@hasna/testers 0.0.3 → 0.0.5

package/README.md

@@ -228,7 +228,7 @@ Screenshots are saved to `~/.testers/screenshots/` organized by:
 Install for Claude Code:
 
 ```bash
-claude mcp add --transport stdio --scope user testers-mcp -- testers-mcp
+claude mcp add --transport stdio --scope user testers -- testers-mcp
 ```
 
 Available tools: `create_scenario`, `list_scenarios`, `run_scenarios`, `get_results`, `get_screenshots`, and more.

package/dist/cli/index.js

@@ -3641,7 +3641,8 @@ async function runAgentLoop(options) {
     screenshotter,
     model,
     runId,
-    maxTurns = 30
+    maxTurns = 30,
+    onStep
   } = options;
   const systemPrompt = [
     "You are an expert QA testing agent. Your job is to thoroughly test web application scenarios.",
@@ -3700,8 +3701,8 @@ async function runAgentLoop(options) {
       }
       const toolUseBlocks = response.content.filter((block) => block.type === "tool_use");
       if (toolUseBlocks.length === 0 && response.stop_reason === "end_turn") {
-        const textBlocks = response.content.filter((block) => block.type === "text");
-        const textReasoning = textBlocks.map((b) => b.text).join(`
+        const textBlocks2 = response.content.filter((block) => block.type === "text");
+        const textReasoning = textBlocks2.map((b) => b.text).join(`
 `);
         return {
           status: "error",
@@ -3712,10 +3713,22 @@ async function runAgentLoop(options) {
         };
       }
       const toolResults = [];
+      const textBlocks = response.content.filter((block) => block.type === "text");
+      if (textBlocks.length > 0 && onStep) {
+        const thinking = textBlocks.map((b) => b.text).join(`
+`);
+        onStep({ type: "thinking", thinking, stepNumber });
+      }
       for (const toolBlock of toolUseBlocks) {
         stepNumber++;
         const toolInput = toolBlock.input;
+        if (onStep) {
+          onStep({ type: "tool_call", toolName: toolBlock.name, toolInput, stepNumber });
+        }
         const execResult = await executeTool(page, screenshotter, toolBlock.name, toolInput, { runId, scenarioSlug, stepNumber });
+        if (onStep) {
+          onStep({ type: "tool_result", toolName: toolBlock.name, toolResult: execResult.result, stepNumber });
+        }
         if (execResult.screenshot) {
           screenshots.push({
             ...execResult.screenshot,
@@ -3824,6 +3837,9 @@ function loadConfig() {
 
 // src/lib/runner.ts
 var eventHandler = null;
+function onRunEvent(handler) {
+  eventHandler = handler;
+}
 function emit(event) {
   if (eventHandler)
     eventHandler(event);
@@ -3858,7 +3874,20 @@ async function runSingleScenario(scenario, runId, options) {
       screenshotter,
       model,
       runId,
-      maxTurns: 30
+      maxTurns: 30,
+      onStep: (stepEvent) => {
+        emit({
+          type: `step:${stepEvent.type}`,
+          scenarioId: scenario.id,
+          scenarioName: scenario.name,
+          runId,
+          toolName: stepEvent.toolName,
+          toolInput: stepEvent.toolInput,
+          toolResult: stepEvent.toolResult,
+          thinking: stepEvent.thinking,
+          stepNumber: stepEvent.stepNumber
+        });
+      }
     });
     for (const ss of agentResult.screenshots) {
       createScreenshot({
@@ -3968,6 +3997,79 @@ async function runByFilter(options) {
   }
   return runBatch(scenarios, options);
 }
+function startRunAsync(options) {
+  const config = loadConfig();
+  const model = resolveModel(options.model ?? config.defaultModel);
+  let scenarios;
+  if (options.scenarioIds && options.scenarioIds.length > 0) {
+    const all = listScenarios({ projectId: options.projectId });
+    scenarios = all.filter((s) => options.scenarioIds.includes(s.id) || options.scenarioIds.includes(s.shortId));
+  } else {
+    scenarios = listScenarios({
+      projectId: options.projectId,
+      tags: options.tags,
+      priority: options.priority
+    });
+  }
+  const parallel = options.parallel ?? 1;
+  const run = createRun({
+    url: options.url,
+    model,
+    headed: options.headed,
+    parallel,
+    projectId: options.projectId
+  });
+  if (scenarios.length === 0) {
+    updateRun(run.id, { status: "passed", total: 0, finished_at: new Date().toISOString() });
+    return { runId: run.id, scenarioCount: 0 };
+  }
+  updateRun(run.id, { status: "running", total: scenarios.length });
+  (async () => {
+    const results = [];
+    try {
+      if (parallel <= 1) {
+        for (const scenario of scenarios) {
+          const result = await runSingleScenario(scenario, run.id, options);
+          results.push(result);
+        }
+      } else {
+        const queue = [...scenarios];
+        const running = [];
+        const processNext = async () => {
+          const scenario = queue.shift();
+          if (!scenario)
+            return;
+          const result = await runSingleScenario(scenario, run.id, options);
+          results.push(result);
+          await processNext();
+        };
+        const workers = Math.min(parallel, scenarios.length);
+        for (let i = 0;i < workers; i++) {
+          running.push(processNext());
+        }
+        await Promise.all(running);
+      }
+      const passed = results.filter((r) => r.status === "passed").length;
+      const failed = results.filter((r) => r.status === "failed" || r.status === "error").length;
+      updateRun(run.id, {
+        status: failed > 0 ? "failed" : "passed",
+        passed,
+        failed,
+        total: scenarios.length,
+        finished_at: new Date().toISOString()
+      });
+      emit({ type: "run:complete", runId: run.id });
+    } catch (error) {
+      const errorMsg = error instanceof Error ? error.message : String(error);
+      updateRun(run.id, {
+        status: "failed",
+        finished_at: new Date().toISOString()
+      });
+      emit({ type: "run:complete", runId: run.id, error: errorMsg });
+    }
+  })();
+  return { runId: run.id, scenarioCount: scenarios.length };
+}
 function estimateCost(model, tokens) {
   const costs = {
     "claude-haiku-4-5-20251001": 0.1,
@@ -5289,8 +5391,17 @@ function deleteAuthPreset(name) {
 
 // src/cli/index.tsx
 import { existsSync as existsSync7, mkdirSync as mkdirSync4 } from "fs";
+function formatToolInput(input) {
+  const parts = [];
+  for (const [key, value] of Object.entries(input)) {
+    const str = typeof value === "string" ? value : JSON.stringify(value);
+    const truncated = str.length > 60 ? str.slice(0, 60) + "..." : str;
+    parts.push(`${key}="${truncated}"`);
+  }
+  return parts.join(" ");
+}
 var program2 = new Command;
-program2.name("testers").version("0.0.1").description("AI-powered browser testing CLI");
+program2.name("testers").version("0.0.4").description("AI-powered browser testing CLI");
 var CONFIG_DIR2 = join6(process.env["HOME"] ?? "~", ".testers");
 var CONFIG_PATH2 = join6(CONFIG_DIR2, "config.json");
 function getActiveProject() {
@@ -5442,9 +5553,75 @@ program2.command("delete <id>").description("Delete a scenario").action((id) =>
 program2.command("run <url> [description]").description("Run test scenarios against a URL").option("-t, --tag <tag>", "Filter by tag (repeatable)", (val, acc) => {
   acc.push(val);
   return acc;
-}, []).option("-s, --scenario <id>", "Run specific scenario ID").option("-p, --priority <level>", "Filter by priority").option("--headed", "Run browser in headed mode", false).option("-m, --model <model>", "AI model to use").option("--parallel <n>", "Number of parallel browsers", "1").option("--json", "Output results as JSON", false).option("-o, --output <filepath>", "Write JSON results to file").option("--timeout <ms>", "Timeout in milliseconds").option("--from-todos", "Import scenarios from todos before running", false).option("--project <id>", "Project ID").action(async (url, description, opts) => {
+}, []).option("-s, --scenario <id>", "Run specific scenario ID").option("-p, --priority <level>", "Filter by priority").option("--headed", "Run browser in headed mode", false).option("-m, --model <model>", "AI model to use").option("--parallel <n>", "Number of parallel browsers", "1").option("--json", "Output results as JSON", false).option("-o, --output <filepath>", "Write JSON results to file").option("--timeout <ms>", "Timeout in milliseconds").option("--from-todos", "Import scenarios from todos before running", false).option("--project <id>", "Project ID").option("-b, --background", "Start run in background and return immediately", false).action(async (url, description, opts) => {
   try {
     const projectId = resolveProject(opts.project);
+    if (opts.fromTodos) {
+      const result = importFromTodos({ projectId });
+      console.log(chalk4.blue(`Imported ${result.imported} scenarios from todos (${result.skipped} skipped)`));
+    }
+    if (opts.background) {
+      if (description) {
+        createScenario({ name: description, description, tags: ["ad-hoc"], projectId });
+      }
+      const { runId, scenarioCount } = startRunAsync({
+        url,
+        tags: opts.tag.length > 0 ? opts.tag : undefined,
+        scenarioIds: opts.scenario ? [opts.scenario] : undefined,
+        priority: opts.priority,
+        model: opts.model,
+        headed: opts.headed,
+        parallel: parseInt(opts.parallel, 10),
+        timeout: opts.timeout ? parseInt(opts.timeout, 10) : undefined,
+        projectId
+      });
+      console.log(chalk4.green(`Run started in background: ${chalk4.bold(runId.slice(0, 8))}`));
+      console.log(chalk4.dim(`  Scenarios: ${scenarioCount}`));
+      console.log(chalk4.dim(`  URL: ${url}`));
+      console.log(chalk4.dim(`  Check progress: testers results ${runId.slice(0, 8)}`));
+      process.exit(0);
+    }
+    if (!opts.json && !opts.output) {
+      onRunEvent((event) => {
+        switch (event.type) {
+          case "scenario:start":
+            console.log(chalk4.blue(`  [start] ${event.scenarioName ?? event.scenarioId}`));
+            break;
+          case "step:thinking":
+            if (event.thinking) {
+              const preview = event.thinking.length > 120 ? event.thinking.slice(0, 120) + "..." : event.thinking;
+              console.log(chalk4.dim(`    [think] ${preview}`));
+            }
+            break;
+          case "step:tool_call":
+            console.log(chalk4.cyan(`    [step ${event.stepNumber}] ${event.toolName}${event.toolInput ? ` ${formatToolInput(event.toolInput)}` : ""}`));
+            break;
+          case "step:tool_result":
+            if (event.toolName === "report_result") {
+              console.log(chalk4.bold(`    [result] ${event.toolResult}`));
+            } else {
+              const resultPreview = (event.toolResult ?? "").length > 100 ? (event.toolResult ?? "").slice(0, 100) + "..." : event.toolResult ?? "";
+              console.log(chalk4.dim(`    [done]  ${resultPreview}`));
+            }
+            break;
+          case "screenshot:captured":
+            console.log(chalk4.dim(`    [screenshot] ${event.screenshotPath}`));
+            break;
+          case "scenario:pass":
+            console.log(chalk4.green(`  [PASS] ${event.scenarioName}`));
+            break;
+          case "scenario:fail":
+            console.log(chalk4.red(`  [FAIL] ${event.scenarioName}`));
+            break;
+          case "scenario:error":
+            console.log(chalk4.yellow(`  [ERR]  ${event.scenarioName}: ${event.error}`));
+            break;
+        }
+      });
+      console.log("");
+      console.log(chalk4.bold(`  Running tests against ${url}`));
+      console.log("");
+    }
     if (description) {
       const scenario = createScenario({
         name: description,
@@ -5475,10 +5652,6 @@ program2.command("run <url> [description]").description("Run test scenarios agai
       }
       process.exit(getExitCode(run2));
     }
-    if (opts.fromTodos) {
-      const result = importFromTodos({ projectId });
-      console.log(chalk4.blue(`Imported ${result.imported} scenarios from todos (${result.skipped} skipped)`));
-    }
     const { run, results } = await runByFilter({
       url,
       tags: opts.tag.length > 0 ? opts.tag : undefined,

package/dist/index.d.ts

@@ -12,7 +12,7 @@ export { loadConfig, resolveModel as resolveModelConfig, getDefaultConfig, } fro
 export { launchBrowser, getPage, closeBrowser, BrowserPool, installBrowser, } from "./lib/browser.js";
 export { Screenshotter, slugify, generateFilename, getScreenshotDir, ensureDir, } from "./lib/screenshotter.js";
 export { createClient, resolveModel, runAgentLoop, executeTool, BROWSER_TOOLS, } from "./lib/ai-client.js";
-export { runSingleScenario, runBatch, runByFilter, onRunEvent, } from "./lib/runner.js";
+export { runSingleScenario, runBatch, runByFilter, startRunAsync, onRunEvent, } from "./lib/runner.js";
 export type { RunOptions, RunEvent, RunEventHandler } from "./lib/runner.js";
 export { formatTerminal, formatJSON, formatSummary, getExitCode, formatRunList, formatScenarioList, formatResultDetail, } from "./lib/reporter.js";
 export { connectToTodos, pullTasks, taskToScenarioInput, importFromTodos, markTodoDone, } from "./lib/todos-connector.js";

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AACA,YAAY,EACV,gBAAgB,EAChB,SAAS,EACT,YAAY,EACZ,WAAW,EACX,UAAU,EACV,QAAQ,EACR,WAAW,EACX,MAAM,EACN,SAAS,EACT,aAAa,EACb,OAAO,EACP,KAAK,EACL,QAAQ,EACR,GAAG,EACH,MAAM,EACN,UAAU,EACV,mBAAmB,EACnB,mBAAmB,EACnB,cAAc,EACd,cAAc,EACd,SAAS,EACT,WAAW,EACX,QAAQ,EACR,mBAAmB,EACnB,mBAAmB,EACnB,cAAc,EACd,UAAU,EACV,aAAa,EACb,gBAAgB,EAChB,aAAa,GACd,MAAM,kBAAkB,CAAC;AAE1B,OAAO,EACL,SAAS,EACT,cAAc,EACd,YAAY,EACZ,eAAe,EACf,UAAU,EACV,aAAa,EACb,iBAAiB,EACjB,eAAe,EACf,qBAAqB,EACrB,gBAAgB,EAChB,mBAAmB,EACnB,oBAAoB,EACpB,YAAY,EACZ,aAAa,EACb,oBAAoB,EACpB,oBAAoB,EACpB,kBAAkB,EAClB,qBAAqB,GACtB,MAAM,kBAAkB,CAAC;AAG1B,OAAO,EACL,WAAW,EACX,aAAa,EACb,aAAa,EACb,gBAAgB,EAChB,GAAG,EACH,IAAI,EACJ,SAAS,GACV,MAAM,kBAAkB,CAAC;AAE1B,OAAO,EACL,cAAc,EACd,WAAW,EACX,oBAAoB,EACpB,aAAa,EACb,cAAc,EACd,cAAc,GACf,MAAM,mBAAmB,CAAC;AAE3B,OAAO,EACL,SAAS,EACT,MAAM,EACN,QAAQ,EACR,SAAS,EACT,SAAS,GACV,MAAM,cAAc,CAAC;AAEtB,OAAO,EACL,YAAY,EACZ,SAAS,EACT,WAAW,EACX,YAAY,EACZ,eAAe,GAChB,MAAM,iBAAiB,CAAC;AAEzB,OAAO,EACL,gBAAgB,EAChB,aAAa,EACb,eAAe,EACf,sBAAsB,GACvB,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EACL,aAAa,EACb,UAAU,EACV,gBAAgB,EAChB,YAAY,EACZ,aAAa,GACd,MAAM,kBAAkB,CAAC;AAE1B,OAAO,EACL,aAAa,EACb,QAAQ,EACR,cAAc,EACd,UAAU,GACX,MAAM,gBAAgB,CAAC;AAExB,OAAO,EACL,cAAc,EACd,WAAW,EACX,aAAa,EACb,cAAc,EACd,cAAc,EACd,mBAAmB,EACnB,aAAa,GACd,MAAM,mBAAmB,CAAC;AAG3B,OAAO,EACL,UAAU,EACV,YAAY,IAAI,kBAAkB,EAClC,gBAAgB,GACjB,MAAM,iBAAiB,CAAC;AAEzB,OAAO,EACL,aAAa,EACb,OAAO,EACP,YAAY,EACZ,WAAW,EACX,cAAc,GACf,MAAM,kBAAkB,CAAC;AAE1B,OAAO,EACL,aAAa,EACb,OAAO,EACP,gBAAgB,EAChB,gBAAgB,EAChB,SAAS,GACV,MAAM,wBAAwB,CAAC;AAEhC,OAAO,EACL,YAAY,EACZ,YAAY,EACZ,YAAY,EACZ,WAAW,EACX,aAAa,GACd,MAAM,oBAAoB,CAAC;AAE5B,OAAO,EACL,iBAAiB,EACjB,QAAQ,EACR,WAAW,EACX,UAAU,GACX,MAAM,iBAAiB,CAAC;AACzB,YAAY,EAAE,UAAU,EAAE,QAAQ,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAE7E,OAAO,EACL,cAAc,EACd,UAAU,EACV,aAAa,EACb,WAAW,EACX,aAAa,EACb,kBAAkB,EAClB,kBAAkB,GACnB,MAAM,mBAAmB,CAAC;AAE3B,OAAO,EACL,cAAc,EACd,SAAS,EACT,mBAAmB,EACnB,eAAe,EACf,YAAY,GACb,MAAM,0BAA0B,CAAC;AAElC,OAAO,EACL,SAAS,EACT,SAAS,EACT,cAAc,EACd,WAAW,EACX,cAAc,GACf,MAAM,oBAAoB,CAAC;AAC5B,YAAY,EAAE,cAAc,EAAE,MAAM,oBAAoB,CAAC;AAEzD,OAAO,EACL,WAAW,EACX,eAAe,EACf,mBAAmB,GACpB,MAAM,eAAe,CAAC;AACvB,YAAY,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAEhD,OAAO,EACL,QAAQ,EACR,gBAAgB,EAChB,iBAAiB,GAClB,MAAM,gBAAgB,CAAC;AACxB,YAAY,EAAE,WAAW,EAAE,UAAU,EAAE,MAAM,gBAAgB,CAAC;AAE9D,OAAO,EACL,QAAQ,EACR,kBAAkB,EAClB,cAAc,GACf,MAAM,eAAe,CAAC;AACvB,YAAY,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AAE9D,OAAO,EACL,WAAW,EACX,iBAAiB,EACjB,kBAAkB,GACnB,MAAM,oBAAoB,CAAC;AAE5B,OAAO,EACL,gBAAgB,EAChB,aAAa,EACb,eAAe,EACf,gBAAgB,GACjB,MAAM,sBAAsB,CAAC;AAE9B,OAAO,EACL,kBAAkB,EAClB,oBAAoB,EACpB,aAAa,GACd,MAAM,iBAAiB,CAAC;AAEzB,OAAO,EACL,cAAc,EACd,WAAW,EACX,mBAAmB,EACnB,eAAe,GAChB,MAAM,gBAAgB,CAAC;AACxB,YAAY,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAC;AAEhE,OAAO,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAC;AAE9C,OAAO,EACL,aAAa,EACb,UAAU,EACV,YAAY,EACZ,aAAa,EACb,gBAAgB,EAChB,WAAW,GACZ,MAAM,mBAAmB,CAAC;AAC3B,YAAY,EAAE,OAAO,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AAEjE,OAAO,EAAE,YAAY,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AACzE,YAAY,EAAE,aAAa,EAAE,MAAM,wBAAwB,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AACA,YAAY,EACV,gBAAgB,EAChB,SAAS,EACT,YAAY,EACZ,WAAW,EACX,UAAU,EACV,QAAQ,EACR,WAAW,EACX,MAAM,EACN,SAAS,EACT,aAAa,EACb,OAAO,EACP,KAAK,EACL,QAAQ,EACR,GAAG,EACH,MAAM,EACN,UAAU,EACV,mBAAmB,EACnB,mBAAmB,EACnB,cAAc,EACd,cAAc,EACd,SAAS,EACT,WAAW,EACX,QAAQ,EACR,mBAAmB,EACnB,mBAAmB,EACnB,cAAc,EACd,UAAU,EACV,aAAa,EACb,gBAAgB,EAChB,aAAa,GACd,MAAM,kBAAkB,CAAC;AAE1B,OAAO,EACL,SAAS,EACT,cAAc,EACd,YAAY,EACZ,eAAe,EACf,UAAU,EACV,aAAa,EACb,iBAAiB,EACjB,eAAe,EACf,qBAAqB,EACrB,gBAAgB,EAChB,mBAAmB,EACnB,oBAAoB,EACpB,YAAY,EACZ,aAAa,EACb,oBAAoB,EACpB,oBAAoB,EACpB,kBAAkB,EAClB,qBAAqB,GACtB,MAAM,kBAAkB,CAAC;AAG1B,OAAO,EACL,WAAW,EACX,aAAa,EACb,aAAa,EACb,gBAAgB,EAChB,GAAG,EACH,IAAI,EACJ,SAAS,GACV,MAAM,kBAAkB,CAAC;AAE1B,OAAO,EACL,cAAc,EACd,WAAW,EACX,oBAAoB,EACpB,aAAa,EACb,cAAc,EACd,cAAc,GACf,MAAM,mBAAmB,CAAC;AAE3B,OAAO,EACL,SAAS,EACT,MAAM,EACN,QAAQ,EACR,SAAS,EACT,SAAS,GACV,MAAM,cAAc,CAAC;AAEtB,OAAO,EACL,YAAY,EACZ,SAAS,EACT,WAAW,EACX,YAAY,EACZ,eAAe,GAChB,MAAM,iBAAiB,CAAC;AAEzB,OAAO,EACL,gBAAgB,EAChB,aAAa,EACb,eAAe,EACf,sBAAsB,GACvB,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EACL,aAAa,EACb,UAAU,EACV,gBAAgB,EAChB,YAAY,EACZ,aAAa,GACd,MAAM,kBAAkB,CAAC;AAE1B,OAAO,EACL,aAAa,EACb,QAAQ,EACR,cAAc,EACd,UAAU,GACX,MAAM,gBAAgB,CAAC;AAExB,OAAO,EACL,cAAc,EACd,WAAW,EACX,aAAa,EACb,cAAc,EACd,cAAc,EACd,mBAAmB,EACnB,aAAa,GACd,MAAM,mBAAmB,CAAC;AAG3B,OAAO,EACL,UAAU,EACV,YAAY,IAAI,kBAAkB,EAClC,gBAAgB,GACjB,MAAM,iBAAiB,CAAC;AAEzB,OAAO,EACL,aAAa,EACb,OAAO,EACP,YAAY,EACZ,WAAW,EACX,cAAc,GACf,MAAM,kBAAkB,CAAC;AAE1B,OAAO,EACL,aAAa,EACb,OAAO,EACP,gBAAgB,EAChB,gBAAgB,EAChB,SAAS,GACV,MAAM,wBAAwB,CAAC;AAEhC,OAAO,EACL,YAAY,EACZ,YAAY,EACZ,YAAY,EACZ,WAAW,EACX,aAAa,GACd,MAAM,oBAAoB,CAAC;AAE5B,OAAO,EACL,iBAAiB,EACjB,QAAQ,EACR,WAAW,EACX,aAAa,EACb,UAAU,GACX,MAAM,iBAAiB,CAAC;AACzB,YAAY,EAAE,UAAU,EAAE,QAAQ,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAE7E,OAAO,EACL,cAAc,EACd,UAAU,EACV,aAAa,EACb,WAAW,EACX,aAAa,EACb,kBAAkB,EAClB,kBAAkB,GACnB,MAAM,mBAAmB,CAAC;AAE3B,OAAO,EACL,cAAc,EACd,SAAS,EACT,mBAAmB,EACnB,eAAe,EACf,YAAY,GACb,MAAM,0BAA0B,CAAC;AAElC,OAAO,EACL,SAAS,EACT,SAAS,EACT,cAAc,EACd,WAAW,EACX,cAAc,GACf,MAAM,oBAAoB,CAAC;AAC5B,YAAY,EAAE,cAAc,EAAE,MAAM,oBAAoB,CAAC;AAEzD,OAAO,EACL,WAAW,EACX,eAAe,EACf,mBAAmB,GACpB,MAAM,eAAe,CAAC;AACvB,YAAY,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAEhD,OAAO,EACL,QAAQ,EACR,gBAAgB,EAChB,iBAAiB,GAClB,MAAM,gBAAgB,CAAC;AACxB,YAAY,EAAE,WAAW,EAAE,UAAU,EAAE,MAAM,gBAAgB,CAAC;AAE9D,OAAO,EACL,QAAQ,EACR,kBAAkB,EAClB,cAAc,GACf,MAAM,eAAe,CAAC;AACvB,YAAY,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AAE9D,OAAO,EACL,WAAW,EACX,iBAAiB,EACjB,kBAAkB,GACnB,MAAM,oBAAoB,CAAC;AAE5B,OAAO,EACL,gBAAgB,EAChB,aAAa,EACb,eAAe,EACf,gBAAgB,GACjB,MAAM,sBAAsB,CAAC;AAE9B,OAAO,EACL,kBAAkB,EAClB,oBAAoB,EACpB,aAAa,GACd,MAAM,iBAAiB,CAAC;AAEzB,OAAO,EACL,cAAc,EACd,WAAW,EACX,mBAAmB,EACnB,eAAe,GAChB,MAAM,gBAAgB,CAAC;AACxB,YAAY,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAC;AAEhE,OAAO,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAC;AAE9C,OAAO,EACL,aAAa,EACb,UAAU,EACV,YAAY,EACZ,aAAa,EACb,gBAAgB,EAChB,WAAW,GACZ,MAAM,mBAAmB,CAAC;AAC3B,YAAY,EAAE,OAAO,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AAEjE,OAAO,EAAE,YAAY,EAAE,iBAAiB,EAAE,MAAM,wBAAwB,CAAC;AACzE,YAAY,EAAE,aAAa,EAAE,MAAM,wBAAwB,CAAC"}

package/dist/index.js

@@ -1951,7 +1951,8 @@ async function runAgentLoop(options) {
     screenshotter,
     model,
     runId,
-    maxTurns = 30
+    maxTurns = 30,
+    onStep
   } = options;
   const systemPrompt = [
     "You are an expert QA testing agent. Your job is to thoroughly test web application scenarios.",
@@ -2010,8 +2011,8 @@ async function runAgentLoop(options) {
       }
       const toolUseBlocks = response.content.filter((block) => block.type === "tool_use");
       if (toolUseBlocks.length === 0 && response.stop_reason === "end_turn") {
-        const textBlocks = response.content.filter((block) => block.type === "text");
-        const textReasoning = textBlocks.map((b) => b.text).join(`
+        const textBlocks2 = response.content.filter((block) => block.type === "text");
+        const textReasoning = textBlocks2.map((b) => b.text).join(`
 `);
         return {
           status: "error",
@@ -2022,10 +2023,22 @@ async function runAgentLoop(options) {
         };
       }
       const toolResults = [];
+      const textBlocks = response.content.filter((block) => block.type === "text");
+      if (textBlocks.length > 0 && onStep) {
+        const thinking = textBlocks.map((b) => b.text).join(`
+`);
+        onStep({ type: "thinking", thinking, stepNumber });
+      }
       for (const toolBlock of toolUseBlocks) {
         stepNumber++;
         const toolInput = toolBlock.input;
+        if (onStep) {
+          onStep({ type: "tool_call", toolName: toolBlock.name, toolInput, stepNumber });
+        }
         const execResult = await executeTool(page, screenshotter, toolBlock.name, toolInput, { runId, scenarioSlug, stepNumber });
+        if (onStep) {
+          onStep({ type: "tool_result", toolName: toolBlock.name, toolResult: execResult.result, stepNumber });
+        }
         if (execResult.screenshot) {
           screenshots.push({
             ...execResult.screenshot,
@@ -2115,7 +2128,20 @@ async function runSingleScenario(scenario, runId, options) {
       screenshotter,
       model,
       runId,
-      maxTurns: 30
+      maxTurns: 30,
+      onStep: (stepEvent) => {
+        emit({
+          type: `step:${stepEvent.type}`,
+          scenarioId: scenario.id,
+          scenarioName: scenario.name,
+          runId,
+          toolName: stepEvent.toolName,
+          toolInput: stepEvent.toolInput,
+          toolResult: stepEvent.toolResult,
+          thinking: stepEvent.thinking,
+          stepNumber: stepEvent.stepNumber
+        });
+      }
     });
     for (const ss of agentResult.screenshots) {
       createScreenshot({
@@ -2225,6 +2251,79 @@ async function runByFilter(options) {
   }
   return runBatch(scenarios, options);
 }
+function startRunAsync(options) {
+  const config = loadConfig();
+  const model = resolveModel2(options.model ?? config.defaultModel);
+  let scenarios;
+  if (options.scenarioIds && options.scenarioIds.length > 0) {
+    const all = listScenarios({ projectId: options.projectId });
+    scenarios = all.filter((s) => options.scenarioIds.includes(s.id) || options.scenarioIds.includes(s.shortId));
+  } else {
+    scenarios = listScenarios({
+      projectId: options.projectId,
+      tags: options.tags,
+      priority: options.priority
+    });
+  }
+  const parallel = options.parallel ?? 1;
+  const run = createRun({
+    url: options.url,
+    model,
+    headed: options.headed,
+    parallel,
+    projectId: options.projectId
+  });
+  if (scenarios.length === 0) {
+    updateRun(run.id, { status: "passed", total: 0, finished_at: new Date().toISOString() });
+    return { runId: run.id, scenarioCount: 0 };
+  }
+  updateRun(run.id, { status: "running", total: scenarios.length });
+  (async () => {
+    const results = [];
+    try {
+      if (parallel <= 1) {
+        for (const scenario of scenarios) {
+          const result = await runSingleScenario(scenario, run.id, options);
+          results.push(result);
+        }
+      } else {
+        const queue = [...scenarios];
+        const running = [];
+        const processNext = async () => {
+          const scenario = queue.shift();
+          if (!scenario)
+            return;
+          const result = await runSingleScenario(scenario, run.id, options);
+          results.push(result);
+          await processNext();
+        };
+        const workers = Math.min(parallel, scenarios.length);
+        for (let i = 0;i < workers; i++) {
+          running.push(processNext());
+        }
+        await Promise.all(running);
+      }
+      const passed = results.filter((r) => r.status === "passed").length;
+      const failed = results.filter((r) => r.status === "failed" || r.status === "error").length;
+      updateRun(run.id, {
+        status: failed > 0 ? "failed" : "passed",
+        passed,
+        failed,
+        total: scenarios.length,
+        finished_at: new Date().toISOString()
+      });
+      emit({ type: "run:complete", runId: run.id });
+    } catch (error) {
+      const errorMsg = error instanceof Error ? error.message : String(error);
+      updateRun(run.id, {
+        status: "failed",
+        finished_at: new Date().toISOString()
+      });
+      emit({ type: "run:complete", runId: run.id, error: errorMsg });
+    }
+  })();
+  return { runId: run.id, scenarioCount: scenarios.length };
+}
 function estimateCost(model, tokens) {
   const costs = {
     "claude-haiku-4-5-20251001": 0.1,
@@ -4418,6 +4517,7 @@ export {
   testWebhook,
   taskToScenarioInput,
   startWatcher,
+  startRunAsync,
   slugify,
   shouldRunAt,
   shortUuid,

package/dist/lib/ai-client.d.ts

@@ -31,6 +31,14 @@ interface ToolExecutionResult {
  * Returns the result string and an optional screenshot capture.
  */
 export declare function executeTool(page: Page, screenshotter: Screenshotter, toolName: string, toolInput: Record<string, unknown>, context: ToolContext): Promise<ToolExecutionResult>;
+export type StepEventHandler = (event: {
+    type: "tool_call" | "tool_result" | "thinking";
+    toolName?: string;
+    toolInput?: Record<string, unknown>;
+    toolResult?: string;
+    thinking?: string;
+    stepNumber: number;
+}) => void;
 interface AgentLoopOptions {
     client: Anthropic;
     page: Page;
@@ -39,6 +47,7 @@ interface AgentLoopOptions {
     model: string;
     runId: string;
     maxTurns?: number;
+    onStep?: StepEventHandler;
 }
 interface AgentLoopResult {
     status: "passed" | "failed" | "error";

package/dist/lib/ai-client.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"ai-client.d.ts","sourceRoot":"","sources":["../../src/lib/ai-client.ts"],"names":[],"mappings":"AAAA,OAAO,SAAS,MAAM,mBAAmB,CAAC;AAC1C,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AACvC,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AAExD,OAAO,KAAK,EAAe,QAAQ,EAAE,MAAM,mBAAmB,CAAC;AAI/D;;;GAGG;AACH,wBAAgB,YAAY,CAAC,YAAY,EAAE,MAAM,GAAG,MAAM,CAKzD;AAID,eAAO,MAAM,aAAa,EAAE,SAAS,CAAC,IAAI,EAyTzC,CAAC;AAIF,UAAU,WAAW;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,UAAU,gBAAgB;IACxB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;CAC9B;AAED,UAAU,mBAAmB;IAC3B,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,CAAC,EAAE,gBAAgB,CAAC;CAC/B;AAED;;;GAGG;AACH,wBAAsB,WAAW,CAC/B,IAAI,EAAE,IAAI,EACV,aAAa,EAAE,aAAa,EAC5B,QAAQ,EAAE,MAAM,EAChB,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAClC,OAAO,EAAE,WAAW,GACnB,OAAO,CAAC,mBAAmB,CAAC,CA8P9B;AAID,UAAU,gBAAgB;IACxB,MAAM,EAAE,SAAS,CAAC;IAClB,IAAI,EAAE,IAAI,CAAC;IACX,QAAQ,EAAE,QAAQ,CAAC;IACnB,aAAa,EAAE,aAAa,CAAC;IAC7B,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,UAAU,eAAe;IACvB,MAAM,EAAE,QAAQ,GAAG,QAAQ,GAAG,OAAO,CAAC;IACtC,SAAS,EAAE,MAAM,CAAC;IAClB,cAAc,EAAE,MAAM,CAAC;IACvB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,KAAK,CAAC;QACjB,QAAQ,EAAE,MAAM,CAAC;QACjB,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,EAAE,MAAM,CAAC;QACf,SAAS,EAAE,MAAM,CAAC;QAClB,MAAM,EAAE,MAAM,CAAC;QACf,UAAU,EAAE,MAAM,CAAC;QACnB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;QAC3B,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;QACvB,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;KAC9B,CAAC,CAAC;CACJ;AAED;;;;GAIG;AACH,wBAAsB,YAAY,CAChC,OAAO,EAAE,gBAAgB,GACxB,OAAO,CAAC,eAAe,CAAC,CAyK1B;AAID;;;GAGG;AACH,wBAAgB,YAAY,CAAC,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAQvD"}
+{"version":3,"file":"ai-client.d.ts","sourceRoot":"","sources":["../../src/lib/ai-client.ts"],"names":[],"mappings":"AAAA,OAAO,SAAS,MAAM,mBAAmB,CAAC;AAC1C,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AACvC,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AAExD,OAAO,KAAK,EAAe,QAAQ,EAAE,MAAM,mBAAmB,CAAC;AAI/D;;;GAGG;AACH,wBAAgB,YAAY,CAAC,YAAY,EAAE,MAAM,GAAG,MAAM,CAKzD;AAID,eAAO,MAAM,aAAa,EAAE,SAAS,CAAC,IAAI,EAyTzC,CAAC;AAIF,UAAU,WAAW;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,UAAU,gBAAgB;IACxB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;CAC9B;AAED,UAAU,mBAAmB;IAC3B,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,CAAC,EAAE,gBAAgB,CAAC;CAC/B;AAED;;;GAGG;AACH,wBAAsB,WAAW,CAC/B,IAAI,EAAE,IAAI,EACV,aAAa,EAAE,aAAa,EAC5B,QAAQ,EAAE,MAAM,EAChB,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAClC,OAAO,EAAE,WAAW,GACnB,OAAO,CAAC,mBAAmB,CAAC,CA8P9B;AAID,MAAM,MAAM,gBAAgB,GAAG,CAAC,KAAK,EAAE;IACrC,IAAI,EAAE,WAAW,GAAG,aAAa,GAAG,UAAU,CAAC;IAC/C,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACpC,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;CACpB,KAAK,IAAI,CAAC;AAEX,UAAU,gBAAgB;IACxB,MAAM,EAAE,SAAS,CAAC;IAClB,IAAI,EAAE,IAAI,CAAC;IACX,QAAQ,EAAE,QAAQ,CAAC;IACnB,aAAa,EAAE,aAAa,CAAC;IAC7B,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,MAAM,CAAC,EAAE,gBAAgB,CAAC;CAC3B;AAED,UAAU,eAAe;IACvB,MAAM,EAAE,QAAQ,GAAG,QAAQ,GAAG,OAAO,CAAC;IACtC,SAAS,EAAE,MAAM,CAAC;IAClB,cAAc,EAAE,MAAM,CAAC;IACvB,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,KAAK,CAAC;QACjB,QAAQ,EAAE,MAAM,CAAC;QACjB,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,EAAE,MAAM,CAAC;QACf,SAAS,EAAE,MAAM,CAAC;QAClB,MAAM,EAAE,MAAM,CAAC;QACf,UAAU,EAAE,MAAM,CAAC;QACnB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;QAC3B,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;QACvB,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;KAC9B,CAAC,CAAC;CACJ;AAED;;;;GAIG;AACH,wBAAsB,YAAY,CAChC,OAAO,EAAE,gBAAgB,GACxB,OAAO,CAAC,eAAe,CAAC,CA6L1B;AAID;;;GAGG;AACH,wBAAgB,YAAY,CAAC,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAQvD"}

package/dist/lib/runner.d.ts

@@ -10,13 +10,18 @@ export interface RunOptions {
     screenshotDir?: string;
 }
 export interface RunEvent {
-    type: "scenario:start" | "scenario:pass" | "scenario:fail" | "scenario:error" | "screenshot:captured" | "run:complete";
+    type: "scenario:start" | "scenario:pass" | "scenario:fail" | "scenario:error" | "screenshot:captured" | "run:complete" | "step:tool_call" | "step:tool_result" | "step:thinking";
     scenarioId?: string;
     scenarioName?: string;
     resultId?: string;
     runId?: string;
     error?: string;
     screenshotPath?: string;
+    toolName?: string;
+    toolInput?: Record<string, unknown>;
+    toolResult?: string;
+    thinking?: string;
+    stepNumber?: number;
 }
 export type RunEventHandler = (event: RunEvent) => void;
 export declare function onRunEvent(handler: RunEventHandler): void;
@@ -33,4 +38,16 @@ export declare function runByFilter(options: RunOptions & {
     run: Run;
     results: Result[];
 }>;
+/**
+ * Start a run asynchronously — creates the run record immediately and returns it,
+ * then executes scenarios in the background. Poll getRun(id) to check progress.
+ */
+export declare function startRunAsync(options: RunOptions & {
+    tags?: string[];
+    priority?: string;
+    scenarioIds?: string[];
+}): {
+    runId: string;
+    scenarioCount: number;
+};
 //# sourceMappingURL=runner.d.ts.map

package/dist/lib/runner.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/lib/runner.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,GAAG,EAAE,MAAM,EAAE,MAAM,mBAAmB,CAAC;AAW/D,MAAM,WAAW,UAAU;IACzB,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,QAAQ;IACvB,IAAI,EAAE,gBAAgB,GAAG,eAAe,GAAG,eAAe,GAAG,gBAAgB,GAAG,qBAAqB,GAAG,cAAc,CAAC;IACvH,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,MAAM,eAAe,GAAG,CAAC,KAAK,EAAE,QAAQ,KAAK,IAAI,CAAC;AAIxD,wBAAgB,UAAU,CAAC,OAAO,EAAE,eAAe,GAAG,IAAI,CAEzD;AAMD,wBAAsB,iBAAiB,CACrC,QAAQ,EAAE,QAAQ,EAClB,KAAK,EAAE,MAAM,EACb,OAAO,EAAE,UAAU,GAClB,OAAO,CAAC,MAAM,CAAC,CAoFjB;AAED,wBAAsB,QAAQ,CAC5B,SAAS,EAAE,QAAQ,EAAE,EACrB,OAAO,EAAE,UAAU,GAClB,OAAO,CAAC;IAAE,GAAG,EAAE,GAAG,CAAC;IAAC,OAAO,EAAE,MAAM,EAAE,CAAA;CAAE,CAAC,CA4D1C;AAED,wBAAsB,WAAW,CAC/B,OAAO,EAAE,UAAU,GAAG;IAAE,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,EAAE,CAAA;CAAE,GACnF,OAAO,CAAC;IAAE,GAAG,EAAE,GAAG,CAAC;IAAC,OAAO,EAAE,MAAM,EAAE,CAAA;CAAE,CAAC,CAuB1C"}
+{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/lib/runner.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,GAAG,EAAE,MAAM,EAAE,MAAM,mBAAmB,CAAC;AAW/D,MAAM,WAAW,UAAU;IACzB,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,QAAQ;IACvB,IAAI,EACA,gBAAgB,GAChB,eAAe,GACf,eAAe,GACf,gBAAgB,GAChB,qBAAqB,GACrB,cAAc,GACd,gBAAgB,GAChB,kBAAkB,GAClB,eAAe,CAAC;IACpB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACpC,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,MAAM,eAAe,GAAG,CAAC,KAAK,EAAE,QAAQ,KAAK,IAAI,CAAC;AAIxD,wBAAgB,UAAU,CAAC,OAAO,EAAE,eAAe,GAAG,IAAI,CAEzD;AAMD,wBAAsB,iBAAiB,CACrC,QAAQ,EAAE,QAAQ,EAClB,KAAK,EAAE,MAAM,EACb,OAAO,EAAE,UAAU,GAClB,OAAO,CAAC,MAAM,CAAC,CAiGjB;AAED,wBAAsB,QAAQ,CAC5B,SAAS,EAAE,QAAQ,EAAE,EACrB,OAAO,EAAE,UAAU,GAClB,OAAO,CAAC;IAAE,GAAG,EAAE,GAAG,CAAC;IAAC,OAAO,EAAE,MAAM,EAAE,CAAA;CAAE,CAAC,CA4D1C;AAED,wBAAsB,WAAW,CAC/B,OAAO,EAAE,UAAU,GAAG;IAAE,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,EAAE,CAAA;CAAE,GACnF,OAAO,CAAC;IAAE,GAAG,EAAE,GAAG,CAAC;IAAC,OAAO,EAAE,MAAM,EAAE,CAAA;CAAE,CAAC,CAuB1C;AAED;;;GAGG;AACH,wBAAgB,aAAa,CAC3B,OAAO,EAAE,UAAU,GAAG;IAAE,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,EAAE,CAAA;CAAE,GACnF;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,aAAa,EAAE,MAAM,CAAA;CAAE,CA+E1C"}

package/dist/mcp/index.js

@@ -5565,7 +5565,8 @@ async function runAgentLoop(options) {
     screenshotter,
     model,
     runId,
-    maxTurns = 30
+    maxTurns = 30,
+    onStep
   } = options;
   const systemPrompt = [
     "You are an expert QA testing agent. Your job is to thoroughly test web application scenarios.",
@@ -5624,8 +5625,8 @@ async function runAgentLoop(options) {
       }
       const toolUseBlocks = response.content.filter((block) => block.type === "tool_use");
       if (toolUseBlocks.length === 0 && response.stop_reason === "end_turn") {
-        const textBlocks = response.content.filter((block) => block.type === "text");
-        const textReasoning = textBlocks.map((b) => b.text).join(`
+        const textBlocks2 = response.content.filter((block) => block.type === "text");
+        const textReasoning = textBlocks2.map((b) => b.text).join(`
 `);
         return {
           status: "error",
@@ -5636,10 +5637,22 @@ async function runAgentLoop(options) {
         };
       }
       const toolResults = [];
+      const textBlocks = response.content.filter((block) => block.type === "text");
+      if (textBlocks.length > 0 && onStep) {
+        const thinking = textBlocks.map((b) => b.text).join(`
+`);
+        onStep({ type: "thinking", thinking, stepNumber });
+      }
       for (const toolBlock of toolUseBlocks) {
         stepNumber++;
         const toolInput = toolBlock.input;
+        if (onStep) {
+          onStep({ type: "tool_call", toolName: toolBlock.name, toolInput, stepNumber });
+        }
         const execResult = await executeTool(page, screenshotter, toolBlock.name, toolInput, { runId, scenarioSlug, stepNumber });
+        if (onStep) {
+          onStep({ type: "tool_result", toolName: toolBlock.name, toolResult: execResult.result, stepNumber });
+        }
         if (execResult.screenshot) {
           screenshots.push({
             ...execResult.screenshot,
@@ -5781,7 +5794,20 @@ async function runSingleScenario(scenario, runId, options) {
       screenshotter,
       model,
       runId,
-      maxTurns: 30
+      maxTurns: 30,
+      onStep: (stepEvent) => {
+        emit({
+          type: `step:${stepEvent.type}`,
+          scenarioId: scenario.id,
+          scenarioName: scenario.name,
+          runId,
+          toolName: stepEvent.toolName,
+          toolInput: stepEvent.toolInput,
+          toolResult: stepEvent.toolResult,
+          thinking: stepEvent.thinking,
+          stepNumber: stepEvent.stepNumber
+        });
+      }
     });
     for (const ss of agentResult.screenshots) {
       createScreenshot({
@@ -5891,6 +5917,79 @@ async function runByFilter(options) {
   }
   return runBatch(scenarios, options);
 }
+function startRunAsync(options) {
+  const config = loadConfig();
+  const model = resolveModel(options.model ?? config.defaultModel);
+  let scenarios;
+  if (options.scenarioIds && options.scenarioIds.length > 0) {
+    const all = listScenarios({ projectId: options.projectId });
+    scenarios = all.filter((s) => options.scenarioIds.includes(s.id) || options.scenarioIds.includes(s.shortId));
+  } else {
+    scenarios = listScenarios({
+      projectId: options.projectId,
+      tags: options.tags,
+      priority: options.priority
+    });
+  }
+  const parallel = options.parallel ?? 1;
+  const run = createRun({
+    url: options.url,
+    model,
+    headed: options.headed,
+    parallel,
+    projectId: options.projectId
+  });
+  if (scenarios.length === 0) {
+    updateRun(run.id, { status: "passed", total: 0, finished_at: new Date().toISOString() });
+    return { runId: run.id, scenarioCount: 0 };
+  }
+  updateRun(run.id, { status: "running", total: scenarios.length });
+  (async () => {
+    const results = [];
+    try {
+      if (parallel <= 1) {
+        for (const scenario of scenarios) {
+          const result = await runSingleScenario(scenario, run.id, options);
+          results.push(result);
+        }
+      } else {
+        const queue = [...scenarios];
+        const running = [];
+        const processNext = async () => {
+          const scenario = queue.shift();
+          if (!scenario)
+            return;
+          const result = await runSingleScenario(scenario, run.id, options);
+          results.push(result);
+          await processNext();
+        };
+        const workers = Math.min(parallel, scenarios.length);
+        for (let i = 0;i < workers; i++) {
+          running.push(processNext());
+        }
+        await Promise.all(running);
+      }
+      const passed = results.filter((r) => r.status === "passed").length;
+      const failed = results.filter((r) => r.status === "failed" || r.status === "error").length;
+      updateRun(run.id, {
+        status: failed > 0 ? "failed" : "passed",
+        passed,
+        failed,
+        total: scenarios.length,
+        finished_at: new Date().toISOString()
+      });
+      emit({ type: "run:complete", runId: run.id });
+    } catch (error) {
+      const errorMsg = error instanceof Error ? error.message : String(error);
+      updateRun(run.id, {
+        status: "failed",
+        finished_at: new Date().toISOString()
+      });
+      emit({ type: "run:complete", runId: run.id, error: errorMsg });
+    }
+  })();
+  return { runId: run.id, scenarioCount: scenarios.length };
+}
 function estimateCost(model, tokens) {
   const costs = {
     "claude-haiku-4-5-20251001": 0.1,
@@ -6343,7 +6442,7 @@ class Scheduler {
 
 // src/mcp/index.ts
 var server = new McpServer({
-  name: "testers-mcp",
+  name: "testers",
   version: "0.0.1"
 });
 server.tool("create_scenario", "Create a new test scenario", {
@@ -6457,18 +6556,15 @@ server.tool("run_scenarios", "Run test scenarios against a URL", {
   parallel: exports_external.number().optional().describe("Number of parallel workers")
 }, async ({ url, tags, scenarioIds, priority, model, headed, parallel }) => {
   try {
-    const { run, results } = await runByFilter({ url, tags, scenarioIds, priority, model, headed, parallel });
-    const passed = results.filter((r) => r.status === "passed").length;
-    const failed = results.filter((r) => r.status === "failed" || r.status === "error").length;
-    const skipped = results.filter((r) => r.status === "skipped").length;
+    const { runId, scenarioCount } = startRunAsync({ url, tags, scenarioIds, priority, model, headed, parallel });
     const text = [
-      `Run ${run.id} \u2014 ${run.status}`,
-      `URL: ${run.url}`,
-      `Total: ${results.length} | Passed: ${passed} | Failed: ${failed} | Skipped: ${skipped}`,
-      `Model: ${run.model}`,
-      `Started: ${run.startedAt}`,
-      run.finishedAt ? `Finished: ${run.finishedAt}` : null
-    ].filter(Boolean).join(`
+      `Run started: ${runId}`,
+      `Scenarios: ${scenarioCount}`,
+      `URL: ${url}`,
+      `Status: running (async)`,
+      ``,
+      `Poll with get_run to check progress.`
+    ].join(`
 `);
     return { content: [{ type: "text", text }] };
   } catch (error) {
@@ -6712,6 +6808,6 @@ async function main() {
   await server.connect(transport);
 }
 main().catch((error) => {
-  console.error("Failed to start testers-mcp:", error);
+  console.error("Failed to start testers:", error);
   process.exit(1);
 });

package/dist/server/index.js

@@ -1499,7 +1499,8 @@ async function runAgentLoop(options) {
     screenshotter,
     model,
     runId,
-    maxTurns = 30
+    maxTurns = 30,
+    onStep
   } = options;
   const systemPrompt = [
     "You are an expert QA testing agent. Your job is to thoroughly test web application scenarios.",
@@ -1558,8 +1559,8 @@ async function runAgentLoop(options) {
       }
       const toolUseBlocks = response.content.filter((block) => block.type === "tool_use");
       if (toolUseBlocks.length === 0 && response.stop_reason === "end_turn") {
-        const textBlocks = response.content.filter((block) => block.type === "text");
-        const textReasoning = textBlocks.map((b) => b.text).join(`
+        const textBlocks2 = response.content.filter((block) => block.type === "text");
+        const textReasoning = textBlocks2.map((b) => b.text).join(`
 `);
         return {
           status: "error",
@@ -1570,10 +1571,22 @@ async function runAgentLoop(options) {
         };
       }
       const toolResults = [];
+      const textBlocks = response.content.filter((block) => block.type === "text");
+      if (textBlocks.length > 0 && onStep) {
+        const thinking = textBlocks.map((b) => b.text).join(`
+`);
+        onStep({ type: "thinking", thinking, stepNumber });
+      }
       for (const toolBlock of toolUseBlocks) {
         stepNumber++;
         const toolInput = toolBlock.input;
+        if (onStep) {
+          onStep({ type: "tool_call", toolName: toolBlock.name, toolInput, stepNumber });
+        }
         const execResult = await executeTool(page, screenshotter, toolBlock.name, toolInput, { runId, scenarioSlug, stepNumber });
+        if (onStep) {
+          onStep({ type: "tool_result", toolName: toolBlock.name, toolResult: execResult.result, stepNumber });
+        }
         if (execResult.screenshot) {
           screenshots.push({
             ...execResult.screenshot,
@@ -1715,7 +1728,20 @@ async function runSingleScenario(scenario, runId, options) {
       screenshotter,
       model,
       runId,
-      maxTurns: 30
+      maxTurns: 30,
+      onStep: (stepEvent) => {
+        emit({
+          type: `step:${stepEvent.type}`,
+          scenarioId: scenario.id,
+          scenarioName: scenario.name,
+          runId,
+          toolName: stepEvent.toolName,
+          toolInput: stepEvent.toolInput,
+          toolResult: stepEvent.toolResult,
+          thinking: stepEvent.thinking,
+          stepNumber: stepEvent.stepNumber
+        });
+      }
     });
     for (const ss of agentResult.screenshots) {
       createScreenshot({

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hasna/testers",
-  "version": "0.0.3",
+  "version": "0.0.5",
   "description": "AI-powered QA testing CLI — spawns cheap AI agents to test web apps with headless browsers",
   "type": "module",
   "main": "dist/index.js",
@@ -28,7 +28,7 @@
     "build:mcp": "bun build src/mcp/index.ts --outdir dist/mcp --target bun --external @modelcontextprotocol/sdk --external @anthropic-ai/sdk --external playwright",
     "build:server": "bun build src/server/index.ts --outdir dist/server --target bun --external @anthropic-ai/sdk --external playwright",
     "build:lib": "bun build src/index.ts --outdir dist --target bun --external playwright --external @anthropic-ai/sdk --external @modelcontextprotocol/sdk",
-    "build:types": "tsc --emitDeclarationOnly --outDir dist",
+    "build:types": "NODE_OPTIONS='--max-old-space-size=8192' tsc --emitDeclarationOnly --outDir dist --skipLibCheck",
     "build:dashboard": "cd dashboard && bun run build",
     "typecheck": "tsc --noEmit",
     "test": "bun test",