vibeview 0.39.0 → 0.41.0

package/dist/{chunk-MYXUGCR5.js → chunk-GDTLLJFL.js}

@@ -274,7 +274,7 @@ var ApiClient = class {
   }
   async getRunStatus(runId) {
     const data = await this.request("GET", `/runs/${runId}`);
-    return { ...data, url: data.run_url ?? `${this.baseUrl}/tests/runs/${runId}` };
+    return { ...data, url: runResultUrl(data, `${this.baseUrl}/tests/runs/${runId}`) };
   }
   /** Request without /ci prefix — for non-CI endpoints like /apps, /tests */
   async requestRaw(method, path, body) {
@@ -772,6 +772,13 @@ var ApiClient = class {
     return this.requestRaw("POST", `/webhooks/${id}/test`);
   }
 };
+function runResultUrl(data, fallback) {
+  const url = data.run_url ?? fallback;
+  if (!data.retry_run_id) return url;
+  const firstId = url.slice(url.lastIndexOf("/") + 1);
+  if (!firstId) return url;
+  return `${url.slice(0, url.length - firstId.length)}${encodeURIComponent(data.retry_run_id)}`;
+}
 function parseContentDispositionFilename(header) {
   if (!header) return void 0;
   const extMatch = header.match(/filename\*\s*=\s*[^']*''([^;]+)/i);
@@ -870,6 +877,10 @@ function createSpinner(text, opts = {}) {
 var EXIT_PASS = 0;
 var EXIT_TEST_FAIL = 1;
 var EXIT_INFRA_ERROR = 2;
+var RUN_COMMANDS = ["run-test", "run-suite", "run-suites"];
+function runCommandParseExitCode(err) {
+  return err.exitCode === 0 ? EXIT_PASS : EXIT_INFRA_ERROR;
+}
 var UsageError = class extends Error {
   constructor(message) {
     super(message);
@@ -1015,6 +1026,8 @@ export {
   EXIT_PASS,
   EXIT_TEST_FAIL,
   EXIT_INFRA_ERROR,
+  RUN_COMMANDS,
+  runCommandParseExitCode,
   UsageError,
   DEVICE_STATUS_FILTERS,
   queueReasonCopy,

package/dist/{chunk-EQN3X62U.js → chunk-MK3WDPVZ.js}

@@ -13,7 +13,7 @@ import {
   resolveArtifact,
   resolveAuth,
   uploadAppBuild
-} from "./chunk-MYXUGCR5.js";
+} from "./chunk-GDTLLJFL.js";
 
 // src/lib/formatters/not-checked.ts
 function notCheckedSteps(steps) {

package/dist/cli.js

@@ -30,7 +30,7 @@ import {
   runAgentVerb,
   runStepsNotChecked,
   stepsNotCheckedLabel
-} from "./chunk-EQN3X62U.js";
+} from "./chunk-MK3WDPVZ.js";
 import {
   ApiClient,
   ApiStatusError,
@@ -40,6 +40,7 @@ import {
   EXIT_TEST_FAIL,
   NetworkError,
   NotFoundError,
+  RUN_COMMANDS,
   UsageError,
   createSpinner,
   errorExitCode,
@@ -49,9 +50,10 @@ import {
   queueReasonCopy,
   reportCommandError,
   resolveAuth,
+  runCommandParseExitCode,
   saveConfig,
   uploadApp
-} from "./chunk-MYXUGCR5.js";
+} from "./chunk-GDTLLJFL.js";
 
 // src/cli.ts
 import { createRequire } from "module";
@@ -125,9 +127,17 @@ function formatHuman(result, verbose) {
   const notCheckedCases = cases.filter((t) => outcomeOf(t) === "not_checked").length;
   const total = cases.length;
   const notCheckedNote = notCheckedCases > 0 ? `, ${notCheckedCases} not checked` : notChecked > 0 ? `, ${stepsNotCheckedLabel(notChecked)}` : "";
-  lines.push(`  ${chalk.dim(`${passed}/${total} cases passed${notCheckedNote}`)}`);
+  const quarantined = cases.filter((t) => t.status === "failed" && t.quarantined === true).length;
+  const quarantinedNote = quarantined > 0 && (result.test_runs ?? []).length > 0 ? `, ${quarantined} failed (quarantined)` : "";
+  lines.push(
+    `  ${chalk.dim(`${passed}/${total} cases passed${notCheckedNote}${quarantinedNote}`)}`
+  );
   if (result.url) {
-    lines.push(`  ${chalk.dim(result.url)}`);
+    lines.push(`  ${chalk.dim(result.first_attempt ? `Retry: ${result.url}` : result.url)}`);
+  }
+  if (result.first_attempt) {
+    const first = result.first_attempt;
+    lines.push(`  ${chalk.dim(`First attempt (${first.status ?? "unknown"}): ${first.url}`)}`);
   }
   const suiteCases = result.test_runs ?? [];
   if (suiteCases.length > 0) {
@@ -647,11 +657,13 @@ function formatToolCall(tc) {
 function runStartedLine(evt) {
   const steps = `${evt.total_steps} step(s)`;
   const model = chalk2.bold(evt.model || "default");
-  if (evt.mode === "replay") return `${chalk2.cyan("\u25B6")} Replay started \u2014 ${steps} (no AI)`;
+  if (evt.mode === "replay") {
+    return `${chalk2.cyan("\u25B6")} No AI run started \u2014 ${steps}, replays only (no AI)`;
+  }
   if (evt.mode === "hybrid") {
-    return `${chalk2.cyan("\u25B6")} Hybrid test started \u2014 ${steps}, replaying the recording, AI fallback model: ${model}`;
+    return `${chalk2.cyan("\u25B6")} Smart run started \u2014 ${steps}, replays recorded and learned paths, AI where needed (model: ${model})`;
   }
-  return `${chalk2.cyan("\u25B6")} AI test started \u2014 ${steps}, model: ${model}`;
+  return `${chalk2.cyan("\u25B6")} AI only run started \u2014 ${steps}, model: ${model}`;
 }
 function displayProgressEvent(evt, ctx) {
   const { spinner } = ctx;
@@ -765,6 +777,37 @@ function parseRunPlatform(value) {
   }
   return value;
 }
+var RUN_MODES = ["hybrid", "ai", "replay"];
+function parseRunMode(value) {
+  if (value === void 0) return void 0;
+  if (!RUN_MODES.includes(value)) {
+    throw new Error(
+      `--mode must be one of ${RUN_MODES.join(", ")} (Smart, AI only, No AI), got "${value}"`
+    );
+  }
+  return value;
+}
+var MAX_TIMEOUT_S = 7200;
+var POSITIVE_INT = /^[1-9][0-9]*$/;
+function parseTimeoutSeconds(value) {
+  const n = POSITIVE_INT.test(value.trim()) ? Number(value.trim()) : NaN;
+  if (!Number.isSafeInteger(n) || n < 1 || n > MAX_TIMEOUT_S) {
+    throw new Error(
+      `--timeout must be a whole number of seconds from 1 to ${MAX_TIMEOUT_S}, got "${value}"`
+    );
+  }
+  return n;
+}
+function parseBuildId(value) {
+  if (value === void 0) return void 0;
+  const n = POSITIVE_INT.test(value.trim()) ? Number(value.trim()) : NaN;
+  if (!Number.isSafeInteger(n)) {
+    throw new Error(
+      `--build must be a build id (a positive whole number, see \`vibeview list-builds\`), got "${value}"`
+    );
+  }
+  return n;
+}
 function refusePlatformWithApp(options) {
   if (options.platform && options.app) {
     throw new Error(
@@ -1181,15 +1224,16 @@ Aggregate: ${aggregate}${aggregateNote}
       for (const cr of sr.case_runs ?? []) {
         const outcome = outcomeOf(cr);
         const neutral = outcome === "not_checked" || outcome === "not_run";
-        const caseMark = neutral ? chalk5.gray("\u25CB") : cr.status === "passed" ? chalk5.green("\u2713") : cr.status === "cancelled" || cr.status === "pending" || cr.status === "running" ? chalk5.yellow("-") : chalk5.red("\u2717");
+        const caseMark = neutral ? chalk5.gray("\u25CB") : cr.status === "passed" ? chalk5.green("\u2713") : cr.status === "cancelled" || cr.status === "pending" || cr.status === "running" ? chalk5.yellow("-") : cr.status === "failed" && cr.quarantined === true ? chalk5.yellow("\u2717") : chalk5.red("\u2717");
         const dur = cr.duration_ms ? chalk5.dim(` (${(cr.duration_ms / 1e3).toFixed(1)}s)`) : "";
         const unchecked = notCheckedSuffix(outcome, cr.steps_not_checked);
-        const label = neutral ? outcome.replace("_", " ") : cr.status ?? "?";
+        const quarantined = cr.status === "failed" && cr.quarantined === true;
+        const label = neutral ? outcome.replace("_", " ") : quarantined ? "failed (quarantined)" : cr.status ?? "?";
         process.stdout.write(
           `      ${caseMark} ${cr.case_name ?? "Case"} ${label}${unchecked ? chalk5.yellow(unchecked) : ""}${dur} ${chalk5.dim(cr.url)}
 `
         );
-        if (cr.error_message && neutral) {
+        if (cr.error_message && (neutral || cr.status === "failed" && cr.quarantined === true)) {
           process.stdout.write(`          ${chalk5.dim(cr.error_message)}
 `);
         } else if (cr.error_message && cr.status !== "passed") {
@@ -1205,7 +1249,15 @@ Aggregate: ${aggregate}${aggregateNote}
 }
 function suiteCounts(sr) {
   const nc = sr.not_checked_tests ?? 0;
-  return `${sr.passed} passed, ${sr.failed} failed${nc > 0 ? `, ${nc} not checked` : ""}`;
+  const q = Math.min(
+    sr.failed,
+    sr.quarantined_failures ?? (sr.case_runs ?? []).filter((c) => c.status === "failed" && c.quarantined === true).length
+  );
+  const parts = [`${sr.passed} passed`];
+  if (q === 0 || sr.failed > q) parts.push(`${sr.failed - q} failed`);
+  if (q > 0) parts.push(`${q} failed (quarantined)`);
+  if (nc > 0) parts.push(`${nc} not checked`);
+  return parts.join(", ");
 }
 
 // src/commands/run-test.ts
@@ -4060,11 +4112,27 @@ function resolveMetadata(opts) {
     process.exit(2);
   }
 }
+function parseRunFlags(opts) {
+  try {
+    return {
+      mode: parseRunMode(opts.mode),
+      timeout: parseTimeoutSeconds(opts.timeout),
+      build: parseBuildId(opts.build)
+    };
+  } catch (err) {
+    process.stderr.write(`error: ${err.message}
+`);
+    process.exit(2);
+  }
+}
 var program = new Command();
 program.name("vibeview").description(
   "VibeView CLI -- run your app on cloud iOS simulators, Android emulators, Apple TV, Android TV and Roku devices: a React Native dev loop (dev), device control for you or your coding agent (ui-tree, tap, mcp), and tests from CI (run-test, run-suites)"
 ).version(pkg.version);
-program.command("run-suite <suite-id>").description("Run all test cases in a suite").option("--token <token>", "API token (or set VIBEVIEW_API_TOKEN)").option("--output <format>", "Output format: human, json, junit", "human").option("--verbose", "Show per-case breakdown", false).option("--timeout <seconds>", "Max wait time in seconds (queue wait plus execution)", "1800").option("--commit-sha <sha>", "Git commit SHA for GitHub status checks").option("--model <model>", "AI model to use").option("--provider <provider>", "AI provider").option("--mode <mode>", "Execution mode: hybrid, ai, replay").option(
+program.command("run-suite <suite-id>").description("Run all test cases in a suite").option("--token <token>", "API token (or set VIBEVIEW_API_TOKEN)").option("--output <format>", "Output format: human, json, junit", "human").option("--verbose", "Show per-case breakdown", false).option("--timeout <seconds>", "Max wait time in seconds (queue wait plus execution)", "1800").option("--commit-sha <sha>", "Git commit SHA for GitHub status checks").option("--model <model>", "AI model to use").option("--provider <provider>", "AI provider").option(
+  "--mode <mode>",
+  "Execution mode: hybrid (Smart, the default), ai (AI only), replay (No AI)"
+).option(
   "--device-category <category>",
   "Device category: phone, tablet, tv (default: tv for a TV platform, else phone)",
   parseDeviceCategory
@@ -4093,12 +4161,12 @@ program.command("run-suite <suite-id>").description("Run all test cases in a sui
 `);
     process.exit(2);
   }
+  const runFlags = parseRunFlags(opts);
   const metadata = resolveMetadata(opts);
   return runSuite(suiteId, {
     ...opts,
     metadata,
-    timeout: parseInt(opts.timeout, 10),
-    build: opts.build ? parseInt(opts.build, 10) : void 0
+    ...runFlags
   });
 });
 program.command("run-suites").description("Run all suites matching a tag filter").requiredOption(
@@ -4109,7 +4177,10 @@ program.command("run-suites").description("Run all suites matching a tag filter"
 ).option("--match <mode>", "any (default) or all", "any").option(
   "--platform <platform>",
   "Filter matched suites to one platform: ios, android, tvos, androidtv, roku (default: no filter)"
-).option("--token <token>", "API token (or set VIBEVIEW_API_TOKEN)").option("--output <format>", "Output format: human, json, junit", "human").option("--verbose", "Show per-suite progress", false).option("--timeout <seconds>", "Per-suite timeout, capped at 2 hours total", "1800").option("--commit-sha <sha>", "Git commit SHA propagated to every SuiteRun").option("--model <model>", "AI model").option("--provider <provider>", "AI provider").option("--mode <mode>", "Execution mode: hybrid, ai, replay").option(
+).option("--token <token>", "API token (or set VIBEVIEW_API_TOKEN)").option("--output <format>", "Output format: human, json, junit", "human").option("--verbose", "Show per-suite progress", false).option("--timeout <seconds>", "Per-suite timeout, capped at 2 hours total", "1800").option("--commit-sha <sha>", "Git commit SHA propagated to every SuiteRun").option("--model <model>", "AI model").option("--provider <provider>", "AI provider").option(
+  "--mode <mode>",
+  "Execution mode: hybrid (Smart, the default), ai (AI only), replay (No AI)"
+).option(
   "--device-category <category>",
   "Device category: phone, tablet, tv (default: tv for a TV platform, else phone)",
   parseDeviceCategory
@@ -4131,15 +4202,18 @@ program.command("run-suites").description("Run all suites matching a tag filter"
 `);
     process.exit(2);
   }
+  const runFlags = parseRunFlags(opts);
   const metadata = resolveMetadata(opts);
   return runSuites({
     ...opts,
     metadata,
-    timeout: parseInt(opts.timeout, 10),
-    build: opts.build ? parseInt(opts.build, 10) : void 0
+    ...runFlags
   });
 });
-program.command("run-test <test-id>").description("Run a single test case").option("--token <token>", "API token").option("--output <format>", "Output format: human, json, junit", "human").option("--verbose", "Show step details", false).option("--timeout <seconds>", "Max wait time in seconds (queue wait plus execution)", "1800").option("--commit-sha <sha>", "Git commit SHA for GitHub status checks").option("--model <model>", "AI model to use").option("--provider <provider>", "AI provider").option("--mode <mode>", "Execution mode: hybrid, ai, replay").option(
+program.command("run-test <test-id>").description("Run a single test case").option("--token <token>", "API token").option("--output <format>", "Output format: human, json, junit", "human").option("--verbose", "Show step details", false).option("--timeout <seconds>", "Max wait time in seconds (queue wait plus execution)", "1800").option("--commit-sha <sha>", "Git commit SHA for GitHub status checks").option("--model <model>", "AI model to use").option("--provider <provider>", "AI provider").option(
+  "--mode <mode>",
+  "Execution mode: hybrid (Smart, the default), ai (AI only), replay (No AI)"
+).option(
   "--device-category <category>",
   "Device category: phone, tablet, tv (default: tv for a TV platform, else phone)",
   parseDeviceCategory
@@ -4168,12 +4242,12 @@ program.command("run-test <test-id>").description("Run a single test case").opti
 `);
     process.exit(2);
   }
+  const runFlags = parseRunFlags(opts);
   const metadata = resolveMetadata(opts);
   return runTest(testId, {
     ...opts,
     metadata,
-    timeout: parseInt(opts.timeout, 10),
-    build: opts.build ? parseInt(opts.build, 10) : void 0
+    ...runFlags
   });
 });
 program.command("list-suites").description("List available test suites").option("--token <token>", "API token").option("--output <format>", "Output format: human, json", "human").option(
@@ -4387,7 +4461,7 @@ program.command("dev-reload").description(
 program.command("mcp").description(
   "Run an MCP stdio server exposing the agent verbs (tap, swipe, type, ...) plus dev_start/dev_stop/dev_reload as tools"
 ).option("--token <token>", "API token (or set VIBEVIEW_API_TOKEN)").action(async (opts) => {
-  const { runMcpServer } = await import("./server-GEY4WZUH.js");
+  const { runMcpServer } = await import("./server-435X37GD.js");
   await runMcpServer({ ...opts, version: pkg.version });
 });
 program.command("agent-setup").description(
@@ -4411,6 +4485,9 @@ ${pressButtonsHelp()}`);
     return runAgentVerb(cmd, positionals, opts);
   });
 }
+for (const name of RUN_COMMANDS) {
+  program.commands.find((c) => c.name() === name)?.exitOverride((err) => process.exit(runCommandParseExitCode(err)));
+}
 if (updateCheckAllowed(process.argv, process.env, process.stderr.isTTY)) {
   void checkForUpdate(pkg.version, { print: (line) => process.stderr.write(`${line}
 `) });

package/dist/{server-GEY4WZUH.js → server-435X37GD.js}

@@ -23,18 +23,19 @@ import {
   outcomeOf,
   parseBuildId,
   parseIdleWarningSeconds,
+  platformLabel,
   platformWithArticle,
   readDevState,
   recordDevUse,
   renderEnvelope,
   resolveMetro,
   startDevSession
-} from "./chunk-EQN3X62U.js";
+} from "./chunk-MK3WDPVZ.js";
 import {
   ApiClient,
   NotFoundError,
   resolveAuth
-} from "./chunk-MYXUGCR5.js";
+} from "./chunk-GDTLLJFL.js";
 
 // src/mcp/server.ts
 import { writeFileSync } from "fs";
@@ -47,7 +48,7 @@ var SESSION_ID_PROPERTY = {
 };
 var AI_USE_PROPERTY = {
   type: "string",
-  description: "smart (default): replays saved paths and uses AI only where the screen changed; no_ai: replays only, no usage credit; ai_only: the AI does every step.",
+  description: "smart (default): replays recorded and learned paths, and uses AI only where the screen changed or a step has no path yet; no_ai: replays only, no AI and no usage credit (a step with no path is not checked); ai_only: the AI does every step.",
   enum: ["smart", "no_ai", "ai_only"]
 };
 var AI_USE_MODE = { smart: "hybrid", no_ai: "replay", ai_only: "ai" };
@@ -59,7 +60,12 @@ var CREATE_TEST_DEF = {
     app: { type: "string", description: "App id (from list_apps)." },
     name: { type: "string", description: "Test name." },
     steps: { type: "string", description: "The steps, one per line." },
-    group: { type: "string", description: "Group id to add the test to (optional)." }
+    group: { type: "string", description: "Group id to add the test to (optional)." },
+    if_exists: {
+      type: "string",
+      description: "When a test with this name already exists for the app: update replaces it, new creates another one beside it, fail refuses. Omitted: an identical test is reused, a different one is refused.",
+      enum: ["fail", "update", "new"]
+    }
   },
   required: ["app", "name", "steps"]
 };
@@ -72,7 +78,7 @@ var RUN_TEST_DEF = {
     session_id: { type: "string", description: "Run on this session instead of a new device." },
     device_model: {
       type: "string",
-      description: "Device model for a new device (from list_device_models); omitted: any free one."
+      description: "Device model for a new device (from list_device_models); omitted: any free one. A model of another platform (same form factor) runs the test's app for that platform: the app linked in its app family, else the app with the same bundle id / package. The reply names the platform and app the run went to."
     },
     ai_use: AI_USE_PROPERTY,
     start_from_current_screen: {
@@ -94,7 +100,7 @@ var RUN_GROUP_DEF = {
     session_id: { type: "string", description: "Run on this session instead of a new device." },
     device_model: {
       type: "string",
-      description: "Device model for a new device; omitted: any free one."
+      description: "Device model for a new device; omitted: any free one. A model of another platform (same form factor) runs the group's app for that platform: the app linked in its app family, else the app with the same bundle id / package. The reply names the platform and app the run went to."
     },
     ai_use: AI_USE_PROPERTY,
     start_from_current_screen: {
@@ -347,6 +353,14 @@ async function explainInactive(sessionId, deps) {
 function findCommand(toolName) {
   return AGENT_COMMANDS.find((c) => toolNameFor(c.verb) === toolName);
 }
+function testToolErrorMessage(err) {
+  const message = err instanceof Error ? err.message : String(err);
+  if (err instanceof NotFoundError && message.startsWith("Not found: ")) {
+    const detail = message.slice("Not found: ".length);
+    if (/not found/i.test(detail)) return detail;
+  }
+  return message;
+}
 async function handleCreateTest(args, deps) {
   const steps = String(args.steps ?? "").split("\n").map((s) => s.trim()).filter(Boolean).map((text) => ({ text }));
   if (!steps.length) return errorResult("create_test: `steps` needs at least one line");
@@ -355,7 +369,8 @@ async function handleCreateTest(args, deps) {
       name: String(args.name),
       app_id: String(args.app),
       steps,
-      group_ids: typeof args.group === "string" && args.group ? [args.group] : void 0
+      group_ids: typeof args.group === "string" && args.group ? [args.group] : void 0,
+      if_exists: args.if_exists === "fail" || args.if_exists === "update" || args.if_exists === "new" ? args.if_exists : void 0
     });
     const verb = created.outcome === "unchanged" ? "Test already exists, unchanged" : created.outcome === "updated" ? "Updated existing test" : "Created test";
     return textResult(
@@ -363,9 +378,17 @@ async function handleCreateTest(args, deps) {
       created
     );
   } catch (err) {
-    return errorResult(`error: ${err instanceof Error ? err.message : String(err)}`);
+    return errorResult(`error: ${testToolErrorMessage(err)}`);
   }
 }
+function runDestination(lanes) {
+  return lanes.map((l) => {
+    const label = platformLabel(l.platform);
+    const device = l.model ? ` (${l.model})` : "";
+    const app = l.app_name ? `: ${l.app_name}, the ${label} app` : "";
+    return ` Running on ${label}${device}${app}.`;
+  }).join("");
+}
 async function handleRun(kind, args, deps) {
   const id = String(kind === "test" ? args.test : args.group);
   const mode = typeof args.ai_use === "string" ? AI_USE_MODE[args.ai_use] : void 0;
@@ -387,12 +410,13 @@ async function handleRun(kind, args, deps) {
       kind === "test" ? { target: "test", test_ids: [id], devices, mode } : { target: "group", group_id: id, devices, mode }
     );
     const queued = r.lanes.some((l) => l.status === "queued") ? " It is waiting for a free device." : "";
+    const where = runDestination(r.lanes);
     return textResult(
-      `Started run ${r.batch_id}.${queued} Check it with get_run run_id=${r.batch_id}.`,
+      `Started run ${r.batch_id}.${where}${queued} Check it with get_run run_id=${r.batch_id}.`,
       r
     );
   } catch (err) {
-    return errorResult(`error: ${err instanceof Error ? err.message : String(err)}`);
+    return errorResult(`error: ${testToolErrorMessage(err)}`);
   }
 }
 function notCheckedNote(status, n) {
@@ -407,11 +431,25 @@ function agentStatus(r) {
 function testsNotChecked(n) {
   return typeof n === "number" && n > 0 ? `, ${n} not checked` : "";
 }
+function quarantinedFailure(t) {
+  return t?.status === "failed" && t?.quarantined === true;
+}
+function quarantinedCount(n) {
+  return typeof n === "number" && n > 0 ? `, ${n} failed (quarantined)` : "";
+}
 function formatRunForAgent(run) {
   const lines = [];
   if (run.lanes) {
+    const rowQuarantined = (r) => typeof r.tests_quarantined === "number" ? r.tests_quarantined : (r.tests ?? []).filter(quarantinedFailure).length;
+    let totalQuarantined = run.summary.tests_quarantined;
+    if (typeof totalQuarantined !== "number") {
+      totalQuarantined = 0;
+      for (const lane of run.lanes) {
+        for (const r of lane.runs ?? []) totalQuarantined += rowQuarantined(r);
+      }
+    }
     lines.push(
-      `${run.name}: ${agentStatus(run)} \u2014 ${run.summary.tests_passed} of ${run.summary.tests_total} tests passed${testsNotChecked(run.summary.tests_not_checked)}`
+      `${run.name}: ${agentStatus(run)} \u2014 ${run.summary.tests_passed} of ${run.summary.tests_total} tests passed${testsNotChecked(run.summary.tests_not_checked)}${quarantinedCount(totalQuarantined)}`
     );
     for (const lane of run.lanes) {
       const detail = [lane.platform, lane.os_version].filter(Boolean).join(" ");
@@ -422,10 +460,32 @@ function formatRunForAgent(run) {
       if (lane.error_message) head += ` \u2014 ${lane.error_message}`;
       lines.push(head);
       for (const r of lane.runs ?? []) {
-        const count = r.kind === "group" && typeof r.tests_total === "number" ? ` (${r.tests_passed ?? 0}/${r.tests_total} tests passed${testsNotChecked(r.tests_not_checked)})` : "";
+        const count = r.kind === "group" && typeof r.tests_total === "number" ? ` (${r.tests_passed ?? 0}/${r.tests_total} tests passed${testsNotChecked(r.tests_not_checked)}${quarantinedCount(rowQuarantined(r))})` : "";
+        const cell = r.kind === "test" ? r.tests?.[0] ?? {} : {};
+        const q = r.kind === "test" && quarantinedFailure(cell) ? " (quarantined)" : "";
+        const why = r.error_message || cell.not_run_reason;
+        const unchecked = r.kind === "test" ? notCheckedNote(outcomeOf(cell), cell.steps_not_checked) : "";
         lines.push(
-          `  ${r.name}: ${agentStatus(r)}${count}${r.error_message ? ` \u2014 ${r.error_message}` : ""}`
+          `  ${r.name}: ${agentStatus(r)}${q}${count}${why ? ` \u2014 ${why}` : ""}${unchecked}`
         );
+        if (r.kind !== "group") continue;
+        for (const t of r.tests ?? []) {
+          if (t.status === "not_run") {
+            const reason = t.not_run_reason || t.error_message;
+            lines.push(`    ${t.name}: not run${reason ? ` \u2014 ${reason}` : ""}`);
+            continue;
+          }
+          const note = notCheckedNote(outcomeOf(t), t.steps_not_checked);
+          if (note) {
+            lines.push(`    ${t.name}: ${agentStatus(t)}${note}`);
+            continue;
+          }
+          if (t.status !== "failed" && t.status !== "error") continue;
+          const tq = quarantinedFailure(t) ? " (quarantined)" : "";
+          lines.push(
+            `    ${t.name}: ${t.status}${tq}${t.error_message ? ` \u2014 ${t.error_message}` : ""}`
+          );
+        }
       }
     }
   } else if (Array.isArray(run.test_runs)) {
@@ -463,7 +523,7 @@ async function handleGetRun(args, deps) {
     const run = kind.kind === "batch" ? await client.getRunBatch(id) : await client.getRunStatus(id);
     return textResult(formatRunForAgent(run), run);
   } catch (err) {
-    return errorResult(`error: ${err instanceof Error ? err.message : String(err)}`);
+    return errorResult(`error: ${testToolErrorMessage(err)}`);
   }
 }
 async function dispatchToolCall(toolName, args, deps, extra = {}) {
@@ -882,7 +942,7 @@ async function handleUploadApp(args, deps) {
   }
   const name = typeof args.name === "string" && args.name ? args.name : void 0;
   const upload = deps.uploadApp ?? (async (f, n) => {
-    const { uploadAppBuild } = await import("./upload-app-UHJLFQKZ.js");
+    const { uploadAppBuild } = await import("./upload-app-ZEQMGZIT.js");
     const { result: result2 } = await uploadAppBuild(f, { ...deps.auth, name: n });
     return result2;
   });
@@ -1278,6 +1338,23 @@ var MCP_INSTRUCTIONS = [
   "A session stopped elsewhere is reported as ended; call dev_start again for a new one.",
   'A session nobody uses ends after the idle timeout. A result that ends with "Session \u2026 ends in N s for inactivity" means act (tap, type, swipe, press) to keep it, or dev_stop if you are done; ui_tree, screenshot, logs, find and wait do not count as use.'
 ].join("\n");
+function toolInputSchema(def, z) {
+  const shape = {};
+  for (const [key, prop] of Object.entries(def.properties)) {
+    const base = prop.enum ? z.enum(prop.enum) : z[prop.type]();
+    const described = base.describe(prop.description);
+    shape[key] = def.required.includes(key) ? described : described.optional();
+  }
+  const valid = Object.keys(def.properties);
+  return z.strictObject(shape, {
+    error: (issue) => {
+      if (issue.code !== "unrecognized_keys") return void 0;
+      const unknown = issue.keys.map((k) => `"${k}"`).join(", ");
+      const takes = valid.length ? `${def.name} takes: ${valid.join(", ")}.` : `${def.name} takes no arguments.`;
+      return `Unknown argument${issue.keys.length === 1 ? "" : "s"} ${unknown}. ${takes}`;
+    }
+  });
+}
 async function createMcpServer(opts) {
   const [{ McpServer }, { z }] = await Promise.all([
     import("@modelcontextprotocol/sdk/server/mcp.js"),
@@ -1296,21 +1373,12 @@ async function createMcpServer(opts) {
     { name: "vibeview", version: opts.version ?? "0.0.0" },
     { instructions: MCP_INSTRUCTIONS }
   );
-  function zodShape(def) {
-    const shape = {};
-    for (const [key, prop] of Object.entries(def.properties)) {
-      const base = prop.enum ? z.enum(prop.enum) : z[prop.type]();
-      const described = base.describe(prop.description);
-      shape[key] = def.required.includes(key) ? described : described.optional();
-    }
-    return shape;
-  }
   for (const def of buildToolDefinitions()) {
     server.registerTool(
       def.name,
       {
         description: def.description,
-        inputSchema: zodShape(def),
+        inputSchema: toolInputSchema(def, z),
         annotations: { readOnlyHint: !def.mutating }
       },
       async (args, extra) => {
@@ -1397,7 +1465,10 @@ export {
   idleNoticeText,
   mcpText,
   progressReporter,
+  runDestination,
   runMcpServer,
   startProgress,
-  startedDevice
+  startedDevice,
+  testToolErrorMessage,
+  toolInputSchema
 };

package/dist/{upload-app-UHJLFQKZ.js → upload-app-ZEQMGZIT.js}

@@ -3,7 +3,7 @@ import {
   describeBuild,
   uploadApp,
   uploadAppBuild
-} from "./chunk-MYXUGCR5.js";
+} from "./chunk-GDTLLJFL.js";
 export {
   describeBuild,
   uploadApp,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vibeview",
-  "version": "0.39.0",
+  "version": "0.41.0",
   "description": "VibeView CLI -- run your app on cloud iOS simulators, Android emulators, Apple TV, Android TV and Roku devices: a React Native dev loop, device control for you or your coding agent (CLI or MCP), and tests from CI",
   "mcpName": "io.github.vibeview/vibeview",
   "type": "module",