to proceed.
+12. You have a limited number of steps. If you spend too many steps reading or planning without validating, the task will fail.
 
 Example tool call sequence for "Add a greet function":
 1. think: { "thought": "I need to create greet.js and export a greet function." }
-2. write_file: { "path": "greet.js", "content": "export function greet(name) { return 'Hello, ' + name + '!'; }\n" }
+2. write_file: { "path": "greet.js", "content": "export function greet(name) { return 'Hello, ' + name + '!'; }\\n" }
 3. publish: { }
-4. finish: { "summary": "Created greet.js with greet(name) function and verified it passes tests.", "success": true }`;
-export const FORCE_ACTION_PROMPT = `You have been thinking without taking action. Stop describing plans and execute the next concrete step using a tool. Use edit_file or run_command.`;
+4. finish: { "summary": "Created greet.js with greet(name) function and verified it passes tests.", "success": true }
+
+Example tool call sequence for "Fix a lint warning":
+1. think: { "thought": "The linter flags var. I need to read src/index.js and replace var with const or let." }
+2. read_file: { "path": "src/index.js" }
+3. edit_file: { "path": "src/index.js", "old_string": "var greeting = 'Hello';", "new_string": "const greeting = 'Hello';" }
+4. publish: { }
+5. finish: { "summary": "Replaced var with const and verified lint passes.", "success": true }`;
+export const FORCE_ACTION_PROMPT = `You have been thinking without taking action. Stop describing plans and execute the next concrete step using a tool. Use edit_file, publish, or run_command.`;
 export const TEXT_TOOLS_SYSTEM_PROMPT = loadPromptFromEnv('OMEGA_TEXT_TOOLS_PROMPT') ??
     `You are Omega, an autonomous software engineering agent running inside a project repository.
 
@@ -68,8 +77,9 @@ Rules:
 - Use edit_file for small changes; write_file only for new files or large rewrites.
 - ALWAYS read a file with read_file before using edit_file on it. Do NOT read a file before write_file when creating a new file.
 - Before creating JavaScript files, read package.json to check whether "type" is "module" and use ESM (export) or CommonJS (module.exports) accordingly.
-- You MUST run validation (pnpm lint, pnpm test, or the project's validation command) after every edit using publish or run_command.
-- Do NOT call finish until validation passes. If validation fails, fix the issue and re-run validation.
+- Immediately after making any edit, call publish to run validation. Do not read more files or call finish before validating.
+- If validation fails, fix the issue and call publish again.
+- Do NOT call finish until validation passes.
 - Do not expose secrets or run destructive commands.
 - Finish only when done. Use summary, not message.
 - Do not repeat the same tool call with the same arguments.
@@ -78,7 +88,7 @@ Rules:
 Example response for "Add a greet function":
 { "tool_calls": [
   { "id": "1", "name": "think", "arguments": { "thought": "Create greet.js with greet(name)." } },
-  { "id": "2", "name": "write_file", "arguments": { "path": "greet.js", "content": "export function greet(name) { return 'Hello, ' + name + '!'; }\n" } },
+  { "id": "2", "name": "write_file", "arguments": { "path": "greet.js", "content": "export function greet(name) { return 'Hello, ' + name + '!'; }\\n" } },
   { "id": "3", "name": "publish", "arguments": {} },
   { "id": "4", "name": "finish", "arguments": { "summary": "Created greet.js and verified tests pass.", "success": true } }
 ] }`;

package/dist/server/node_modules/@omega/agent/dist/prompts.js.map

@@ -1 +1 @@
-{"version":3,"file":"prompts.js","sourceRoot":"","sources":["../src/prompts.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,SAAS,CAAC;AAEzB,SAAS,iBAAiB,CAAC,IAAY;IACrC,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACjC,IAAI,MAAM;QAAE,OAAO,MAAM,CAAC;IAC1B,MAAM,OAAO,GAAG,GAAG,IAAI,OAAO,CAAC;IAC/B,MAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IACtC,IAAI,QAAQ,EAAE,CAAC;QACb,IAAI,CAAC;YACH,OAAO,EAAE,CAAC,YAAY,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;QAC5C,CAAC;QAAC,MAAM,CAAC;YACP,0BAA0B;QAC5B,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,MAAM,CAAC,MAAM,mBAAmB,GAC9B,iBAAiB,CAAC,qBAAqB,CAAC;IACxC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;sHA8BoH,CAAC;AAEvH,MAAM,CAAC,MAAM,mBAAmB,GAAG,oJAAoJ,CAAC;AAExL,MAAM,CAAC,MAAM,wBAAwB,GACnC,iBAAiB,CAAC,yBAAyB,CAAC;IAC5C;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IAgCE,CAAC;AAEL,MAAM,UAAU,iBAAiB,CAAC,OAAgB;IAChD,IAAI,CAAC,OAAO,IAAI,OAAO,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,mBAAmB,CAAC;IACxE,OAAO,GAAG,mBAAmB,YAAY,OAAO,OAAO,CAAC;AAC1D,CAAC;AAED,MAAM,UAAU,0BAA0B,CAAC,OAAgB;IACzD,IAAI,CAAC,OAAO,IAAI,OAAO,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,wBAAwB,CAAC;IAC7E,OAAO,GAAG,wBAAwB,YAAY,OAAO,OAAO,CAAC;AAC/D,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,KAAa,EAAE,WAAoB;IACjE,MAAM,KAAK,GAAG,CAAC,SAAS,KAAK,EAAE,CAAC,CAAC;IACjC,IAAI,WAAW;QAAE,KAAK,CAAC,IAAI,CAAC,gBAAgB,WAAW,EAAE,CAAC,CAAC;IAC3D,KAAK,CAAC,IAAI,CAAC,iDAAiD,CAAC,CAAC;IAC9D,OAAO,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AAC5B,CAAC;AAED,MAAM,UAAU,qBAAqB,CACnC,IAA6C,EAC7C,OAAiD;IAEjD,MAAM,YAAY,GAAG,CAAC,SAAS,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;IAC7C,IAAI,IAAI,CAAC,WAAW;QAAE,YAAY,CAAC,IAAI,CAAC,gBAAgB,IAAI,CAAC,WAAW,EAAE,CAAC,CAAC;IAC5E,YAAY,CAAC,IAAI,CAAC,mEAAmE,CAAC,CAAC;IACvF,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,sBAAsB,OAAO;SAC3D,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC,UAAU,MAAM,CAAC,CAAC,MAAM,EAAE,CAAC;SAC5C,IAAI,CAAC,MAAM,CAAC,mCAAmC,CAAC;AACrD,CAAC"}
+{"version":3,"file":"prompts.js","sourceRoot":"","sources":["../src/prompts.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,SAAS,CAAC;AAEzB,SAAS,iBAAiB,CAAC,IAAY;IACrC,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACjC,IAAI,MAAM;QAAE,OAAO,MAAM,CAAC;IAC1B,MAAM,OAAO,GAAG,GAAG,IAAI,OAAO,CAAC;IAC/B,MAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IACtC,IAAI,QAAQ,EAAE,CAAC;QACb,IAAI,CAAC;YACH,OAAO,EAAE,CAAC,YAAY,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;QAC5C,CAAC;QAAC,MAAM,CAAC;YACP,0BAA0B;QAC5B,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,MAAM,CAAC,MAAM,mBAAmB,GAC9B,iBAAiB,CAAC,qBAAqB,CAAC;IACxC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;+FAuC6F,CAAC;AAEhG,MAAM,CAAC,MAAM,mBAAmB,GAAG,8JAA8J,CAAC;AAElM,MAAM,CAAC,MAAM,wBAAwB,GACnC,iBAAiB,CAAC,yBAAyB,CAAC;IAC5C;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IAiCE,CAAC;AAEL,MAAM,UAAU,iBAAiB,CAAC,OAAgB;IAChD,IAAI,CAAC,OAAO,IAAI,OAAO,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,mBAAmB,CAAC;IACxE,OAAO,GAAG,mBAAmB,YAAY,OAAO,OAAO,CAAC;AAC1D,CAAC;AAED,MAAM,UAAU,0BAA0B,CAAC,OAAgB;IACzD,IAAI,CAAC,OAAO,IAAI,OAAO,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,wBAAwB,CAAC;IAC7E,OAAO,GAAG,wBAAwB,YAAY,OAAO,OAAO,CAAC;AAC/D,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,KAAa,EAAE,WAAoB;IACjE,MAAM,KAAK,GAAG,CAAC,SAAS,KAAK,EAAE,CAAC,CAAC;IACjC,IAAI,WAAW;QAAE,KAAK,CAAC,IAAI,CAAC,gBAAgB,WAAW,EAAE,CAAC,CAAC;IAC3D,KAAK,CAAC,IAAI,CAAC,iDAAiD,CAAC,CAAC;IAC9D,OAAO,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AAC5B,CAAC;AAED,MAAM,UAAU,qBAAqB,CACnC,IAA6C,EAC7C,OAAiD;IAEjD,MAAM,YAAY,GAAG,CAAC,SAAS,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;IAC7C,IAAI,IAAI,CAAC,WAAW;QAAE,YAAY,CAAC,IAAI,CAAC,gBAAgB,IAAI,CAAC,WAAW,EAAE,CAAC,CAAC;IAC5E,YAAY,CAAC,IAAI,CAAC,mEAAmE,CAAC,CAAC;IACvF,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,sBAAsB,OAAO;SAC3D,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC,UAAU,MAAM,CAAC,CAAC,MAAM,EAAE,CAAC;SAC5C,IAAI,CAAC,MAAM,CAAC,mCAAmC,CAAC;AACrD,CAAC"}

package/dist/server/pglite-data/postmaster.pid

@@ -1,6 +1,6 @@
 -42
 /pglite/data
-1783134189
+1783146074
 5432
 
 

package/dist/server/src/app.ts

@@ -7,7 +7,7 @@ import { providerRoutes } from './routes/providers.js';
 import { routerRoutes } from './routes/router.js';
 import { metricsRoutes } from './routes/metrics.js';
 import { promptVersionRoutes } from './routes/prompt-versions.js';
-import { reportRoutes } from './routes/reports.js';
+import { benchmarkRoutes } from './routes/benchmarks.js';
 
 export const app: express.Express = express();
 
@@ -20,7 +20,7 @@ app.use('/providers', providerRoutes(prisma));
 app.use('/router', routerRoutes(prisma));
 app.use('/metrics', metricsRoutes(prisma));
 app.use('/prompt-versions', promptVersionRoutes(prisma));
-app.use('/reports', reportRoutes());
+app.use('/benchmarks', benchmarkRoutes(prisma));
 
 app.use((err: unknown, _req: express.Request, res: express.Response, _next: express.NextFunction) => {
   console.error(err);

package/dist/server/src/routes/benchmarks.ts

@@ -0,0 +1,152 @@
+import { Router } from 'express';
+import type { PrismaClient } from '@omega/db';
+import fs from 'node:fs/promises';
+import path from 'node:path';
+import { fileURLToPath } from 'node:url';
+import { spawn } from 'node:child_process';
+import { z } from 'zod';
+import { asyncHandler } from '../lib/async-handler.js';
+
+const __filename = fileURLToPath(import.meta.url);
+const root = path.resolve(__filename, '..', '..', '..', '..');
+const reportsDir = path.join(root, '.omega', 'reports');
+const statusFile = path.join(root, '.omega', 'bench-run-status.json');
+
+const runSchema = z.object({
+  suite: z.enum(['synthetic', 'deep-swe']).optional(),
+  nTasks: z.number().int().positive().optional(),
+  provider: z.string().optional(),
+  model: z.string().optional(),
+  timeout: z.number().int().positive().optional(),
+});
+
+interface RunStatus {
+  running: boolean;
+  pid?: number;
+  output?: string;
+}
+
+async function readStatus(): Promise<RunStatus> {
+  try {
+    const raw = await fs.readFile(statusFile, 'utf-8');
+    return JSON.parse(raw) as RunStatus;
+  } catch {
+    return { running: false };
+  }
+}
+
+async function writeStatus(status: RunStatus): Promise<void> {
+  await fs.mkdir(path.dirname(statusFile), { recursive: true });
+  await fs.writeFile(statusFile, JSON.stringify(status, null, 2), 'utf-8');
+}
+
+function isReportFile(name: string, prefix: string): boolean {
+  return name.startsWith(prefix) && name.endsWith('.json');
+}
+
+async function listReports(prefix: string): Promise<string[]> {
+  try {
+    const files = await fs.readdir(reportsDir);
+    return files
+      .filter((f) => isReportFile(f, prefix))
+      .sort()
+      .reverse();
+  } catch {
+    return [];
+  }
+}
+
+export function benchmarkRoutes(_prisma: PrismaClient): Router {
+  const r = Router();
+
+  r.get('/reports', asyncHandler(async (_req, res) => {
+    const [benchmark, ab] = await Promise.all([
+      listReports('benchmark-'),
+      listReports('ab-'),
+    ]);
+    res.json({ benchmark, ab });
+  }));
+
+  r.get('/reports/:file', asyncHandler(async (req, res) => {
+    const file = req.params.file;
+    if (!isReportFile(file, 'benchmark-') && !isReportFile(file, 'ab-')) {
+      res.status(400).json({ error: 'Invalid report file' });
+      return;
+    }
+    const filePath = path.join(reportsDir, file);
+    const resolved = path.resolve(filePath);
+    const resolvedReportsDir = path.resolve(reportsDir);
+    if (!resolved.startsWith(resolvedReportsDir + path.sep) && resolved !== resolvedReportsDir) {
+      res.status(400).json({ error: 'Invalid report path' });
+      return;
+    }
+    try {
+      const raw = await fs.readFile(filePath, 'utf-8');
+      const data = JSON.parse(raw) as unknown;
+      res.json(data);
+    } catch {
+      res.status(404).json({ error: 'Report not found' });
+    }
+  }));
+
+  r.post('/run', asyncHandler(async (req, res) => {
+    const body = runSchema.parse(req.body);
+    const status = await readStatus();
+    if (status.running) {
+      res.status(409).json({ error: 'Benchmark run already in progress', pid: status.pid });
+      return;
+    }
+
+    const args = ['apps/cli/dist/index.js', 'bench', 'run'];
+    if (body.suite) {
+      args.push('--suite', body.suite);
+    }
+    if (body.nTasks) {
+      args.push('--n-tasks', String(body.nTasks));
+    }
+    if (body.timeout) {
+      args.push('--timeout', String(body.timeout));
+    }
+    if (body.provider) {
+      args.push('--provider', body.provider);
+    }
+    if (body.model) {
+      args.push('--model', body.model);
+    }
+
+    const child = spawn('node', args, {
+      cwd: root,
+      detached: true,
+      stdio: ['ignore', 'pipe', 'pipe'],
+    });
+
+    let output = '';
+    child.stdout.on('data', (chunk: Buffer) => {
+      output += chunk.toString();
+    });
+    child.stderr.on('data', (chunk: Buffer) => {
+      output += chunk.toString();
+    });
+
+    child.on('exit', () => {
+      void (async () => {
+        const current = await readStatus();
+        if (current.pid === child.pid) {
+          await writeStatus({ running: false, output });
+        }
+      })();
+    });
+
+    child.unref();
+
+    await writeStatus({ running: true, pid: child.pid, output: '' });
+    res.status(202).json({ pid: child.pid, status: 'started' });
+  }));
+
+  r.get('/run-status', asyncHandler(async (_req, res) => {
+    const status = await readStatus();
+    res.json(status);
+  }));
+
+  return r;
+}