omnikey-cli 1.0.36 → 1.0.37

package/backend-dist/agent/agentPrompts.js

@@ -17,28 +17,17 @@ ${hasTaskInstructions
 - Default to a \`<shell_script>\` for anything involving the machine, network, files, processes, env vars, or system state — never answer from training data alone.
 - **Read vs write:** For open-ended/ambiguous requests run safe read-only commands first to understand the current state. When the user **explicitly** asks to create, update, delete, configure, or run something — do it directly; no need to ask for confirmation unless the scope is genuinely unclear.
 - **Package installation:** Install any package required to complete the task. Include the install step as its own phase so you can confirm it succeeded before building on it. Prefer project-local or user scope; avoid \`sudo\`/admin unless the user explicitly asks.
-${config_1.config.browserDebugPort !== undefined ? `- **Browser automation:** When the user explicitly asks to interact with a browser (click a button, fill a form, check a page, take a screenshot, etc.), generate \`<shell_script>\` blocks that use Node.js and \`playwright-core\` — one phase at a time (phasing rules below apply).
+${config_1.config.browserDebugPort !== undefined
+        ? `- **Browser automation:** Use browser automation proactively when needed to complete the task.
+  - Do NOT wait for explicit user wording like "use browser" if interaction is obviously required to get the final result.
+  - If \`web_search\` / \`web_fetch\` do not provide enough usable context (blocked pages, incomplete data, client-rendered content, auth walls, dynamic tables, hidden details, repeated low-value fetch results), immediately switch to Playwright-based browser interaction.
+  - Generate \`<shell_script>\` blocks that use Node.js and \`playwright-core\` — one phase at a time (phasing rules below apply).
   - **Phase 1 — ensure deps:** Check and install \`playwright-core\` if missing:
     \`node -e "require('/tmp/playwright-runner/node_modules/playwright-core')" 2>/dev/null || npm install --prefix /tmp/playwright-runner playwright-core --silent\`
-  - **Phase 2 — connect & navigate:** Try CDP first; fall back to the existing debug profile. Reuse an open tab if the URL already matches — never open a duplicate.
-    \`\`\`js
-    const { chromium } = require('/tmp/playwright-runner/node_modules/playwright-core');
-    let browser, page;
-    try {
-      browser = await chromium.connectOverCDP('http://localhost:${config_1.config.browserDebugPort}');
-      const pages = browser.contexts().flatMap(c => c.pages());
-      page = pages.find(p => p.url().startsWith(TARGET_URL)) ?? null;
-      if (page) { await page.bringToFront(); }
-      else { page = await browser.contexts()[0].newPage(); await page.goto(TARGET_URL, { waitUntil: 'domcontentloaded', timeout: 15000 }); }
-    } catch {
-      const ctx = await chromium.launchPersistentContext('${config_1.config.browserDebugUserDataDir}', { executablePath: '${config_1.config.browserDebugExecutable}', headless: false });
-      browser = ctx;
-      page = ctx.pages().find(p => p.url().startsWith(TARGET_URL)) ?? await ctx.newPage();
-      if (!page.url().startsWith(TARGET_URL)) await page.goto(TARGET_URL, { waitUntil: 'domcontentloaded', timeout: 15000 });
-    }
-    \`\`\`
-  - **Phase 3+ — one action per script:** Each subsequent script reconnects the same way, finds the already-open tab, performs exactly one action (click / type / select / screenshot / read text), prints the result, then calls \`browser.disconnect()\` (CDP) or just exits (profile launch — leaves the window open).
-  - Always inline Node.js via a bash heredoc so the script is self-contained. Print structured output to stdout so it returns as \`TERMINAL OUTPUT:\`.` : ''}
+  - **Phase 2 — connect & navigate:** Connect to the running browser via CDP at \`http://localhost:${config_1.config.browserDebugPort}\`. If CDP fails, fall back to launching a persistent context using the debug profile at \`${config_1.config.browserDebugUserDataDir}\` with the executable at \`${config_1.config.browserDebugExecutable}\` (headless: false). Once connected, navigate to any URL required by the task — open any page needed, reusing an existing tab if the URL already matches or creating a new one if not. There is no restriction on which sites or pages you can visit; open whatever is necessary to complete the task.
+  - **Phase 3+ — one action per script:** Each subsequent script reconnects via the same CDP endpoint (\`http://localhost:${config_1.config.browserDebugPort}\`) or profile fallback, finds the already-open tab (or reopens it), performs exactly one action (click, type, select, scroll, screenshot, read text, extract data, fill forms, etc.), prints the result to stdout, then calls \`browser.disconnect()\` (CDP) or exits (profile launch). You may perform any interaction the task requires — reading content, extracting structured data, submitting forms, navigating between pages, or capturing screenshots.
+  - Always inline Node.js via a bash heredoc so the script is self-contained. Print structured output to stdout so it returns as \`TERMINAL OUTPUT:\`.`
+        : ''}
 - Use ${!isWindows ? 'bash (macOS/Linux)' : 'PowerShell'}. Every script must be self-contained and ready to run as-is.
 - Skip the script only for purely factual/conversational requests with no live data dependency (e.g. "what is 2+2").
 

package/backend-dist/agent/agentServer.js

@@ -171,7 +171,6 @@ async function runToolLoop(initialResult, session, sessionId, send, log, tools,
     return result;
 }
 const aiModel = (0, ai_client_1.getDefaultModel)(config_1.config.aiProvider, 'smart');
-const MAX_TURNS = 20;
 // ─── DB helpers ───────────────────────────────────────────────────────────────
 async function persistSessionToDB(sessionId, state) {
     try {
@@ -276,7 +275,7 @@ ${prompt}
     };
     // Persist immediately so that GET /sessions picks it up right away.
     try {
-        await agentSession_1.AgentSession.findOrCreate({
+        const [dbSession, created] = await agentSession_1.AgentSession.findOrCreate({
             where: { id: sessionId, subscriptionId: subscription.id },
             defaults: {
                 id: sessionId,
@@ -288,6 +287,25 @@ ${prompt}
                 lastActiveAt: new Date(),
             },
         });
+        if (!created) {
+            const history = JSON.parse(dbSession.historyJson || '[]');
+            const existingEntry = {
+                subscription,
+                history,
+                turns: dbSession.turns,
+            };
+            log.info('Reused existing agent session row from DB during create path', {
+                sessionId,
+                subscriptionId: subscription.id,
+                turns: existingEntry.turns,
+            });
+            return {
+                sessionState: existingEntry,
+                hasStoredPrompt: history
+                    .filter((h) => h.role === 'user')
+                    .some((h) => typeof h.content === 'string' && h.content.includes('<stored_instructions>')),
+            };
+        }
         // Prune oldest sessions after each creation so the cap is always respected.
         void enforceSessionCap(subscription.id, log);
     }
@@ -313,14 +331,6 @@ async function runAgentTurnInternal(sessionId, subscription, clientMessage, send
         subscriptionId: subscription.id,
         turn: session.turns,
     });
-    const effectiveMaxTurns = options?.maxTurns ?? MAX_TURNS;
-    // On the final iteration, instruct the LLM to provide a consolidated answer.
-    if (session.turns === effectiveMaxTurns) {
-        (0, utils_1.pushToSessionHistory)(logger_1.logger, session, {
-            role: 'system',
-            content: 'Provide a single, final, concise answer based on the entire conversation so far. Wrap the answer in a <final_answer>...</final_answer> block and do not ask for further input or mention additional shell scripts to run. Do not include any <shell_script> block in this response.',
-        });
-    }
     // Append the client message as user content, marking terminal
     // output and errors in the text so the agent can reason about them.
     let userContent = clientMessage.content || '';
@@ -369,10 +379,7 @@ async function runAgentTurnInternal(sessionId, subscription, clientMessage, send
             }
         }
     }
-    // On the final turn we omit tools so the model is forced to emit a
-    // plain text <final_answer> rather than issuing another tool call.
-    const isFinalTurn = session.turns >= effectiveMaxTurns;
-    const tools = isFinalTurn ? undefined : (0, utils_1.buildAvailableTools)();
+    const tools = (0, utils_1.buildAvailableTools)();
     const recordUsage = async (result) => {
         const usage = result.usage;
         if (!usage)
@@ -433,7 +440,7 @@ async function runAgentTurnInternal(sessionId, subscription, clientMessage, send
         }
         // If the model requested web tool calls, execute them and get a follow-up
         // response before deciding what to send to the client.
-        if (!isFinalTurn && result.finish_reason === 'tool_calls') {
+        if (result.finish_reason === 'tool_calls') {
             log.info('Running web tool calls to gather information', {
                 sessionId,
                 subscriptionId: subscription.id,
@@ -488,6 +495,9 @@ async function runAgentTurnInternal(sessionId, subscription, clientMessage, send
                             'No plain text. No other format.',
                         ].join('\n'),
                 });
+                // DB-only session state: persist before recursive handoff so the
+                // follow-up turn reads the latest history and turn count.
+                await persistSessionToDB(sessionId, session);
                 await runAgentTurnInternal(sessionId, subscription, {
                     sender: 'agent',
                     session_id: sessionId,
@@ -497,15 +507,9 @@ async function runAgentTurnInternal(sessionId, subscription, clientMessage, send
                 return;
             }
         }
-        // Ensure that a proper <final_answer> block is produced for the
-        // desktop clients once we reach the final turn. If the model did
-        // not emit either a <shell_script> or <final_answer> tag on the
-        // MAX_TURNS turn, we treat this as the final natural-language answer
-        // and wrap it in <final_answer> tags so the client can stop
-        // waiting and paste the result.
         const hasShellScriptTag = content.includes('<shell_script>');
         const hasFinalAnswerTag = content.includes('<final_answer>');
-        if (hasShellScriptTag && !isFinalTurn) {
+        if (hasShellScriptTag) {
             log.info('Completed agent turn. Sending back scripts, waiting for results.', {
                 sessionId,
                 subscriptionId: subscription.id,
@@ -516,6 +520,10 @@ async function runAgentTurnInternal(sessionId, subscription, clientMessage, send
                 role: 'assistant',
                 content,
             });
+            // Persist before sending so that if the send callback triggers a new
+            // runAgentTurn immediately (e.g. cron shell-script loop), the DB already
+            // has the updated turn count and history.
+            await persistSessionToDB(sessionId, session);
             send({
                 session_id: sessionId,
                 sender: 'agent',
@@ -525,8 +533,8 @@ async function runAgentTurnInternal(sessionId, subscription, clientMessage, send
             });
             return;
         }
-        if (isFinalTurn || hasFinalAnswerTag) {
-            log.info('Finalizing agent session after max turns or final answer tag', {
+        if (hasFinalAnswerTag) {
+            log.info('Finalizing agent session after final answer tag', {
                 sessionId,
                 subscriptionId: subscription.id,
                 turns: session.turns,

package/backend-dist/index.js

@@ -8,7 +8,6 @@ const cors_1 = __importDefault(require("cors"));
 const path_1 = __importDefault(require("path"));
 const fs_1 = __importDefault(require("fs"));
 const zlib_1 = __importDefault(require("zlib"));
-const child_process_1 = require("child_process");
 const subscriptionRoutes_1 = require("./subscriptionRoutes");
 const featureRoutes_1 = require("./featureRoutes");
 const db_1 = require("./db");
@@ -24,7 +23,6 @@ require("./models/scheduledJob");
 const bucket_adapter_1 = require("./bucket-adapter");
 const app = (0, express_1.default)();
 const PORT = Number(config_1.config.port);
-const IS_CRON_CHILD = process.env.OMNIKEY_CRON_CHILD === '1';
 app.set('trust proxy', 1);
 app.use((0, cors_1.default)());
 app.use(express_1.default.json());
@@ -167,38 +165,9 @@ app.get('*', (_req, res) => {
     res.sendFile(path_1.default.join(process.cwd(), 'public', 'index.html'));
 });
 let server = null;
-let cronChildProcess = null;
-function startCronChildProcess() {
-    if (IS_CRON_CHILD || cronChildProcess)
-        return;
-    const childPort = PORT + 1;
-    const entry = process.argv[1] || __filename;
-    cronChildProcess = (0, child_process_1.fork)(entry, [], {
-        env: {
-            ...process.env,
-            OMNIKEY_CRON_CHILD: '1',
-            OMNIKEY_PORT: String(childPort),
-        },
-        execArgv: process.execArgv,
-        stdio: 'inherit',
-    });
-    logger_1.logger.info('Spawned cron child process.', {
-        pid: cronChildProcess.pid,
-        port: childPort,
-    });
-    cronChildProcess.on('exit', (code, signal) => {
-        logger_1.logger.warn('Cron child process exited.', { code, signal });
-        cronChildProcess = null;
-    });
-}
 async function start() {
     try {
         await (0, db_1.initDatabase)(logger_1.logger);
-        if (IS_CRON_CHILD) {
-            logger_1.logger.info('Starting cron child process mode.', { port: PORT });
-            (0, scheduledJobExecutor_1.startScheduledJobExecutor)();
-            return;
-        }
         server = app.listen(PORT, () => {
             logger_1.logger.info(`Enhancer API listening on http://localhost:${PORT}`, {
                 isSelfHosted: config_1.config.isSelfHosted,
@@ -212,7 +181,7 @@ async function start() {
             (0, agentServer_1.attachAgentWebSocketServer)(server);
         }
         if (config_1.config.isSelfHosted) {
-            startCronChildProcess();
+            (0, scheduledJobExecutor_1.startScheduledJobExecutor)();
         }
     }
     catch (err) {
@@ -223,15 +192,6 @@ async function start() {
 start();
 function gracefulShutdown(signal) {
     logger_1.logger.info(`Received ${signal}. Starting graceful shutdown...`);
-    if (cronChildProcess) {
-        cronChildProcess.kill('SIGTERM');
-        cronChildProcess = null;
-    }
-    if (IS_CRON_CHILD) {
-        logger_1.logger.info('Cron child process exiting.');
-        process.exit(0);
-        return;
-    }
     if (!server) {
         logger_1.logger.info('Server was not started or already closed. Exiting process.');
         process.exit(0);

package/backend-dist/scheduledJobExecutor.js

@@ -144,7 +144,7 @@ function runCronJob(job, subscription, sessionId) {
                         content: output,
                         is_terminal_output: true,
                         is_error: isError,
-                    }, send, logger_1.logger, { maxTurns: MAX_CRON_TURNS, isCronJob: true }).catch((err) => settle(err instanceof Error ? err : new Error(String(err))));
+                    }, send, logger_1.logger, { isCronJob: true }).catch((err) => settle(err instanceof Error ? err : new Error(String(err))));
                     return;
                 }
                 if (FINAL_ANSWER_RE.test(content)) {
@@ -158,7 +158,7 @@ function runCronJob(job, subscription, sessionId) {
             sender: 'user',
             content: job.prompt,
             platform: job.platform ?? undefined,
-        }, send, logger_1.logger, { maxTurns: MAX_CRON_TURNS, isCronJob: true }).catch((err) => settle(err instanceof Error ? err : new Error(String(err))));
+        }, send, logger_1.logger, { isCronJob: true }).catch((err) => settle(err instanceof Error ? err : new Error(String(err))));
     });
 }
 async function executeJob(job) {

package/package.json

@@ -4,7 +4,7 @@
     "access": "public",
     "registry": "https://registry.npmjs.org/"
   },
-  "version": "1.0.36",
+  "version": "1.0.37",
   "description": "CLI for onboarding users to Omnikey AI and configuring OPENAI_API_KEY. Use Yarn for install/build.",
   "engines": {
     "node": ">=14.0.0",