@hasna/testers 0.0.15 → 0.0.16

package/dist/cli/index.js

@@ -6,39 +6,60 @@ var __defProp = Object.defineProperty;
 var __getOwnPropNames = Object.getOwnPropertyNames;
 var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
 var __hasOwnProp = Object.prototype.hasOwnProperty;
+function __accessProp(key) {
+  return this[key];
+}
+var __toESMCache_node;
+var __toESMCache_esm;
 var __toESM = (mod, isNodeMode, target) => {
+  var canCache = mod != null && typeof mod === "object";
+  if (canCache) {
+    var cache = isNodeMode ? __toESMCache_node ??= new WeakMap : __toESMCache_esm ??= new WeakMap;
+    var cached = cache.get(mod);
+    if (cached)
+      return cached;
+  }
   target = mod != null ? __create(__getProtoOf(mod)) : {};
   const to = isNodeMode || !mod || !mod.__esModule ? __defProp(target, "default", { value: mod, enumerable: true }) : target;
   for (let key of __getOwnPropNames(mod))
     if (!__hasOwnProp.call(to, key))
       __defProp(to, key, {
-        get: () => mod[key],
+        get: __accessProp.bind(mod, key),
         enumerable: true
       });
+  if (canCache)
+    cache.set(mod, to);
   return to;
 };
-var __moduleCache = /* @__PURE__ */ new WeakMap;
 var __toCommonJS = (from) => {
-  var entry = __moduleCache.get(from), desc;
+  var entry = (__moduleCache ??= new WeakMap).get(from), desc;
   if (entry)
     return entry;
   entry = __defProp({}, "__esModule", { value: true });
-  if (from && typeof from === "object" || typeof from === "function")
-    __getOwnPropNames(from).map((key) => !__hasOwnProp.call(entry, key) && __defProp(entry, key, {
-      get: () => from[key],
-      enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable
-    }));
+  if (from && typeof from === "object" || typeof from === "function") {
+    for (var key of __getOwnPropNames(from))
+      if (!__hasOwnProp.call(entry, key))
+        __defProp(entry, key, {
+          get: __accessProp.bind(from, key),
+          enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable
+        });
+  }
   __moduleCache.set(from, entry);
   return entry;
 };
+var __moduleCache;
 var __commonJS = (cb, mod) => () => (mod || cb((mod = { exports: {} }).exports, mod), mod.exports);
+var __returnValue = (v) => v;
+function __exportSetter(name, newValue) {
+  this[name] = __returnValue.bind(null, newValue);
+}
 var __export = (target, all) => {
   for (var name in all)
     __defProp(target, name, {
       get: all[name],
       enumerable: true,
       configurable: true,
-      set: (newValue) => all[name] = () => newValue
+      set: __exportSetter.bind(all, name)
     });
 };
 var __esm = (fn, res) => () => (fn && (res = fn(fn = 0)), res);
@@ -2122,9 +2143,12 @@ function scenarioFromRow(row) {
     assertions: JSON.parse(row.assertions || "[]"),
     personaId: row.persona_id ?? null,
     scenarioType: row.scenario_type ?? "browser",
+    requiredRole: row.required_role ?? null,
     version: row.version,
     createdAt: row.created_at,
-    updatedAt: row.updated_at
+    updatedAt: row.updated_at,
+    lastPassedAt: row.last_passed_at ?? null,
+    lastPassedUrl: row.last_passed_url ?? null
   };
 }
 function runFromRow(row) {
@@ -2164,7 +2188,8 @@ function resultFromRow(row) {
     metadata: row.metadata ? JSON.parse(row.metadata) : null,
     createdAt: row.created_at,
     personaId: row.persona_id ?? null,
-    personaName: row.persona_name ?? null
+    personaName: row.persona_name ?? null,
+    failureAnalysis: row.failure_analysis ? JSON.parse(row.failure_analysis) : null
   };
 }
 function screenshotFromRow(row) {
@@ -2232,6 +2257,7 @@ function flowFromRow(row) {
   };
 }
 function personaFromRow(row) {
+  const hasAuth = row.auth_email && row.auth_password;
   return {
     id: row.id,
     shortId: row.short_id,
@@ -2240,13 +2266,23 @@ function personaFromRow(row) {
     description: row.description,
     role: row.role,
     instructions: row.instructions,
-    traits: JSON.parse(row.traits),
-    goals: JSON.parse(row.goals),
+    traits: JSON.parse(row.traits || "[]"),
+    goals: JSON.parse(row.goals || "[]"),
+    behaviors: JSON.parse(row.behaviors || "[]"),
+    expertiseLevel: row.expertise_level || "intermediate",
+    demographics: JSON.parse(row.demographics || "{}"),
+    painPoints: JSON.parse(row.pain_points || "[]"),
     metadata: row.metadata ? JSON.parse(row.metadata) : null,
     enabled: row.enabled === 1,
     version: row.version,
     createdAt: row.created_at,
-    updatedAt: row.updated_at
+    updatedAt: row.updated_at,
+    auth: hasAuth ? {
+      email: row.auth_email,
+      password: row.auth_password,
+      loginPath: row.auth_login_path ?? "/login",
+      cookies: row.auth_cookies ? JSON.parse(row.auth_cookies) : null
+    } : null
   };
 }
 function apiCheckFromRow(row) {
@@ -2288,12 +2324,14 @@ function apiCheckResultFromRow(row) {
     createdAt: row.created_at
   };
 }
-var MODEL_MAP, VersionConflictError, BrowserError, AIClientError, TodosConnectionError, ScheduleNotFoundError, DependencyCycleError;
+var MODEL_MAP, VersionConflictError, BrowserError, AIClientError, TodosConnectionError, ScheduleNotFoundError, BudgetExceededError, DependencyCycleError;
 var init_types = __esm(() => {
   MODEL_MAP = {
     quick: "claude-haiku-4-5-20251001",
     thorough: "claude-sonnet-4-6-20260311",
-    deep: "claude-opus-4-6-20260311"
+    deep: "claude-opus-4-6-20260311",
+    "cerebras-fast": "llama-3.1-8b",
+    "cerebras-smart": "llama-3.3-70b"
   };
   VersionConflictError = class VersionConflictError extends Error {
     constructor(entity, id) {
@@ -2325,6 +2363,12 @@ var init_types = __esm(() => {
       this.name = "ScheduleNotFoundError";
     }
   };
+  BudgetExceededError = class BudgetExceededError extends Error {
+    constructor(estimatedCents, capCents) {
+      super(`Estimated run cost ($${(estimatedCents / 100).toFixed(2)}) exceeds budget cap ($${(capCents / 100).toFixed(2)}). Pass skipBudgetCheck: true to override.`);
+      this.name = "BudgetExceededError";
+    }
+  };
   DependencyCycleError = class DependencyCycleError extends Error {
     constructor(scenarioId, dependsOn) {
       super(`Adding dependency ${dependsOn} to ${scenarioId} would create a cycle`);
@@ -2734,6 +2778,26 @@ CREATE TABLE IF NOT EXISTS golden_check_results (
 CREATE INDEX IF NOT EXISTS idx_golden_project ON golden_answers(project_id);
 CREATE INDEX IF NOT EXISTS idx_golden_enabled ON golden_answers(enabled);
 CREATE INDEX IF NOT EXISTS idx_golden_results_golden ON golden_check_results(golden_id);
+  `,
+    `
+ALTER TABLE results ADD COLUMN failure_analysis TEXT;
+  `,
+    `
+ALTER TABLE personas ADD COLUMN behaviors TEXT DEFAULT '[]';
+ALTER TABLE personas ADD COLUMN expertise_level TEXT DEFAULT 'intermediate';
+ALTER TABLE personas ADD COLUMN demographics TEXT DEFAULT '{}';
+ALTER TABLE personas ADD COLUMN pain_points TEXT DEFAULT '[]';
+  `,
+    `
+ALTER TABLE scenarios ADD COLUMN last_passed_at TEXT;
+ALTER TABLE scenarios ADD COLUMN last_passed_url TEXT;
+  `,
+    `
+ALTER TABLE personas ADD COLUMN auth_email TEXT;
+ALTER TABLE personas ADD COLUMN auth_password TEXT;
+ALTER TABLE personas ADD COLUMN auth_login_path TEXT DEFAULT '/login';
+ALTER TABLE personas ADD COLUMN auth_cookies TEXT;
+ALTER TABLE scenarios ADD COLUMN required_role TEXT;
   `
   ];
 });
@@ -2741,6 +2805,7 @@ CREATE INDEX IF NOT EXISTS idx_golden_results_golden ON golden_check_results(gol
 // src/db/scenarios.ts
 var exports_scenarios = {};
 __export(exports_scenarios, {
+  updateScenarioPassedCache: () => updateScenarioPassedCache,
   updateScenario: () => updateScenario,
   listScenarios: () => listScenarios,
   getScenarioByShortId: () => getScenarioByShortId,
@@ -2834,7 +2899,32 @@ function listScenarios(filter) {
     params.push(filter.offset);
   }
   const rows = db2.query(sql).all(...params);
-  return rows.map(scenarioFromRow);
+  const scenarios = rows.map(scenarioFromRow);
+  if (scenarios.length === 0)
+    return scenarios;
+  const scenarioIds = scenarios.map((s) => s.id);
+  const placeholders = scenarioIds.map(() => "?").join(",");
+  const statsRows = db2.query(`
+    SELECT scenario_id,
+           COUNT(*) as total,
+           SUM(CASE WHEN status = 'passed' THEN 1 ELSE 0 END) as passed
+    FROM (
+      SELECT scenario_id, status
+      FROM results
+      WHERE scenario_id IN (${placeholders})
+      ORDER BY created_at DESC
+    )
+    GROUP BY scenario_id
+  `).all(...scenarioIds);
+  const statsMap = new Map(statsRows.map((r) => [r.scenario_id, r]));
+  return scenarios.map((s) => {
+    const stats = statsMap.get(s.id);
+    return {
+      ...s,
+      flakinessScore: stats ? stats.passed / stats.total : null,
+      recentRunCount: stats?.total ?? 0
+    };
+  });
 }
 function updateScenario(id, input, version) {
   const db2 = getDatabase();
@@ -2955,6 +3045,10 @@ function findStaleScenarios(days) {
     lastRunAt: row.last_run_at
   }));
 }
+function updateScenarioPassedCache(id, url) {
+  const db2 = getDatabase();
+  db2.query("UPDATE scenarios SET last_passed_at = ?, last_passed_url = ? WHERE id = ?").run(now(), url, id);
+}
 function deleteScenario(id) {
   const db2 = getDatabase();
   const scenario = getScenario(id);
@@ -3195,6 +3289,10 @@ function updateResult(id, updates) {
     sets.push("metadata = ?");
     params.push(JSON.stringify(updates.metadata));
   }
+  if (updates.failureAnalysis !== undefined) {
+    sets.push("failure_analysis = ?");
+    params.push(updates.failureAnalysis !== null ? JSON.stringify(updates.failureAnalysis) : null);
+  }
   if (sets.length === 0) {
     return existing;
   }
@@ -3398,6 +3496,412 @@ var init_browser_lightpanda = __esm(() => {
   init_types();
 });
 
+// src/lib/browser-bun.ts
+var exports_browser_bun = {};
+__export(exports_browser_bun, {
+  isBunWebViewAvailable: () => isBunWebViewAvailable,
+  BunWebViewSession: () => BunWebViewSession
+});
+import { join as join2 } from "path";
+import { mkdirSync as mkdirSync2 } from "fs";
+import { homedir as homedir2 } from "os";
+function isBunWebViewAvailable() {
+  return typeof globalThis.Bun !== "undefined" && typeof globalThis.Bun.WebView !== "undefined";
+}
+function getProfileDir(profileName) {
+  const base = process.env["TESTERS_BROWSER_DATA_DIR"] ?? join2(homedir2(), ".testers", "browser");
+  const dir = join2(base, "profiles", profileName);
+  mkdirSync2(dir, { recursive: true });
+  return dir;
+}
+var BunWebViewSession;
+var init_browser_bun = __esm(() => {
+  BunWebViewSession = class BunWebViewSession {
+    view;
+    _sessionId;
+    _eventListeners = new Map;
+    constructor(opts = {}) {
+      if (!isBunWebViewAvailable()) {
+        throw new Error("Bun.WebView is not available. Install Bun canary: bun upgrade --canary");
+      }
+      const BunWebView = globalThis.Bun.WebView;
+      const constructorOpts = {
+        width: opts.width ?? 1280,
+        height: opts.height ?? 720
+      };
+      if (opts.profile) {
+        constructorOpts.dataStore = { directory: getProfileDir(opts.profile) };
+      } else {
+        constructorOpts.dataStore = "ephemeral";
+      }
+      if (opts.onConsole) {
+        constructorOpts.console = opts.onConsole;
+      }
+      this.view = new BunWebView(constructorOpts);
+      this.view.onNavigated = (url) => {
+        this._emit("navigated", url);
+      };
+      this.view.onNavigationFailed = (error) => {
+        this._emit("navigationfailed", error);
+      };
+    }
+    async goto(url, opts) {
+      await this.view.navigate(url);
+      await new Promise((r) => setTimeout(r, 200));
+    }
+    async goBack() {
+      await this.view.goBack();
+    }
+    async goForward() {
+      await this.view.goForward();
+    }
+    async reload() {
+      await this.view.reload();
+    }
+    async evaluate(fnOrExpr, ...args) {
+      let expr;
+      if (typeof fnOrExpr === "function") {
+        const serializedArgs = args.map((a) => JSON.stringify(a)).join(", ");
+        expr = `(${fnOrExpr.toString()})(${serializedArgs})`;
+      } else {
+        expr = fnOrExpr;
+      }
+      return this.view.evaluate(expr);
+    }
+    async screenshot(opts) {
+      const uint8 = await this.view.screenshot();
+      return Buffer.from(uint8);
+    }
+    async click(selector, opts) {
+      await this.view.click(selector, opts ? { button: opts.button } : undefined);
+    }
+    async type(selector, text, opts) {
+      try {
+        await this.view.click(selector);
+      } catch {}
+      await this.view.type(text);
+    }
+    async fill(selector, value) {
+      await this.view.evaluate(`
+      (() => {
+        const el = document.querySelector(${JSON.stringify(selector)});
+        if (el) { el.value = ''; el.dispatchEvent(new Event('input')); }
+      })()
+    `);
+      await this.type(selector, value);
+    }
+    async press(key, opts) {
+      await this.view.press(key, opts);
+    }
+    async scroll(direction, amount) {
+      const dx = direction === "left" ? -amount : direction === "right" ? amount : 0;
+      const dy = direction === "up" ? -amount : direction === "down" ? amount : 0;
+      await this.view.scroll(dx, dy);
+    }
+    async scrollIntoView(selector) {
+      await this.view.scrollTo(selector);
+    }
+    async hover(selector) {
+      try {
+        await this.view.scrollTo(selector);
+      } catch {}
+    }
+    async resize(width, height) {
+      await this.view.resize(width, height);
+    }
+    async $(selector) {
+      const exists = await this.view.evaluate(`!!document.querySelector(${JSON.stringify(selector)})`);
+      if (!exists)
+        return null;
+      return {
+        textContent: async () => this.view.evaluate(`document.querySelector(${JSON.stringify(selector)})?.textContent ?? null`)
+      };
+    }
+    async $$(selector) {
+      const count = await this.view.evaluate(`document.querySelectorAll(${JSON.stringify(selector)}).length`);
+      return Array.from({ length: count }, (_, i) => ({
+        textContent: async () => this.view.evaluate(`document.querySelectorAll(${JSON.stringify(selector)})[${i}]?.textContent ?? null`)
+      }));
+    }
+    async inputValue(selector) {
+      return this.view.evaluate(`document.querySelector(${JSON.stringify(selector)})?.value ?? ''`);
+    }
+    async isChecked(selector) {
+      return this.view.evaluate(`!!(document.querySelector(${JSON.stringify(selector)})?.checked)`);
+    }
+    async isVisible(selector) {
+      return this.view.evaluate(`
+      (() => {
+        const el = document.querySelector(${JSON.stringify(selector)});
+        if (!el) return false;
+        const style = window.getComputedStyle(el);
+        return style.display !== 'none' && style.visibility !== 'hidden' && el.offsetWidth > 0;
+      })()
+    `);
+    }
+    async isEnabled(selector) {
+      return this.view.evaluate(`!(document.querySelector(${JSON.stringify(selector)})?.disabled)`);
+    }
+    async selectOption(selector, value) {
+      await this.view.evaluate(`
+      (() => {
+        const el = document.querySelector(${JSON.stringify(selector)});
+        if (el) {
+          el.value = ${JSON.stringify(value)};
+          el.dispatchEvent(new Event('change'));
+        }
+      })()
+    `);
+      return [value];
+    }
+    async check(selector) {
+      await this.view.evaluate(`
+      (() => {
+        const el = document.querySelector(${JSON.stringify(selector)});
+        if (el && !el.checked) { el.checked = true; el.dispatchEvent(new Event('change')); }
+      })()
+    `);
+    }
+    async uncheck(selector) {
+      await this.view.evaluate(`
+      (() => {
+        const el = document.querySelector(${JSON.stringify(selector)});
+        if (el && el.checked) { el.checked = false; el.dispatchEvent(new Event('change')); }
+      })()
+    `);
+    }
+    async setInputFiles(selector, files) {
+      throw new Error("File upload not supported in Bun.WebView engine. Use engine: 'playwright' instead.");
+    }
+    getByRole(role, opts) {
+      const name = opts?.name?.toString() ?? "";
+      const selector = name ? `[role="${role}"][aria-label*="${name}"], ${role}[aria-label*="${name}"]` : `[role="${role}"], ${role}`;
+      return {
+        click: (clickOpts) => this.click(selector, clickOpts),
+        fill: (value) => this.fill(selector, value),
+        check: () => this.check(selector),
+        uncheck: () => this.uncheck(selector),
+        isVisible: () => this.isVisible(selector),
+        textContent: () => this.view.evaluate(`document.querySelector(${JSON.stringify(selector)})?.textContent ?? null`),
+        inputValue: () => this.inputValue(selector),
+        first: () => ({
+          click: (clickOpts) => this.click(selector, clickOpts),
+          fill: (value) => this.fill(selector, value),
+          textContent: () => this.view.evaluate(`document.querySelector(${JSON.stringify(selector)})?.textContent ?? null`),
+          isVisible: () => this.isVisible(selector),
+          hover: () => this.hover(selector),
+          boundingBox: async () => null,
+          scrollIntoViewIfNeeded: () => this.scrollIntoView(selector),
+          evaluate: (fn) => this.view.evaluate(`(${fn.toString()})(document.querySelector(${JSON.stringify(selector)}))`),
+          waitFor: (opts2) => {
+            return new Promise((resolve, reject) => {
+              const timeout = opts2?.timeout ?? 1e4;
+              const start = Date.now();
+              const check = async () => {
+                const visible = await this.isVisible(selector);
+                if (visible)
+                  return resolve();
+                if (Date.now() - start > timeout)
+                  return reject(new Error(`Timeout waiting for ${selector}`));
+                setTimeout(check, 100);
+              };
+              check();
+            });
+          }
+        }),
+        count: async () => {
+          const count = await this.view.evaluate(`document.querySelectorAll(${JSON.stringify(selector)}).length`);
+          return count;
+        },
+        nth: (n) => ({
+          click: (clickOpts) => this.click(selector, clickOpts),
+          textContent: () => this.view.evaluate(`document.querySelectorAll(${JSON.stringify(selector)})[${n}]?.textContent ?? null`),
+          isVisible: () => this.isVisible(selector)
+        })
+      };
+    }
+    getByText(text, opts) {
+      const selector = opts?.exact ? `*:is(button, a, span, div, p, h1, h2, h3, h4, label)` : "*";
+      return {
+        first: () => ({
+          click: async (clickOpts) => {
+            await this.view.evaluate(`
+            (() => {
+              const text = ${JSON.stringify(text)};
+              const all = document.querySelectorAll('*');
+              for (const el of all) {
+                if (el.children.length === 0 && el.textContent?.trim() === text) {
+                  el.click(); return;
+                }
+              }
+              for (const el of all) {
+                if (el.textContent?.includes(text)) { el.click(); return; }
+              }
+            })()
+          `);
+          },
+          waitFor: (waitOpts) => {
+            const timeout = waitOpts?.timeout ?? 1e4;
+            return new Promise((resolve, reject) => {
+              const start = Date.now();
+              const check = async () => {
+                const found = await this.view.evaluate(`document.body?.textContent?.includes(${JSON.stringify(text)})`);
+                if (found)
+                  return resolve();
+                if (Date.now() - start > timeout)
+                  return reject(new Error(`Timeout: text "${text}" not found`));
+                setTimeout(check, 100);
+              };
+              check();
+            });
+          }
+        })
+      };
+    }
+    locator(selector) {
+      return {
+        click: (opts) => this.click(selector, opts),
+        fill: (value) => this.fill(selector, value),
+        scrollIntoViewIfNeeded: () => this.scrollIntoView(selector),
+        first: () => this.getByRole("*").first(),
+        evaluate: (fn) => this.view.evaluate(`(${fn.toString()})(document.querySelector(${JSON.stringify(selector)}))`),
+        waitFor: (opts) => {
+          const timeout = opts?.timeout ?? 1e4;
+          return new Promise((resolve, reject) => {
+            const start = Date.now();
+            const check = async () => {
+              const exists = await this.view.evaluate(`!!document.querySelector(${JSON.stringify(selector)})`);
+              if (exists)
+                return resolve();
+              if (Date.now() - start > timeout)
+                return reject(new Error(`Timeout: ${selector}`));
+              setTimeout(check, 100);
+            };
+            check();
+          });
+        }
+      };
+    }
+    url() {
+      return this.view.url;
+    }
+    async title() {
+      return this.view.title || await this.evaluate("document.title");
+    }
+    viewportSize() {
+      return { width: 1280, height: 720 };
+    }
+    async waitForLoadState(state, opts) {
+      await new Promise((r) => setTimeout(r, 200));
+    }
+    async waitForURL(pattern, opts) {
+      const timeout = opts?.timeout ?? 30000;
+      const start = Date.now();
+      while (Date.now() - start < timeout) {
+        const url = this.view.url;
+        const matches = pattern instanceof RegExp ? pattern.test(url) : url.includes(pattern);
+        if (matches)
+          return;
+        await new Promise((r) => setTimeout(r, 100));
+      }
+      throw new Error(`Timeout waiting for URL to match ${pattern}`);
+    }
+    async waitForSelector(selector, opts) {
+      const timeout = opts?.timeout ?? 1e4;
+      const start = Date.now();
+      while (Date.now() - start < timeout) {
+        const exists = await this.view.evaluate(`!!document.querySelector(${JSON.stringify(selector)})`);
+        if (exists)
+          return;
+        await new Promise((r) => setTimeout(r, 100));
+      }
+      throw new Error(`Timeout waiting for ${selector}`);
+    }
+    async setContent(html) {
+      await this.view.navigate(`data:text/html,${encodeURIComponent(html)}`);
+      await new Promise((r) => setTimeout(r, 100));
+    }
+    async content() {
+      return this.view.evaluate("document.documentElement.outerHTML");
+    }
+    async addInitScript(script) {
+      const expr = typeof script === "function" ? `(${script.toString()})()` : script;
+      await this.view.evaluate(expr);
+    }
+    keyboard = {
+      press: (key) => this.view.press(key)
+    };
+    context() {
+      return {
+        close: async () => {
+          await this.close();
+        },
+        newPage: async () => {
+          throw new Error("Multi-tab not supported in Bun.WebView. Use engine: 'playwright'");
+        },
+        cookies: async () => [],
+        addCookies: async (_) => {},
+        clearCookies: async () => {},
+        newCDPSession: async () => {
+          throw new Error("CDP session via context not available in Bun.WebView. Use view.cdp() when shipped.");
+        },
+        route: async (_pattern, _handler) => {
+          throw new Error("Network interception not supported in Bun.WebView. Use engine: 'cdp' or 'playwright'.");
+        },
+        unrouteAll: async () => {},
+        pages: () => [],
+        addInitScript: async (script) => {
+          await this.addInitScript(script);
+        }
+      };
+    }
+    on(event, handler) {
+      if (!this._eventListeners.has(event))
+        this._eventListeners.set(event, []);
+      this._eventListeners.get(event).push(handler);
+      return this;
+    }
+    off(event, handler) {
+      const listeners = this._eventListeners.get(event) ?? [];
+      this._eventListeners.set(event, listeners.filter((l) => l !== handler));
+      return this;
+    }
+    _emit(event, ...args) {
+      for (const handler of this._eventListeners.get(event) ?? []) {
+        try {
+          handler(...args);
+        } catch {}
+      }
+    }
+    async pdf(_opts) {
+      throw new Error("PDF generation not supported in Bun.WebView. Use engine: 'playwright'.");
+    }
+    coverage = {
+      startJSCoverage: async () => {},
+      stopJSCoverage: async () => [],
+      startCSSCoverage: async () => {},
+      stopCSSCoverage: async () => []
+    };
+    setSessionId(id) {
+      this._sessionId = id;
+    }
+    getSessionId() {
+      return this._sessionId;
+    }
+    getNativeView() {
+      return this.view;
+    }
+    async close() {
+      try {
+        await this.view.close();
+      } catch {}
+    }
+    [Symbol.asyncDispose]() {
+      return this.close();
+    }
+  };
+});
+
 // src/lib/browser.ts
 var exports_browser = {};
 __export(exports_browser, {
@@ -3419,6 +3923,22 @@ async function launchBrowser(options) {
     }
     return launchLightpanda2({ viewport: options?.viewport });
   }
+  if (engine === "bun") {
+    const { isBunWebViewAvailable: isBunWebViewAvailable2, BunWebViewSession: BunWebViewSession2 } = await Promise.resolve().then(() => (init_browser_bun(), exports_browser_bun));
+    if (!isBunWebViewAvailable2()) {
+      throw new BrowserError("Bun.WebView not available. Upgrade to Bun canary: bun upgrade --canary");
+    }
+    const session = new BunWebViewSession2({
+      width: options?.viewport?.width ?? 1280,
+      height: options?.viewport?.height ?? 720
+    });
+    return {
+      newContext: async () => ({ newPage: async () => session, close: async () => {} }),
+      close: async () => session.close(),
+      contexts: () => [],
+      _bunSession: session
+    };
+  }
   const headless = options?.headless ?? true;
   const viewport = options?.viewport ?? DEFAULT_VIEWPORT;
   try {
@@ -3440,6 +3960,12 @@ async function getPage(browser, options) {
     const { getLightpandaPage: getLightpandaPage2 } = await Promise.resolve().then(() => (init_browser_lightpanda(), exports_browser_lightpanda));
     return getLightpandaPage2(browser, options);
   }
+  if (engine === "bun") {
+    const bunSession = browser._bunSession;
+    if (bunSession)
+      return bunSession;
+    throw new BrowserError("Bun.WebView session not found on browser instance");
+  }
   const viewport = options?.viewport ?? DEFAULT_VIEWPORT;
   try {
     const context = await browser.newContext({
@@ -3459,6 +3985,12 @@ async function closeBrowser(browser, engine) {
     const { closeLightpanda: closeLightpanda2 } = await Promise.resolve().then(() => (init_browser_lightpanda(), exports_browser_lightpanda));
     return closeLightpanda2(browser);
   }
+  if (engine === "bun") {
+    const bunSession = browser._bunSession;
+    if (bunSession)
+      await bunSession.close();
+    return;
+  }
   try {
     await browser.close();
   } catch (error) {
@@ -3528,6 +4060,9 @@ async function launchBrowserEngine(engine, config) {
     }
     return launchLightpanda2({ viewport: config.viewport });
   }
+  if (engine === "bun") {
+    return launchBrowser({ headless: config.headless, viewport: config.viewport, engine: "bun" });
+  }
   return chromium2.launch({
     headless: config.headless,
     args: ["--no-sandbox", "--disable-setuid-sandbox"]
@@ -3662,8 +4197,8 @@ async function scanA11y(options) {
 var AXE_CDN = "https://cdn.jsdelivr.net/npm/axe-core@4/axe.min.js";
 
 // src/lib/config.ts
-import { homedir as homedir2 } from "os";
-import { join as join2 } from "path";
+import { homedir as homedir3 } from "os";
+import { join as join3 } from "path";
 import { readFileSync, existsSync as existsSync2 } from "fs";
 function getDefaultConfig() {
   return {
@@ -3675,7 +4210,7 @@ function getDefaultConfig() {
       timeout: 60000
     },
     screenshots: {
-      dir: join2(homedir2(), ".testers", "screenshots"),
+      dir: join3(homedir3(), ".testers", "screenshots"),
       format: "png",
       quality: 90,
       fullPage: false
@@ -3701,7 +4236,8 @@ function loadConfig() {
     todosDbPath: fileConfig.todosDbPath,
     judgeModel: fileConfig.judgeModel,
     judgeProvider: fileConfig.judgeProvider,
-    selfHeal: fileConfig.selfHeal ?? false
+    selfHeal: fileConfig.selfHeal ?? false,
+    conversationsSpace: fileConfig.conversationsSpace
   };
   const envModel = process.env["TESTERS_MODEL"];
   if (envModel) {
@@ -3720,8 +4256,8 @@ function loadConfig() {
 var CONFIG_DIR, CONFIG_PATH;
 var init_config = __esm(() => {
   init_types();
-  CONFIG_DIR = join2(homedir2(), ".testers");
-  CONFIG_PATH = join2(CONFIG_DIR, "config.json");
+  CONFIG_DIR = join3(homedir3(), ".testers");
+  CONFIG_PATH = join3(CONFIG_DIR, "config.json");
 });
 
 // src/lib/healer.ts
@@ -4163,6 +4699,8 @@ async function runAgentLoop(options) {
 Instructions: ${persona.instructions}` : "",
     persona.traits.length > 0 ? `Traits: ${persona.traits.join(", ")}` : "",
     persona.goals.length > 0 ? `Goals: ${persona.goals.join("; ")}` : "",
+    persona.behaviors && persona.behaviors.length > 0 ? `Behaviors: ${persona.behaviors.join("; ")}` : "",
+    persona.painPoints && persona.painPoints.length > 0 ? `Pain points: ${persona.painPoints.join("; ")}` : "",
     "",
     "Stay in character throughout the test. Your observations, choices, and priorities should reflect this persona."
   ].filter(Boolean).join(`
@@ -4213,6 +4751,15 @@ Instructions: ${persona.instructions}` : "",
   const isOpenAICompat = "provider" in client;
   try {
     for (let turn = 0;turn < maxTurns; turn++) {
+      if (persona && turn > 0 && turn % 5 === 0) {
+        messages = [
+          ...messages,
+          {
+            role: "user",
+            content: `[Reminder: You are ${persona.name} \u2014 ${persona.role}. Traits: ${persona.traits.join(", ")}. Stay in character.]`
+          }
+        ];
+      }
       const response = isOpenAICompat ? await callOpenAICompatible({
         baseUrl: client.baseUrl,
         apiKey: client.apiKey,
@@ -4307,6 +4854,8 @@ function detectProvider(model) {
     return "openai";
   if (model.startsWith("gemini-"))
     return "google";
+  if (model.startsWith("llama-") || model.startsWith("qwen-") || model.includes("cerebras"))
+    return "cerebras";
   return "anthropic";
 }
 function createClient(apiKey) {
@@ -4400,6 +4949,12 @@ function createClientForModel(model, apiKey) {
       throw new AIClientError("No Google API key. Set GOOGLE_API_KEY or pass it explicitly.");
     return { provider: "google", baseUrl: "https://generativelanguage.googleapis.com/v1beta/openai", apiKey: key };
   }
+  if (provider === "cerebras") {
+    const key = apiKey ?? process.env["CEREBRAS_API_KEY"];
+    if (!key)
+      throw new AIClientError("No Cerebras API key. Set CEREBRAS_API_KEY or pass it explicitly.");
+    return { provider: "cerebras", baseUrl: "https://api.cerebras.ai/v1", apiKey: key };
+  }
   return createClient(apiKey);
 }
 var BROWSER_TOOLS;
@@ -4760,19 +5315,21 @@ function resolveJudgeModel(config) {
       apiKey = process.env["OPENAI_API_KEY"];
     else if (provider === "google")
       apiKey = process.env["GOOGLE_API_KEY"];
+    else if (provider === "cerebras")
+      apiKey = process.env["CEREBRAS_API_KEY"];
   }
   if (!apiKey) {
-    apiKey = process.env["ANTHROPIC_API_KEY"] ?? process.env["OPENAI_API_KEY"] ?? process.env["GOOGLE_API_KEY"] ?? globalConfig.anthropicApiKey;
+    apiKey = process.env["ANTHROPIC_API_KEY"] ?? process.env["CEREBRAS_API_KEY"] ?? process.env["OPENAI_API_KEY"] ?? process.env["GOOGLE_API_KEY"] ?? globalConfig.anthropicApiKey;
     if (!apiKey)
-      throw new AIClientError("No API key found for judge. Set ANTHROPIC_API_KEY, OPENAI_API_KEY, or GOOGLE_API_KEY.");
+      throw new AIClientError("No API key found for judge. Set ANTHROPIC_API_KEY, CEREBRAS_API_KEY, OPENAI_API_KEY, or GOOGLE_API_KEY.");
   }
   return { model, provider, apiKey };
 }
 async function callJudge(prompt, config) {
   const { model, provider, apiKey } = resolveJudgeModel(config);
   const threshold = 0.7;
-  if (provider === "openai" || provider === "google") {
-    const baseUrl = provider === "openai" ? "https://api.openai.com/v1" : "https://generativelanguage.googleapis.com/v1beta/openai";
+  if (provider === "openai" || provider === "google" || provider === "cerebras") {
+    const baseUrl = provider === "openai" ? "https://api.openai.com/v1" : provider === "cerebras" ? "https://api.cerebras.ai/v1" : "https://generativelanguage.googleapis.com/v1beta/openai";
     const resp2 = await callOpenAICompatible({
       baseUrl,
       apiKey,
@@ -5322,84 +5879,437 @@ var init_eval_runner = __esm(() => {
   init_pipeline_runner();
 });
 
-// src/db/personas.ts
-function createPersona(input) {
-  const db2 = getDatabase();
-  const id = uuid();
-  const short_id = shortUuid();
-  const timestamp = now();
-  db2.query(`
-    INSERT INTO personas (id, short_id, project_id, name, description, role, instructions, traits, goals, metadata, enabled, version, created_at, updated_at)
-    VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, 1, ?, ?)
-  `).run(id, short_id, input.projectId ?? null, input.name, input.description ?? "", input.role, input.instructions ?? "", JSON.stringify(input.traits ?? []), JSON.stringify(input.goals ?? []), input.metadata ? JSON.stringify(input.metadata) : "{}", input.enabled === false ? 0 : 1, timestamp, timestamp);
-  return getPersona(id);
-}
-function getPersona(id) {
-  const db2 = getDatabase();
-  let row = db2.query("SELECT * FROM personas WHERE id = ?").get(id);
-  if (row)
-    return personaFromRow(row);
-  row = db2.query("SELECT * FROM personas WHERE short_id = ?").get(id);
-  if (row)
-    return personaFromRow(row);
-  return null;
-}
-function listPersonas(filter) {
-  const db2 = getDatabase();
-  const conditions = [];
-  const params = [];
-  if (filter?.globalOnly) {
-    conditions.push("project_id IS NULL");
-  } else if (filter?.projectId) {
-    conditions.push("(project_id = ? OR project_id IS NULL)");
-    params.push(filter.projectId);
+// src/lib/failure-analyzer.ts
+function analyzeFailure(error, reasoning) {
+  const combinedText = [error, reasoning].filter(Boolean).join(" ");
+  if (!combinedText.trim())
+    return null;
+  const errorText = error ?? "";
+  const reasoningText = reasoning ?? "";
+  if (/waiting for selector/i.test(errorText) || /not found/i.test(errorText) || /No element/i.test(errorText) || /waiting for selector/i.test(reasoningText) || /could not find element/i.test(reasoningText) || /element not found/i.test(reasoningText)) {
+    const selectorMatch = errorText.match(/'([^']+)'/) ?? reasoningText.match(/'([^']+)'/);
+    const affectedElement = selectorMatch ? selectorMatch[1] : undefined;
+    const stepMatch = reasoningText.match(/step\s+(\d+)/i);
+    const stepNumber = stepMatch ? parseInt(stepMatch[1], 10) : undefined;
+    return {
+      type: "selector_not_found",
+      affectedElement,
+      stepNumber,
+      confidence: affectedElement ? "high" : "medium"
+    };
   }
-  if (filter?.enabled !== undefined) {
-    conditions.push("enabled = ?");
-    params.push(filter.enabled ? 1 : 0);
+  if (/assert/i.test(errorText) || /expected/i.test(errorText) || /to equal/i.test(errorText) || /to be/i.test(errorText) || /\bgot\b/.test(errorText) || /assertion.*failed/i.test(reasoningText) || /expected.*but.*got/i.test(reasoningText)) {
+    const expectedActualMatch = errorText.match(/expected[:\s]+(['"]?)([^'"]+)\1[,\s]+(?:got|received|actual)[:\s]+(['"]?)([^'"]+)\3/i);
+    const toEqualMatch = errorText.match(/expected[:\s]+(['"]?)([^'"]+)\1\s+to\s+equal\s+(['"]?)([^'"]+)\3/i);
+    let expected;
+    let actual;
+    if (expectedActualMatch) {
+      expected = expectedActualMatch[2];
+      actual = expectedActualMatch[4];
+    } else if (toEqualMatch) {
+      expected = toEqualMatch[4];
+      actual = toEqualMatch[2];
+    }
+    const stepMatch = reasoningText.match(/step\s+(\d+)/i);
+    const stepNumber = stepMatch ? parseInt(stepMatch[1], 10) : undefined;
+    return {
+      type: "assertion_failed",
+      expected,
+      actual,
+      stepNumber,
+      confidence: expected && actual ? "high" : "medium"
+    };
   }
-  let sql = "SELECT * FROM personas";
-  if (conditions.length > 0) {
-    sql += " WHERE " + conditions.join(" AND ");
+  if (/timeout/i.test(errorText) || /timed out/i.test(errorText) || /Timeout/i.test(reasoningText) || /timed out/i.test(reasoningText)) {
+    const stepMatch = reasoningText.match(/step\s+(\d+)/i);
+    const stepNumber = stepMatch ? parseInt(stepMatch[1], 10) : undefined;
+    return {
+      type: "timeout",
+      stepNumber,
+      confidence: "high"
+    };
   }
-  sql += " ORDER BY created_at DESC";
-  if (filter?.limit) {
-    sql += " LIMIT ?";
-    params.push(filter.limit);
+  if (/\b401\b/.test(errorText) || /\b403\b/.test(errorText) || /login/i.test(errorText) || /unauthorized/i.test(errorText) || /\bauth\b/i.test(errorText) || /\b401\b/.test(reasoningText) || /\b403\b/.test(reasoningText) || /unauthorized/i.test(reasoningText) || /authentication/i.test(reasoningText)) {
+    return {
+      type: "auth_error",
+      confidence: "high"
+    };
   }
-  if (filter?.offset) {
-    sql += " OFFSET ?";
-    params.push(filter.offset);
+  if (/ECONNREFUSED/i.test(errorText) || /ENOTFOUND/i.test(errorText) || /fetch failed/i.test(errorText) || /network/i.test(errorText) || /ECONNREFUSED/i.test(reasoningText) || /fetch failed/i.test(reasoningText) || /connection refused/i.test(reasoningText)) {
+    return {
+      type: "network_error",
+      confidence: "high"
+    };
   }
-  const rows = db2.query(sql).all(...params);
-  return rows.map(personaFromRow);
-}
-function deletePersona(id) {
-  const db2 = getDatabase();
-  const persona = getPersona(id);
-  if (!persona)
-    return false;
-  const result = db2.query("DELETE FROM personas WHERE id = ?").run(persona.id);
-  return result.changes > 0;
+  if (/\beval\b/i.test(errorText) || /evaluate/i.test(errorText) || /\bscript\b/i.test(errorText) || /\beval\b/i.test(reasoningText) || /evaluate/i.test(reasoningText)) {
+    return {
+      type: "eval_failed",
+      confidence: "medium"
+    };
+  }
+  return {
+    type: "unknown",
+    confidence: "low"
+  };
 }
-var init_personas = __esm(() => {
-  init_types();
-  init_database();
-});
 
-// src/lib/screenshotter.ts
-import { mkdirSync as mkdirSync2, existsSync as existsSync3, writeFileSync } from "fs";
-import { join as join3 } from "path";
-import { homedir as homedir3 } from "os";
-function slugify(text) {
-  return text.toLowerCase().replace(/[^a-z0-9]+/g, "-").replace(/^-+|-+$/g, "");
+// src/lib/costs.ts
+import chalk from "chalk";
+function getDateFilter(period) {
+  switch (period) {
+    case "day":
+      return "AND r.created_at >= date('now', 'start of day')";
+    case "week":
+      return "AND r.created_at >= date('now', '-7 days')";
+    case "month":
+      return "AND r.created_at >= date('now', '-30 days')";
+    case "all":
+      return "";
+  }
 }
-function generateFilename(stepNumber, action) {
-  const padded = String(stepNumber).padStart(3, "0");
-  const slug = slugify(action);
-  return `${padded}_${slug}.png`;
+function getPeriodDays(period) {
+  switch (period) {
+    case "day":
+      return 1;
+    case "week":
+      return 7;
+    case "month":
+      return 30;
+    case "all":
+      return 30;
+  }
 }
-function formatDate(date) {
+function loadBudgetConfig() {
+  const config = loadConfig();
+  const budget = config.budget;
+  return {
+    maxPerRunCents: budget?.maxPerRunCents ?? 50,
+    maxPerDayCents: budget?.maxPerDayCents ?? 500,
+    warnAtPercent: budget?.warnAtPercent ?? 0.8
+  };
+}
+function getCostSummary(options) {
+  const db2 = getDatabase();
+  const period = options?.period ?? "month";
+  const projectId = options?.projectId;
+  const dateFilter = getDateFilter(period);
+  const projectFilter = projectId ? "AND ru.project_id = ?" : "";
+  const projectParams = projectId ? [projectId] : [];
+  const totalsRow = db2.query(`SELECT
+        COALESCE(SUM(r.cost_cents), 0) as total_cost,
+        COALESCE(SUM(r.tokens_used), 0) as total_tokens,
+        COUNT(DISTINCT r.run_id) as run_count
+      FROM results r
+      JOIN runs ru ON r.run_id = ru.id
+      WHERE 1=1 ${dateFilter} ${projectFilter}`).get(...projectParams);
+  const modelRows = db2.query(`SELECT
+        r.model,
+        COALESCE(SUM(r.cost_cents), 0) as cost_cents,
+        COALESCE(SUM(r.tokens_used), 0) as tokens,
+        COUNT(DISTINCT r.run_id) as runs
+      FROM results r
+      JOIN runs ru ON r.run_id = ru.id
+      WHERE 1=1 ${dateFilter} ${projectFilter}
+      GROUP BY r.model
+      ORDER BY cost_cents DESC`).all(...projectParams);
+  const byModel = {};
+  for (const row of modelRows) {
+    byModel[row.model] = {
+      costCents: row.cost_cents,
+      tokens: row.tokens,
+      runs: row.runs
+    };
+  }
+  const scenarioRows = db2.query(`SELECT
+        r.scenario_id,
+        COALESCE(s.name, r.scenario_id) as name,
+        COALESCE(SUM(r.cost_cents), 0) as cost_cents,
+        COALESCE(SUM(r.tokens_used), 0) as tokens,
+        COUNT(DISTINCT r.run_id) as runs
+      FROM results r
+      JOIN runs ru ON r.run_id = ru.id
+      LEFT JOIN scenarios s ON r.scenario_id = s.id
+      WHERE 1=1 ${dateFilter} ${projectFilter}
+      GROUP BY r.scenario_id
+      ORDER BY cost_cents DESC
+      LIMIT 10`).all(...projectParams);
+  const byScenario = scenarioRows.map((row) => ({
+    scenarioId: row.scenario_id,
+    name: row.name,
+    costCents: row.cost_cents,
+    tokens: row.tokens,
+    runs: row.runs
+  }));
+  const runCount = totalsRow.run_count;
+  const avgCostPerRun = runCount > 0 ? totalsRow.total_cost / runCount : 0;
+  const periodDays = getPeriodDays(period);
+  const estimatedMonthlyCents = periodDays > 0 ? totalsRow.total_cost / periodDays * 30 : 0;
+  return {
+    period,
+    totalCostCents: totalsRow.total_cost,
+    totalTokens: totalsRow.total_tokens,
+    runCount,
+    byModel,
+    byScenario,
+    avgCostPerRun,
+    estimatedMonthlyCents
+  };
+}
+function modelToCostKey(model) {
+  const exact = COST_PER_SCENARIO_CENTS[model];
+  if (exact !== undefined)
+    return exact;
+  const lower = model.toLowerCase();
+  if (lower.includes("opus"))
+    return COST_PER_SCENARIO_CENTS["opus"];
+  if (lower.includes("sonnet"))
+    return COST_PER_SCENARIO_CENTS["sonnet"];
+  if (lower.includes("haiku"))
+    return COST_PER_SCENARIO_CENTS["haiku"];
+  if (lower.includes("gpt-4o-mini"))
+    return COST_PER_SCENARIO_CENTS["gpt-4o-mini"];
+  if (lower.includes("gpt-4o"))
+    return COST_PER_SCENARIO_CENTS["gpt-4o"];
+  if (lower.includes("gemini-2.0-flash") || lower.includes("gemini-flash"))
+    return COST_PER_SCENARIO_CENTS["gemini-2.0-flash"];
+  if (lower.includes("gemini-1.5-pro") || lower.includes("gemini-pro"))
+    return COST_PER_SCENARIO_CENTS["gemini-1.5-pro"];
+  if (lower.includes("llama-3.3") || lower.includes("llama3.3"))
+    return COST_PER_SCENARIO_CENTS["llama-3.3-70b"];
+  if (lower.includes("llama"))
+    return COST_PER_SCENARIO_CENTS["llama-3.1-8b"];
+  return 10;
+}
+function estimateRunCostCents(scenarioCount, model, samples = 1) {
+  const costPerScenario = modelToCostKey(model);
+  return scenarioCount * costPerScenario * Math.max(1, samples);
+}
+function getCostsByScenario(options) {
+  const db2 = getDatabase();
+  const period = options?.period ?? "month";
+  const projectId = options?.projectId;
+  const dateFilter = getDateFilter(period);
+  const projectFilter = projectId ? "AND ru.project_id = ?" : "";
+  const projectParams = projectId ? [projectId] : [];
+  const rows = db2.query(`SELECT
+        r.scenario_id,
+        COALESCE(s.name, r.scenario_id) as name,
+        COUNT(DISTINCT r.run_id) as run_count,
+        COALESCE(SUM(r.cost_cents), 0) as total_cost_cents
+      FROM results r
+      JOIN runs ru ON r.run_id = ru.id
+      LEFT JOIN scenarios s ON r.scenario_id = s.id
+      WHERE 1=1 ${dateFilter} ${projectFilter}
+      GROUP BY r.scenario_id
+      ORDER BY total_cost_cents DESC`).all(...projectParams);
+  return rows.map((row) => ({
+    scenarioId: row.scenario_id,
+    name: row.name,
+    runCount: row.run_count,
+    totalCostCents: row.total_cost_cents,
+    avgCostPerRunCents: row.run_count > 0 ? row.total_cost_cents / row.run_count : 0
+  }));
+}
+function formatCostsByScenarioTerminal(rows, period) {
+  const lines = [];
+  lines.push("");
+  lines.push(chalk.bold(`  Cost by Scenario (${period})`));
+  lines.push("");
+  if (rows.length === 0) {
+    lines.push(chalk.dim("  No cost data found."));
+    lines.push("");
+    return lines.join(`
+`);
+  }
+  lines.push(`  ${"Scenario".padEnd(40)} ${"Runs".padEnd(8)} ${"Total Cost".padEnd(14)} Avg/Run`);
+  lines.push(`  ${"\u2500".repeat(40)} ${"\u2500".repeat(8)} ${"\u2500".repeat(14)} ${"\u2500".repeat(10)}`);
+  for (const row of rows) {
+    const label = row.name.length > 38 ? row.name.slice(0, 35) + "..." : row.name;
+    lines.push(`  ${label.padEnd(40)} ${String(row.runCount).padEnd(8)} ${formatDollars(row.totalCostCents).padEnd(14)} ${formatDollars(row.avgCostPerRunCents)}`);
+  }
+  lines.push("");
+  return lines.join(`
+`);
+}
+function checkBudget(estimatedCostCents) {
+  const budget = loadBudgetConfig();
+  if (estimatedCostCents > budget.maxPerRunCents) {
+    return {
+      allowed: false,
+      warning: `Estimated cost (${formatDollars(estimatedCostCents)}) exceeds per-run limit (${formatDollars(budget.maxPerRunCents)})`
+    };
+  }
+  const todaySummary = getCostSummary({ period: "day" });
+  const projectedDaily = todaySummary.totalCostCents + estimatedCostCents;
+  if (projectedDaily > budget.maxPerDayCents) {
+    return {
+      allowed: false,
+      warning: `Daily spending (${formatDollars(todaySummary.totalCostCents)}) + this run (${formatDollars(estimatedCostCents)}) would exceed daily limit (${formatDollars(budget.maxPerDayCents)})`
+    };
+  }
+  if (projectedDaily > budget.maxPerDayCents * budget.warnAtPercent) {
+    return {
+      allowed: true,
+      warning: `Approaching daily limit: ${formatDollars(projectedDaily)} of ${formatDollars(budget.maxPerDayCents)} (${Math.round(projectedDaily / budget.maxPerDayCents * 100)}%)`
+    };
+  }
+  return { allowed: true };
+}
+function formatDollars(cents) {
+  return `$${(cents / 100).toFixed(2)}`;
+}
+function formatTokens(tokens) {
+  if (tokens >= 1e6)
+    return `${(tokens / 1e6).toFixed(1)}M`;
+  if (tokens >= 1000)
+    return `${(tokens / 1000).toFixed(1)}K`;
+  return String(tokens);
+}
+function formatCostsTerminal(summary) {
+  const lines = [];
+  lines.push("");
+  lines.push(chalk.bold(`  Cost Summary (${summary.period})`));
+  lines.push("");
+  lines.push(`  Total:     ${chalk.yellow(formatDollars(summary.totalCostCents))} (${formatTokens(summary.totalTokens)} tokens across ${summary.runCount} runs)`);
+  lines.push(`  Avg/run:   ${chalk.yellow(formatDollars(summary.avgCostPerRun))}`);
+  lines.push(`  Est/month: ${chalk.yellow(formatDollars(summary.estimatedMonthlyCents))}`);
+  const modelEntries = Object.entries(summary.byModel);
+  if (modelEntries.length > 0) {
+    lines.push("");
+    lines.push(chalk.bold("  By Model"));
+    lines.push(`  ${"Model".padEnd(40)} ${"Cost".padEnd(12)} ${"Tokens".padEnd(12)} Runs`);
+    lines.push(`  ${"\u2500".repeat(40)} ${"\u2500".repeat(12)} ${"\u2500".repeat(12)} ${"\u2500".repeat(6)}`);
+    for (const [model, data] of modelEntries) {
+      lines.push(`  ${model.padEnd(40)} ${formatDollars(data.costCents).padEnd(12)} ${formatTokens(data.tokens).padEnd(12)} ${data.runs}`);
+    }
+  }
+  if (summary.byScenario.length > 0) {
+    lines.push("");
+    lines.push(chalk.bold("  Scenarios by Cost (most expensive first)"));
+    lines.push(`  ${"Scenario".padEnd(40)} ${"Total Cost".padEnd(12)} ${"Avg/Run".padEnd(12)} ${"Runs".padEnd(6)} Tokens`);
+    lines.push(`  ${"\u2500".repeat(40)} ${"\u2500".repeat(12)} ${"\u2500".repeat(12)} ${"\u2500".repeat(6)} ${"\u2500".repeat(10)}`);
+    for (const s of summary.byScenario) {
+      const label = s.name.length > 38 ? s.name.slice(0, 35) + "..." : s.name;
+      const avgPerRun = s.runs > 0 ? s.costCents / s.runs : 0;
+      lines.push(`  ${label.padEnd(40)} ${formatDollars(s.costCents).padEnd(12)} ${formatDollars(avgPerRun).padEnd(12)} ${String(s.runs).padEnd(6)} ${formatTokens(s.tokens)}`);
+    }
+  }
+  lines.push("");
+  return lines.join(`
+`);
+}
+function formatCostsJSON(summary) {
+  return JSON.stringify(summary, null, 2);
+}
+function formatCostsCsv(summary) {
+  const lines = [];
+  lines.push("scenario,runs,total_cost_cents,avg_cost_cents,tokens");
+  for (const s of summary.byScenario) {
+    const avgCostCents = s.runs > 0 ? s.costCents / s.runs : 0;
+    const name = s.name.includes(",") ? `"${s.name.replace(/"/g, '""')}"` : s.name;
+    lines.push(`${name},${s.runs},${s.costCents},${avgCostCents.toFixed(2)},${s.tokens}`);
+  }
+  return lines.join(`
+`);
+}
+var COST_PER_SCENARIO_CENTS;
+var init_costs = __esm(() => {
+  init_database();
+  init_config();
+  COST_PER_SCENARIO_CENTS = {
+    haiku: 5,
+    sonnet: 30,
+    opus: 150,
+    "claude-haiku": 5,
+    "claude-sonnet": 30,
+    "claude-opus": 150,
+    "gpt-4o-mini": 3,
+    "gpt-4o": 25,
+    "gemini-2.0-flash": 2,
+    "gemini-1.5-pro": 20,
+    "llama-3.1-8b": 1,
+    "llama-3.3-70b": 3
+  };
+});
+
+// src/db/personas.ts
+function createPersona(input) {
+  const db2 = getDatabase();
+  const id = uuid();
+  const short_id = shortUuid();
+  const timestamp = now();
+  db2.query(`
+    INSERT INTO personas (id, short_id, project_id, name, description, role, instructions, traits, goals, behaviors, expertise_level, demographics, pain_points, metadata, enabled, auth_email, auth_password, auth_login_path, version, created_at, updated_at)
+    VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, 1, ?, ?)
+  `).run(id, short_id, input.projectId ?? null, input.name, input.description ?? "", input.role, input.instructions ?? "", JSON.stringify(input.traits ?? []), JSON.stringify(input.goals ?? []), JSON.stringify(input.behaviors ?? []), input.expertiseLevel ?? "intermediate", JSON.stringify(input.demographics ?? {}), JSON.stringify(input.painPoints ?? []), input.metadata ? JSON.stringify(input.metadata) : "{}", input.enabled === false ? 0 : 1, input.authEmail ?? null, input.authPassword ?? null, input.authLoginPath ?? null, timestamp, timestamp);
+  return getPersona(id);
+}
+function getPersona(id) {
+  const db2 = getDatabase();
+  let row = db2.query("SELECT * FROM personas WHERE id = ?").get(id);
+  if (row)
+    return personaFromRow(row);
+  row = db2.query("SELECT * FROM personas WHERE short_id = ?").get(id);
+  if (row)
+    return personaFromRow(row);
+  return null;
+}
+function listPersonas(filter) {
+  const db2 = getDatabase();
+  const conditions = [];
+  const params = [];
+  if (filter?.globalOnly) {
+    conditions.push("project_id IS NULL");
+  } else if (filter?.projectId) {
+    conditions.push("(project_id = ? OR project_id IS NULL)");
+    params.push(filter.projectId);
+  }
+  if (filter?.enabled !== undefined) {
+    conditions.push("enabled = ?");
+    params.push(filter.enabled ? 1 : 0);
+  }
+  let sql = "SELECT * FROM personas";
+  if (conditions.length > 0) {
+    sql += " WHERE " + conditions.join(" AND ");
+  }
+  sql += " ORDER BY created_at DESC";
+  if (filter?.limit) {
+    sql += " LIMIT ?";
+    params.push(filter.limit);
+  }
+  if (filter?.offset) {
+    sql += " OFFSET ?";
+    params.push(filter.offset);
+  }
+  const rows = db2.query(sql).all(...params);
+  return rows.map(personaFromRow);
+}
+function deletePersona(id) {
+  const db2 = getDatabase();
+  const persona = getPersona(id);
+  if (!persona)
+    return false;
+  const result = db2.query("DELETE FROM personas WHERE id = ?").run(persona.id);
+  return result.changes > 0;
+}
+var init_personas = __esm(() => {
+  init_types();
+  init_database();
+});
+
+// src/lib/screenshotter.ts
+import { mkdirSync as mkdirSync3, existsSync as existsSync3, writeFileSync } from "fs";
+import { join as join4 } from "path";
+import { homedir as homedir4 } from "os";
+function slugify(text) {
+  return text.toLowerCase().replace(/[^a-z0-9]+/g, "-").replace(/^-+|-+$/g, "");
+}
+function generateFilename(stepNumber, action) {
+  const padded = String(stepNumber).padStart(3, "0");
+  const slug = slugify(action);
+  return `${padded}_${slug}.png`;
+}
+function formatDate(date) {
   return date.toISOString().slice(0, 10);
 }
 function formatTime(date) {
@@ -5410,11 +6320,11 @@ function getScreenshotDir(baseDir, runId, scenarioSlug, projectName, timestamp)
   const project = projectName ?? "default";
   const dateDir = formatDate(now2);
   const timeDir = `${formatTime(now2)}_${runId.slice(0, 8)}`;
-  return join3(baseDir, project, dateDir, timeDir, scenarioSlug);
+  return join4(baseDir, project, dateDir, timeDir, scenarioSlug);
 }
 function ensureDir(dirPath) {
   if (!existsSync3(dirPath)) {
-    mkdirSync2(dirPath, { recursive: true });
+    mkdirSync3(dirPath, { recursive: true });
   }
 }
 function writeMetaSidecar(screenshotPath, meta) {
@@ -5425,10 +6335,10 @@ function writeMetaSidecar(screenshotPath, meta) {
 }
 async function generateThumbnail(page, screenshotDir, filename) {
   try {
-    const thumbDir = join3(screenshotDir, "_thumbnail");
+    const thumbDir = join4(screenshotDir, "_thumbnail");
     ensureDir(thumbDir);
     const thumbFilename = filename.replace(/\.(png|jpeg)$/, ".thumb.$1");
-    const thumbPath = join3(thumbDir, thumbFilename);
+    const thumbPath = join4(thumbDir, thumbFilename);
     const viewport = page.viewportSize();
     if (viewport) {
       await page.screenshot({
@@ -5462,14 +6372,16 @@ class Screenshotter {
     const action = options.description ?? options.action;
     const dir = getScreenshotDir(this.baseDir, options.runId, options.scenarioSlug, this.projectName, this.runTimestamp);
     const filename = generateFilename(options.stepNumber, action);
-    const filePath = join3(dir, filename);
+    const filePath = join4(dir, filename);
     ensureDir(dir);
-    await page.screenshot({
+    const screenshotOpts = {
       path: filePath,
       fullPage: this.fullPage,
-      type: this.format,
-      quality: this.format === "jpeg" ? this.quality : undefined
-    });
+      type: this.format
+    };
+    if (this.format === "jpeg")
+      screenshotOpts.quality = this.quality;
+    await page.screenshot(screenshotOpts);
     const viewport = page.viewportSize() ?? { width: 0, height: 0 };
     const pageUrl = page.url();
     const timestamp = new Date().toISOString();
@@ -5497,14 +6409,16 @@ class Screenshotter {
     const action = options.description ?? options.action;
     const dir = getScreenshotDir(this.baseDir, options.runId, options.scenarioSlug, this.projectName, this.runTimestamp);
     const filename = generateFilename(options.stepNumber, action);
-    const filePath = join3(dir, filename);
+    const filePath = join4(dir, filename);
     ensureDir(dir);
-    await page.screenshot({
+    const ssOpts2 = {
       path: filePath,
       fullPage: true,
-      type: this.format,
-      quality: this.format === "jpeg" ? this.quality : undefined
-    });
+      type: this.format
+    };
+    if (this.format === "jpeg")
+      ssOpts2.quality = this.quality;
+    await page.screenshot(ssOpts2);
     const viewport = page.viewportSize() ?? { width: 0, height: 0 };
     const pageUrl = page.url();
     const timestamp = new Date().toISOString();
@@ -5532,13 +6446,15 @@ class Screenshotter {
     const action = options.description ?? options.action;
     const dir = getScreenshotDir(this.baseDir, options.runId, options.scenarioSlug, this.projectName, this.runTimestamp);
     const filename = generateFilename(options.stepNumber, action);
-    const filePath = join3(dir, filename);
+    const filePath = join4(dir, filename);
     ensureDir(dir);
-    await page.locator(selector).screenshot({
+    const ssOpts3 = {
       path: filePath,
-      type: this.format,
-      quality: this.format === "jpeg" ? this.quality : undefined
-    });
+      type: this.format
+    };
+    if (this.format === "jpeg")
+      ssOpts3.quality = this.quality;
+    await page.locator(selector).screenshot(ssOpts3);
     const viewport = page.viewportSize() ?? { width: 0, height: 0 };
     const pageUrl = page.url();
     const timestamp = new Date().toISOString();
@@ -5564,7 +6480,7 @@ class Screenshotter {
 }
 var DEFAULT_BASE_DIR;
 var init_screenshotter = __esm(() => {
-  DEFAULT_BASE_DIR = join3(homedir3(), ".testers", "screenshots");
+  DEFAULT_BASE_DIR = join4(homedir4(), ".testers", "screenshots");
 });
 
 // src/lib/webhooks.ts
@@ -5735,13 +6651,13 @@ async function pushFailedRunToLogs(run, failedResults, scenarios) {
 // src/lib/todos-connector.ts
 import { Database as Database2 } from "bun:sqlite";
 import { existsSync as existsSync4 } from "fs";
-import { join as join4 } from "path";
-import { homedir as homedir4 } from "os";
+import { join as join5 } from "path";
+import { homedir as homedir5 } from "os";
 function resolveTodosDbPath() {
   const envPath = process.env["TODOS_DB_PATH"];
   if (envPath)
     return envPath;
-  return join4(homedir4(), ".todos", "todos.db");
+  return join5(homedir5(), ".todos", "todos.db");
 }
 function connectToTodos() {
   const dbPath = resolveTodosDbPath();
@@ -5927,6 +6843,45 @@ async function notifyFailureToConversations(run, failedResults, scenarios) {
     });
   } catch {}
 }
+async function notifyRunToConversations(run, results, options) {
+  const baseUrl = process.env["TESTERS_CONVERSATIONS_URL"];
+  const space = options?.spaceId ?? process.env["TESTERS_CONVERSATIONS_SPACE"];
+  if (!baseUrl || !space)
+    return;
+  const passRate = run.total > 0 ? (run.passed / run.total * 100).toFixed(0) : "0";
+  const statusIcon = run.status === "passed" ? "\u2705" : run.status === "failed" ? "\u274C" : "\u26A0\uFE0F";
+  const durationSec = run.finishedAt && run.startedAt ? ((new Date(run.finishedAt).getTime() - new Date(run.startedAt).getTime()) / 1000).toFixed(1) : null;
+  const lines = [
+    `${statusIcon} **Testers run ${run.status.toUpperCase()}** \u2014 ${run.passed}/${run.total} scenarios (${passRate}% pass rate)`,
+    ``,
+    `**URL:** ${run.url}`,
+    `**Run ID:** \`${run.id}\``,
+    `**Model:** ${run.model}`,
+    durationSec ? `**Duration:** ${durationSec}s` : null
+  ].filter((l) => l !== null);
+  if (run.status === "failed") {
+    const failedResults = results.filter((r) => r.status === "failed" || r.status === "error");
+    const failLines = failedResults.slice(0, 5).map((r) => {
+      const err = r.error ? ` \u2014 ${r.error.slice(0, 100)}` : "";
+      return `  \u274C ${r.scenarioId.slice(0, 8)}${err}`;
+    });
+    if (failLines.length > 0) {
+      lines.push(``, `**Failures:**`);
+      lines.push(...failLines);
+      if (failedResults.length > 5)
+        lines.push(`  \u2026 and ${failedResults.length - 5} more`);
+    }
+  }
+  const message = lines.join(`
+`);
+  try {
+    await fetch(`${baseUrl.replace(/\/$/, "")}/api/spaces/${encodeURIComponent(space)}/messages`, {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({ content: message, from: "testers" })
+    });
+  } catch {}
+}
 var init_failure_pipeline = __esm(() => {
   init_todos_connector();
 });
@@ -6132,10 +7087,35 @@ async function runSingleScenario(scenario, runId, options) {
   const config = loadConfig();
   if (options.selfHeal !== undefined)
     config.selfHeal = options.selfHeal;
-  const model = resolveModel(options.model ?? scenario.model ?? config.defaultModel);
-  const client = createClientForModel(model, options.apiKey ?? config.anthropicApiKey);
+  let effectiveOptions = options;
+  if (options.minimal) {
+    effectiveOptions = {
+      ...options,
+      engine: options.engine ?? "playwright"
+    };
+    try {
+      const { isLightpandaAvailable: isLightpandaAvailable2 } = await Promise.resolve().then(() => (init_browser_lightpanda(), exports_browser_lightpanda)).catch(() => ({ isLightpandaAvailable: () => false }));
+      if (isLightpandaAvailable2())
+        effectiveOptions = { ...effectiveOptions, engine: "lightpanda" };
+    } catch {}
+  }
+  const model = resolveModel(effectiveOptions.minimal ? "quick" : effectiveOptions.model ?? scenario.model ?? config.defaultModel);
+  if (options.cacheMaxAgeMs && options.cacheMaxAgeMs > 0 && scenario.lastPassedAt && scenario.lastPassedUrl === options.url) {
+    const age = Date.now() - new Date(scenario.lastPassedAt).getTime();
+    if (age < options.cacheMaxAgeMs) {
+      const cached = createResult({ runId, scenarioId: scenario.id, model, stepsTotal: 0 });
+      return updateResult(cached.id, {
+        status: "passed",
+        reasoning: `Cache hit: passed ${Math.round(age / 1000)}s ago at ${options.url}`,
+        stepsCompleted: 0,
+        durationMs: 0,
+        tokensUsed: 0
+      });
+    }
+  }
+  const client = createClientForModel(model, effectiveOptions.apiKey ?? config.anthropicApiKey);
   const screenshotter = new Screenshotter({
-    baseDir: options.screenshotDir ?? config.screenshots.dir
+    baseDir: effectiveOptions.screenshotDir ?? config.screenshots.dir
   });
   const resolvedPersonaId = options.personaId ?? scenario.personaId;
   const persona = resolvedPersonaId ? getPersona(resolvedPersonaId) : null;
@@ -6151,12 +7131,20 @@ async function runSingleScenario(scenario, runId, options) {
   let browser = null;
   let page = null;
   try {
-    browser = await launchBrowser({ headless: !(options.headed ?? false), engine: options.engine });
+    browser = await launchBrowser({ headless: !(effectiveOptions.headed ?? false), engine: effectiveOptions.engine });
     page = await getPage(browser, {
       viewport: config.browser.viewport
     });
     const targetUrl = scenario.targetPath ? `${options.url.replace(/\/$/, "")}${scenario.targetPath}` : options.url;
     const scenarioTimeout = scenario.timeoutMs ?? options.timeout ?? config.browser.timeout ?? 60000;
+    const consoleErrors = [];
+    page.on("console", (msg) => {
+      if (msg.type() === "error")
+        consoleErrors.push(msg.text());
+    });
+    page.on("pageerror", (err) => {
+      consoleErrors.push(err.message);
+    });
     await page.goto(targetUrl, { timeout: Math.min(scenarioTimeout, 30000) });
     const stepStartTimes = new Map;
     const agentResult = await withTimeout(runAgentLoop({
@@ -6166,15 +7154,17 @@ async function runSingleScenario(scenario, runId, options) {
       screenshotter,
       model,
       runId,
-      maxTurns: 30,
-      a11y: options.a11y,
+      maxTurns: effectiveOptions.minimal ? 10 : 30,
+      a11y: effectiveOptions.a11y,
       persona: persona ? {
         name: persona.name,
         role: persona.role,
         description: persona.description,
         instructions: persona.instructions,
         traits: persona.traits,
-        goals: persona.goals
+        goals: persona.goals,
+        behaviors: persona.behaviors,
+        painPoints: persona.painPoints
       } : null,
       onStep: (stepEvent) => {
         let stepDurationMs;
@@ -6201,7 +7191,7 @@ async function runSingleScenario(scenario, runId, options) {
         });
       }
     }), scenarioTimeout, scenario.name);
-    if (options.engine !== "lightpanda") {
+    if (options.engine !== "lightpanda" && options.engine !== "bun") {
       for (const ss of agentResult.screenshots) {
         try {
           createScreenshot({
@@ -6219,8 +7209,8 @@ async function runSingleScenario(scenario, runId, options) {
         } catch {}
       }
     }
-    const lightpandaNote = options.engine === "lightpanda" ? " (Running with Lightpanda \u2014 no screenshots)" : "";
-    const updatedResult = updateResult(result.id, {
+    const lightpandaNote = options.engine === "lightpanda" ? " (Running with Lightpanda \u2014 no screenshots)" : options.engine === "bun" ? " (Running with Bun.WebView \u2014 native, ~11x faster)" : "";
+    let updatedResult = updateResult(result.id, {
       status: agentResult.status,
       reasoning: agentResult.reasoning ? agentResult.reasoning + lightpandaNote : lightpandaNote || undefined,
       stepsCompleted: agentResult.stepsCompleted,
@@ -6228,29 +7218,53 @@ async function runSingleScenario(scenario, runId, options) {
       tokensUsed: agentResult.tokensUsed,
       costCents: estimateCost(model, agentResult.tokensUsed)
     });
+    if (agentResult.status === "failed" || agentResult.status === "error") {
+      const failureAnalysis = analyzeFailure(null, agentResult.reasoning ?? null);
+      if (failureAnalysis) {
+        updatedResult = updateResult(result.id, { failureAnalysis });
+      }
+    }
+    if (agentResult.status === "passed") {
+      try {
+        updateScenarioPassedCache(scenario.id, options.url);
+      } catch {}
+    }
     const eventType = agentResult.status === "passed" ? "scenario:pass" : "scenario:fail";
     emit({ type: eventType, scenarioId: scenario.id, scenarioName: scenario.name, resultId: result.id, runId });
     return updatedResult;
   } catch (error) {
     const errorMsg = error instanceof Error ? error.message : String(error);
-    const updatedResult = updateResult(result.id, {
+    let updatedResult = updateResult(result.id, {
       status: "error",
       error: errorMsg,
       durationMs: Date.now() - new Date(result.createdAt).getTime()
     });
+    const failureAnalysis = analyzeFailure(errorMsg, null);
+    if (failureAnalysis) {
+      updatedResult = updateResult(result.id, { failureAnalysis });
+    }
     emit({ type: "scenario:error", scenarioId: scenario.id, scenarioName: scenario.name, error: errorMsg, runId });
     return updatedResult;
   } finally {
     if (browser)
-      await closeBrowser(browser, options.engine);
+      await closeBrowser(browser, effectiveOptions.engine);
   }
 }
 async function runBatch(scenarios, options) {
   const config = loadConfig();
-  const model = resolveModel(options.model ?? config.defaultModel);
-  const parallel = options.parallel ?? 1;
+  const model = resolveModel(options.minimal ? "quick" : options.model ?? config.defaultModel);
+  const parallel = options.minimal ? Math.max(5, options.parallel ?? 1) : options.parallel ?? 1;
   const samples = options.samples ?? 1;
   const flakinessThreshold = options.flakinessThreshold ?? 0.95;
+  if (!options.skipBudgetCheck) {
+    const cap = options.maxCostCents ?? config.defaultMaxCostCents;
+    if (cap !== undefined && cap > 0) {
+      const estimated = estimateRunCostCents(scenarios.length, model, samples);
+      if (estimated > cap) {
+        throw new BudgetExceededError(estimated, cap);
+      }
+    }
+  }
   const run = createRun({
     url: options.url,
     model,
@@ -6392,6 +7406,10 @@ async function runBatch(scenarios, options) {
     createFailureTasks(finalRun, failedResults, scenarios).catch(() => {});
     notifyFailureToConversations(finalRun, failedResults, scenarios).catch(() => {});
   }
+  const conversationsSpaceId = config.conversationsSpace ?? process.env["TESTERS_CONVERSATIONS_SPACE"];
+  if (conversationsSpaceId) {
+    notifyRunToConversations(finalRun, results, { spaceId: conversationsSpaceId }).catch(() => {});
+  }
   return { run: finalRun, results };
 }
 async function runByFilter(options) {
@@ -6429,6 +7447,16 @@ function startRunAsync(options) {
       priority: options.priority
     });
   }
+  if (!options.skipBudgetCheck) {
+    const cap = options.maxCostCents ?? config.defaultMaxCostCents;
+    if (cap !== undefined && cap > 0 && scenarios.length > 0) {
+      const samples = options.samples ?? 1;
+      const estimated = estimateRunCostCents(scenarios.length, model, samples);
+      if (estimated > cap) {
+        throw new BudgetExceededError(estimated, cap);
+      }
+    }
+  }
   const parallel = options.parallel ?? 1;
   const run = createRun({
     url: options.url,
@@ -6505,9 +7533,11 @@ function estimateCost(model, tokens) {
 }
 var eventHandler = null;
 var init_runner = __esm(() => {
+  init_types();
   init_eval_runner();
   init_runs();
   init_results();
+  init_costs();
   init_screenshots();
   init_scenarios();
   init_personas();
@@ -6532,7 +7562,7 @@ __export(exports_reporter, {
   formatJSON: () => formatJSON,
   formatActionableSummary: () => formatActionableSummary
 });
-import chalk from "chalk";
+import chalk2 from "chalk";
 function useEmoji() {
   return !process.env["NO_COLOR"] && process.argv.indexOf("--no-color") === -1;
 }
@@ -6540,13 +7570,13 @@ function formatTerminal(run, results, options) {
   const lines = [];
   const failedOnly = options?.failedOnly ?? false;
   lines.push("");
-  lines.push(chalk.bold(`  Run ${run.id.slice(0, 8)} \u2014 ${run.url}`));
-  lines.push(chalk.dim(`  Model: ${run.model} | Parallel: ${run.parallel} | Headed: ${run.headed ? "yes" : "no"}`));
+  lines.push(chalk2.bold(`  Run ${run.id.slice(0, 8)} \u2014 ${run.url}`));
+  lines.push(chalk2.dim(`  Model: ${run.model} | Parallel: ${run.parallel} | Headed: ${run.headed ? "yes" : "no"}`));
   lines.push("");
   if (failedOnly) {
     const passedCount = results.filter((r) => r.status === "passed").length;
     if (passedCount > 0) {
-      lines.push(chalk.dim(`  (${passedCount} passed scenario${passedCount !== 1 ? "s" : ""} hidden \u2014 use without --failed-only to see all)`));
+      lines.push(chalk2.dim(`  (${passedCount} passed scenario${passedCount !== 1 ? "s" : ""} hidden \u2014 use without --failed-only to see all)`));
       lines.push("");
     }
   }
@@ -6564,28 +7594,28 @@ function formatTerminal(run, results, options) {
     const emoji = useEmoji();
     switch (result.status) {
       case "passed":
-        statusIcon = emoji ? "\u2705" : chalk.green("PASS");
-        statusColor = chalk.green;
+        statusIcon = emoji ? "\u2705" : chalk2.green("PASS");
+        statusColor = chalk2.green;
         break;
       case "failed":
-        statusIcon = emoji ? "\u274C" : chalk.red("FAIL");
-        statusColor = chalk.red;
+        statusIcon = emoji ? "\u274C" : chalk2.red("FAIL");
+        statusColor = chalk2.red;
         break;
       case "error":
-        statusIcon = emoji ? "\u26A0\uFE0F " : chalk.yellow("ERR ");
-        statusColor = chalk.yellow;
+        statusIcon = emoji ? "\u26A0\uFE0F " : chalk2.yellow("ERR ");
+        statusColor = chalk2.yellow;
         break;
       default:
-        statusIcon = emoji ? "\u23ED\uFE0F " : chalk.dim("SKIP");
-        statusColor = chalk.dim;
+        statusIcon = emoji ? "\u23ED\uFE0F " : chalk2.dim("SKIP");
+        statusColor = chalk2.dim;
         break;
     }
-    lines.push(`  ${statusIcon}  ${statusColor(name)}  ${chalk.dim(duration)}  ${chalk.dim(`${screenshotCount} screenshots`)}`);
+    lines.push(`  ${statusIcon}  ${statusColor(name)}  ${chalk2.dim(duration)}  ${chalk2.dim(`${screenshotCount} screenshots`)}`);
     if (result.reasoning && (result.status === "failed" || result.status === "error")) {
-      lines.push(chalk.dim(`         ${result.reasoning}`));
+      lines.push(chalk2.dim(`         ${result.reasoning}`));
     }
     if (result.error) {
-      lines.push(chalk.red(`         ${result.error}`));
+      lines.push(chalk2.red(`         ${result.error}`));
     }
   }
   lines.push("");
@@ -6596,9 +7626,9 @@ function formatTerminal(run, results, options) {
 }
 function formatSummary(run) {
   const duration = run.finishedAt ? `${((new Date(run.finishedAt).getTime() - new Date(run.startedAt).getTime()) / 1000).toFixed(1)}s` : "running";
-  const passedStr = chalk.green(`${run.passed} passed`);
-  const failedStr = run.failed > 0 ? chalk.red(` ${run.failed} failed`) : "";
-  const totalStr = chalk.dim(` (${run.total} total)`);
+  const passedStr = chalk2.green(`${run.passed} passed`);
+  const failedStr = run.failed > 0 ? chalk2.red(` ${run.failed} failed`) : "";
+  const totalStr = chalk2.dim(` (${run.total} total)`);
   return `  ${passedStr}${failedStr}${totalStr} in ${duration}`;
 }
 function formatActionableSummary(run, results) {
@@ -6609,18 +7639,18 @@ function formatActionableSummary(run, results) {
   const passStr = `${emoji ? "\u2705" : "PASS"} ${passedCount} passed`;
   const failStr = failedCount > 0 ? `  ${emoji ? "\u274C" : "FAIL"} ${failedCount} failed` : "";
   const lines = [];
-  lines.push(`  ${chalk.bold(passStr)}${failedCount > 0 ? chalk.bold(failStr) : ""}`);
+  lines.push(`  ${chalk2.bold(passStr)}${failedCount > 0 ? chalk2.bold(failStr) : ""}`);
   if (failedCount > 0) {
-    lines.push(chalk.dim(`  retry failed: testers retry ${shortId}  |  view: testers results ${shortId}`));
+    lines.push(chalk2.dim(`  retry failed: testers retry ${shortId}  |  view: testers results ${shortId}`));
   } else {
-    lines.push(chalk.dim(`  view: testers results ${shortId}`));
+    lines.push(chalk2.dim(`  view: testers results ${shortId}`));
   }
   const totalCostCents = results.reduce((sum, r) => sum + (r.costCents ?? 0), 0);
   const totalTokens = results.reduce((sum, r) => sum + (r.tokensUsed ?? 0), 0);
   if (totalTokens > 0) {
     const costStr = `$${(totalCostCents / 100).toFixed(4)}`;
     const tokensStr = totalTokens.toLocaleString();
-    lines.push(chalk.dim(`  ${emoji ? "\uD83D\uDCB0" : "cost:"} Cost: ${costStr} (${tokensStr} tokens)`));
+    lines.push(chalk2.dim(`  ${emoji ? "\uD83D\uDCB0" : "cost:"} Cost: ${costStr} (${tokensStr} tokens)`));
   }
   return lines.join(`
 `);
@@ -6685,19 +7715,19 @@ function getExitCode(run) {
 function formatRunList(runs) {
   const lines = [];
   lines.push("");
-  lines.push(chalk.bold("  Recent Runs"));
+  lines.push(chalk2.bold("  Recent Runs"));
   lines.push("");
   if (runs.length === 0) {
-    lines.push(chalk.dim("  No runs found."));
+    lines.push(chalk2.dim("  No runs found."));
     lines.push("");
     return lines.join(`
 `);
   }
   for (const run of runs) {
-    const statusIcon = run.status === "passed" ? chalk.green("PASS") : run.status === "failed" ? chalk.red("FAIL") : run.status === "running" ? chalk.blue("RUN ") : chalk.dim(run.status.toUpperCase().padEnd(4));
+    const statusIcon = run.status === "passed" ? chalk2.green("PASS") : run.status === "failed" ? chalk2.red("FAIL") : run.status === "running" ? chalk2.blue("RUN ") : chalk2.dim(run.status.toUpperCase().padEnd(4));
     const date = new Date(run.startedAt).toLocaleString();
     const id = run.id.slice(0, 8);
-    lines.push(`  ${statusIcon}  ${chalk.dim(id)}  ${run.url}  ${chalk.dim(`${run.passed}/${run.total}`)}  ${chalk.dim(date)}`);
+    lines.push(`  ${statusIcon}  ${chalk2.dim(id)}  ${run.url}  ${chalk2.dim(`${run.passed}/${run.total}`)}  ${chalk2.dim(date)}`);
   }
   lines.push("");
   return lines.join(`
@@ -6715,32 +7745,33 @@ function getScenarioRunStats(scenarioId) {
 function formatScenarioList(scenarios) {
   const lines = [];
   lines.push("");
-  lines.push(chalk.bold("  Scenarios"));
+  lines.push(chalk2.bold("  Scenarios"));
   lines.push("");
   if (scenarios.length === 0) {
-    lines.push(chalk.dim("  No scenarios found. Use 'testers add' to create one."));
+    lines.push(chalk2.dim("  No scenarios found. Use 'testers add' to create one."));
     lines.push("");
     return lines.join(`
 `);
   }
   for (const s of scenarios) {
-    const priorityColor = s.priority === "critical" ? chalk.red : s.priority === "high" ? chalk.yellow : s.priority === "medium" ? chalk.blue : chalk.dim;
-    const tags = s.tags.length > 0 ? chalk.dim(` [${s.tags.join(", ")}]`) : "";
-    let lastStatusIcon = chalk.dim("\u2014");
-    let passRateStr = chalk.dim("\u2014");
+    const priorityColor = s.priority === "critical" ? chalk2.red : s.priority === "high" ? chalk2.yellow : s.priority === "medium" ? chalk2.blue : chalk2.dim;
+    const tags = s.tags.length > 0 ? chalk2.dim(` [${s.tags.join(", ")}]`) : "";
+    let lastStatusIcon = chalk2.dim("\u2014");
+    let passRateStr = chalk2.dim("\u2014");
     if (s.id) {
       const stats = getScenarioRunStats(s.id);
       if (stats.lastStatus === "passed")
-        lastStatusIcon = chalk.green("\u2713");
+        lastStatusIcon = chalk2.green("\u2713");
       else if (stats.lastStatus === "failed")
-        lastStatusIcon = chalk.red("\u2717");
+        lastStatusIcon = chalk2.red("\u2717");
       else if (stats.lastStatus === "error")
-        lastStatusIcon = chalk.yellow("!");
+        lastStatusIcon = chalk2.yellow("!");
       else if (stats.lastStatus === "skipped")
-        lastStatusIcon = chalk.dim("~");
-      passRateStr = stats.passRate === "\u2014" ? chalk.dim("\u2014") : chalk.dim(stats.passRate);
+        lastStatusIcon = chalk2.dim("~");
+      passRateStr = stats.passRate === "\u2014" ? chalk2.dim("\u2014") : chalk2.dim(stats.passRate);
     }
-    lines.push(`  ${chalk.cyan(s.shortId)}  ${s.name}  ${priorityColor(s.priority)}${tags}  ${lastStatusIcon} ${passRateStr}`);
+    const flakinessStr = s.flakinessScore !== null && s.flakinessScore !== undefined && s.flakinessScore < 0.8 ? chalk2.yellow(` \u26A1 flaky (${Math.round(s.flakinessScore * 100)}%)`) : "";
+    lines.push(`  ${chalk2.cyan(s.shortId)}  ${s.name}  ${priorityColor(s.priority)}${tags}${flakinessStr}  ${lastStatusIcon} ${passRateStr}`);
   }
   lines.push("");
   return lines.join(`
@@ -6750,30 +7781,30 @@ function formatResultDetail(result, screenshots) {
   const lines = [];
   const scenario = getScenario(result.scenarioId);
   lines.push("");
-  lines.push(chalk.bold(`  Result ${result.id.slice(0, 8)}`));
+  lines.push(chalk2.bold(`  Result ${result.id.slice(0, 8)}`));
   if (scenario) {
     lines.push(`  Scenario: ${scenario.shortId} \u2014 ${scenario.name}`);
   }
-  lines.push(`  Status: ${result.status === "passed" ? chalk.green("PASSED") : chalk.red(result.status.toUpperCase())}`);
+  lines.push(`  Status: ${result.status === "passed" ? chalk2.green("PASSED") : chalk2.red(result.status.toUpperCase())}`);
   lines.push(`  Model: ${result.model}`);
   lines.push(`  Duration: ${(result.durationMs / 1000).toFixed(1)}s`);
   lines.push(`  Steps: ${result.stepsCompleted}/${result.stepsTotal}`);
   lines.push(`  Tokens: ${result.tokensUsed} (~$${(result.costCents / 100).toFixed(4)})`);
   if (result.reasoning) {
     lines.push("");
-    lines.push(chalk.bold("  Reasoning:"));
+    lines.push(chalk2.bold("  Reasoning:"));
     lines.push(`  ${result.reasoning}`);
   }
   if (result.error) {
     lines.push("");
-    lines.push(chalk.red.bold("  Error:"));
-    lines.push(chalk.red(`  ${result.error}`));
+    lines.push(chalk2.red.bold("  Error:"));
+    lines.push(chalk2.red(`  ${result.error}`));
   }
   if (screenshots.length > 0) {
     lines.push("");
-    lines.push(chalk.bold(`  Screenshots (${screenshots.length}):`));
+    lines.push(chalk2.bold(`  Screenshots (${screenshots.length}):`));
     for (const ss of screenshots) {
-      lines.push(`  ${chalk.dim(`${String(ss.stepNumber).padStart(3, "0")}`)} ${ss.action} \u2014 ${chalk.dim(ss.filePath)}`);
+      lines.push(`  ${chalk2.dim(`${String(ss.stepNumber).padStart(3, "0")}`)} ${ss.action} \u2014 ${chalk2.dim(ss.filePath)}`);
     }
   }
   lines.push("");
@@ -7298,6 +8329,60 @@ async function postGitHubComment(run, results, options) {
   }
 }
 
+// src/lib/affected.ts
+var exports_affected = {};
+__export(exports_affected, {
+  matchFilesToScenarios: () => matchFilesToScenarios
+});
+function globToRegex(glob) {
+  const escaped = glob.replace(/[.+^${}()|[\]\\]/g, "\\$&").replace(/\*\*/g, "\x00DS\x00").replace(/\*/g, "[^/]*").replace(/\x00DS\x00/g, ".*");
+  return new RegExp(`^${escaped}$`, "i");
+}
+function matchFilesToScenarios(filePaths, scenarios, mappings = []) {
+  if (filePaths.length === 0)
+    return scenarios;
+  const compiledMappings = mappings.map((m) => ({
+    regex: globToRegex(m.glob),
+    tags: m.tags
+  }));
+  const normPaths = filePaths.map((p) => p.replace(/\\/g, "/").toLowerCase());
+  const matchedIds = new Set;
+  for (const scenario of scenarios) {
+    let matched = false;
+    if (!matched) {
+      for (const { regex, tags } of compiledMappings) {
+        if (normPaths.some((fp) => regex.test(fp)) && tags.some((tag) => scenario.tags.includes(tag))) {
+          matched = true;
+          break;
+        }
+      }
+    }
+    if (!matched && scenario.targetPath) {
+      const segments = scenario.targetPath.replace(/^\//, "").split("/").filter((s) => s.length > 2);
+      if (segments.some((seg) => normPaths.some((fp) => fp.includes(seg.toLowerCase())))) {
+        matched = true;
+      }
+    }
+    if (!matched) {
+      for (const tag of scenario.tags) {
+        if (tag.length > 2 && normPaths.some((fp) => fp.includes(tag.toLowerCase()))) {
+          matched = true;
+          break;
+        }
+      }
+    }
+    if (!matched) {
+      const nameWords = scenario.name.toLowerCase().split(/[\s\-_/]+/).filter((w) => w.length > 3);
+      if (nameWords.some((word) => normPaths.some((fp) => fp.includes(word)))) {
+        matched = true;
+      }
+    }
+    if (matched)
+      matchedIds.add(scenario.id);
+  }
+  return scenarios.filter((s) => matchedIds.has(s.id));
+}
+
 // src/lib/compliance-report.ts
 var exports_compliance_report = {};
 __export(exports_compliance_report, {
@@ -7963,60 +9048,6 @@ var init_recorder = __esm(() => {
   init_scenarios();
 });
 
-// src/lib/affected.ts
-var exports_affected = {};
-__export(exports_affected, {
-  matchFilesToScenarios: () => matchFilesToScenarios
-});
-function globToRegex(glob) {
-  const escaped = glob.replace(/[.+^${}()|[\]\\]/g, "\\$&").replace(/\*\*/g, "\x00DS\x00").replace(/\*/g, "[^/]*").replace(/\x00DS\x00/g, ".*");
-  return new RegExp(`^${escaped}$`, "i");
-}
-function matchFilesToScenarios(filePaths, scenarios, mappings = []) {
-  if (filePaths.length === 0)
-    return scenarios;
-  const compiledMappings = mappings.map((m) => ({
-    regex: globToRegex(m.glob),
-    tags: m.tags
-  }));
-  const normPaths = filePaths.map((p) => p.replace(/\\/g, "/").toLowerCase());
-  const matchedIds = new Set;
-  for (const scenario of scenarios) {
-    let matched = false;
-    if (!matched) {
-      for (const { regex, tags } of compiledMappings) {
-        if (normPaths.some((fp) => regex.test(fp)) && tags.some((tag) => scenario.tags.includes(tag))) {
-          matched = true;
-          break;
-        }
-      }
-    }
-    if (!matched && scenario.targetPath) {
-      const segments = scenario.targetPath.replace(/^\//, "").split("/").filter((s) => s.length > 2);
-      if (segments.some((seg) => normPaths.some((fp) => fp.includes(seg.toLowerCase())))) {
-        matched = true;
-      }
-    }
-    if (!matched) {
-      for (const tag of scenario.tags) {
-        if (tag.length > 2 && normPaths.some((fp) => fp.includes(tag.toLowerCase()))) {
-          matched = true;
-          break;
-        }
-      }
-    }
-    if (!matched) {
-      const nameWords = scenario.name.toLowerCase().split(/[\s\-_/]+/).filter((w) => w.length > 3);
-      if (nameWords.some((word) => normPaths.some((fp) => fp.includes(word)))) {
-        matched = true;
-      }
-    }
-    if (matched)
-      matchedIds.add(scenario.id);
-  }
-  return scenarios.filter((s) => matchedIds.has(s.id));
-}
-
 // src/lib/git-watch.ts
 var exports_git_watch = {};
 __export(exports_git_watch, {
@@ -9081,6 +10112,102 @@ var init_health_scan = __esm(() => {
   init_todos_connector();
 });
 
+// src/db/seed-personas.ts
+var exports_seed_personas = {};
+__export(exports_seed_personas, {
+  seedDefaultPersonas: () => seedDefaultPersonas,
+  DEFAULT_PERSONAS: () => DEFAULT_PERSONAS
+});
+function seedDefaultPersonas() {
+  const existing = listPersonas({ globalOnly: true });
+  if (existing.length > 0)
+    return { seeded: 0, skipped: DEFAULT_PERSONAS.length };
+  let seeded = 0;
+  for (const p of DEFAULT_PERSONAS) {
+    try {
+      createPersona(p);
+      seeded++;
+    } catch {}
+  }
+  return { seeded, skipped: DEFAULT_PERSONAS.length - seeded };
+}
+var DEFAULT_PERSONAS;
+var init_seed_personas = __esm(() => {
+  init_personas();
+  DEFAULT_PERSONAS = [
+    {
+      name: "First-Time User",
+      role: "first-time user who has never used this app",
+      description: "A new user encountering the product for the first time. No prior knowledge of the interface or flows.",
+      instructions: "Explore cautiously. Read labels carefully before clicking. If something is unclear, hesitate and try the most obvious option. Notice and comment on anything confusing or unexpected.",
+      traits: ["cautious", "reads-instructions", "easily-confused", "asks-why"],
+      goals: ["complete the main task", "understand what the app does", "not make mistakes"],
+      behaviors: ["reads every label before clicking", "hovers over elements before interacting", "notices missing help text"],
+      painPoints: ["unclear error messages", "no onboarding guidance", "confusing navigation labels"]
+    },
+    {
+      name: "Power User",
+      role: "experienced power user who uses this app daily",
+      description: "A seasoned user who knows all shortcuts and wants to accomplish tasks as fast as possible.",
+      instructions: "Move fast. Use keyboard shortcuts where possible. Skip tutorials. Go directly to the feature. Get frustrated by extra clicks.",
+      traits: ["fast", "impatient", "keyboard-first", "efficiency-focused"],
+      goals: ["accomplish tasks as fast as possible", "avoid unnecessary steps", "use advanced features"],
+      behaviors: ["skips onboarding modals immediately", "uses keyboard shortcuts", "ignores decorative UI elements"],
+      painPoints: ["forced multi-step wizards", "no keyboard shortcuts", "slow page loads"]
+    },
+    {
+      name: "Mobile User",
+      role: "user on a mobile device with a small screen",
+      description: "A user on a phone with limited screen space and touch-based interaction.",
+      instructions: "Simulate touch interactions (tap = click). Notice if buttons are too small to tap. Check if text is readable. Look for horizontal scroll issues.",
+      traits: ["touch-based", "limited-screen", "on-the-go", "interrupted"],
+      goals: ["complete tasks on a small screen", "find mobile-optimized flows", "identify layout issues"],
+      behaviors: ["taps instead of clicking", "scrolls vertically to find content", "notices overflow and truncated text"],
+      painPoints: ["tiny tap targets", "horizontal scrolling", "desktop-only modals"]
+    },
+    {
+      name: "Accessibility User",
+      role: "user with accessibility needs relying on keyboard navigation",
+      description: "A user who navigates primarily via keyboard and relies on semantic HTML and ARIA labels.",
+      instructions: "Navigate using Tab, Enter, Escape, and arrow keys only. Note if focus indicators are visible. Check if interactive elements have accessible labels.",
+      traits: ["keyboard-only", "screen-reader-compatible", "focus-dependent"],
+      goals: ["complete all tasks via keyboard", "verify accessibility compliance", "identify ARIA issues"],
+      behaviors: ["tabs through all interactive elements", "checks for visible focus rings", "reads aria-labels aloud"],
+      painPoints: ["missing focus indicators", "unlabeled icon buttons", "focus traps in modals"]
+    },
+    {
+      name: "Security Auditor",
+      role: "security-focused tester looking for vulnerabilities",
+      description: "A security professional trying to find injection vulnerabilities, unauthorized access, and data leaks.",
+      instructions: "Try edge cases in every input field. Attempt to access other users' data. Test form validation boundaries. Check for sensitive data exposure.",
+      traits: ["suspicious", "boundary-testing", "adversarial", "detail-oriented"],
+      goals: ["find security vulnerabilities", "test input validation", "verify authorization controls"],
+      behaviors: ["enters SQL/XSS payloads in input fields", "manipulates URL parameters", "checks network responses for sensitive data"],
+      painPoints: ["no input sanitization", "verbose error messages exposing internals", "missing CSRF protection"]
+    },
+    {
+      name: "Non-Technical User",
+      role: "non-technical user unfamiliar with software conventions",
+      description: "A user without technical background who is confused by jargon and relies on visual cues.",
+      instructions: "Avoid technical terminology. Be confused by 'API key', 'JSON', 'endpoint'. Look for visual cues and icons. Read all text literally.",
+      traits: ["non-technical", "jargon-confused", "visual-learner", "literal-reader"],
+      goals: ["understand what to do from visual cues only", "complete basic tasks", "identify confusing terminology"],
+      behaviors: ["reads every word on the screen", "looks for visual icons to understand actions", "asks 'what does this mean?'"],
+      painPoints: ["technical jargon without explanation", "developer-facing error codes", "settings with no plain-language descriptions"]
+    },
+    {
+      name: "Skeptical Buyer",
+      role: "potential customer evaluating the product before purchasing",
+      description: "A prospect who is not yet committed, looking for value and trust signals before converting.",
+      instructions: "Look for pricing, terms, and trust signals. Check for social proof. Try to find limitations. Look for hidden costs. Notice anything that creates doubt.",
+      traits: ["evaluating", "price-conscious", "trust-seeking", "skeptical"],
+      goals: ["evaluate whether the product is worth it", "find the pricing", "understand limitations"],
+      behaviors: ["looks for pricing before signing up", "reads reviews and testimonials", "searches for 'free trial' or 'no credit card'"],
+      painPoints: ["hidden pricing", "no free tier or trial", "unclear cancellation policy"]
+    }
+  ];
+});
+
 // src/lib/persona-diff.ts
 var exports_persona_diff = {};
 __export(exports_persona_diff, {
@@ -9639,6 +10766,170 @@ var init_golden_monitor = __esm(() => {
   init_golden_answers();
 });
 
+// src/lib/hybrid-runner.ts
+var exports_hybrid_runner = {};
+__export(exports_hybrid_runner, {
+  runHybridScenario: () => runHybridScenario
+});
+async function runDeterministicStep(page, step, baseUrl) {
+  try {
+    switch (step.type) {
+      case "navigate": {
+        const url = step.url.startsWith("http") ? step.url : `${baseUrl.replace(/\/$/, "")}${step.url}`;
+        await page.goto(url, { timeout: 30000 });
+        return { ok: true };
+      }
+      case "click":
+        await page.click(step.selector, { timeout: 1e4 });
+        return { ok: true };
+      case "fill":
+        await page.fill(step.selector, step.value, { timeout: 1e4 });
+        return { ok: true };
+      case "wait":
+        await new Promise((r) => setTimeout(r, step.ms));
+        return { ok: true };
+      case "wait_for":
+        await page.waitForSelector(step.selector, { timeout: step.timeoutMs ?? 1e4 });
+        return { ok: true };
+      case "screenshot":
+        await page.screenshot({ fullPage: false });
+        return { ok: true };
+      case "assert_text": {
+        const text = await page.locator(step.selector).textContent({ timeout: 5000 });
+        const actual = text ?? "";
+        const ok = step.contains ? actual.includes(step.expected) : actual.trim() === step.expected.trim();
+        if (!ok)
+          return { ok: false, error: `Expected "${step.expected}", got "${actual}"` };
+        return { ok: true };
+      }
+      case "assert_visible": {
+        const shouldBeVisible = step.visible !== false;
+        const count = await page.locator(step.selector).count();
+        const isVisible = count > 0;
+        if (shouldBeVisible && !isVisible)
+          return { ok: false, error: `Expected ${step.selector} to be visible` };
+        if (!shouldBeVisible && isVisible)
+          return { ok: false, error: `Expected ${step.selector} to be hidden` };
+        return { ok: true };
+      }
+      default:
+        return { ok: true };
+    }
+  } catch (err) {
+    return { ok: false, error: err instanceof Error ? err.message : String(err) };
+  }
+}
+async function runHybridScenario(scenario, options) {
+  const config = loadConfig();
+  const baseUrl = scenario.baseUrl ?? options?.baseUrl ?? "http://localhost:3000";
+  const startTime = Date.now();
+  const stepResults = [];
+  let browser = null;
+  let page = null;
+  try {
+    browser = await launchBrowser({ headless: true });
+    page = await getPage(browser, { viewport: config.browser.viewport });
+    const screenshotter = new Screenshotter({ baseDir: options?.screenshotDir ?? config.screenshots.dir });
+    for (let i = 0;i < scenario.steps.length; i++) {
+      const step = scenario.steps[i];
+      const stepStart = Date.now();
+      if (step.type === "ai" || step.type === "ai_verify") {
+        const model = resolveModel(step.model ?? scenario.model ?? config.defaultModel);
+        const client = createClientForModel(model, options?.apiKey ?? config.anthropicApiKey);
+        const instruction = step.type === "ai_verify" ? `Verify the following assertion about the current page state: "${step.assertion}". Do NOT navigate. Just inspect the page and call report_result with pass or fail.` : step.instruction;
+        const syntheticScenario = {
+          id: `hybrid-step-${i}`,
+          shortId: `hs-${i}`,
+          projectId: null,
+          name: `${scenario.name} \u2014 step ${i + 1}`,
+          description: instruction,
+          steps: [instruction],
+          tags: [],
+          priority: "medium",
+          model,
+          timeoutMs: scenario.timeoutMs ?? 60000,
+          targetPath: null,
+          requiresAuth: false,
+          authConfig: null,
+          metadata: null,
+          assertions: [],
+          personaId: null,
+          scenarioType: "browser",
+          requiredRole: null,
+          version: 1,
+          createdAt: new Date().toISOString(),
+          updatedAt: new Date().toISOString(),
+          lastPassedAt: null,
+          lastPassedUrl: null
+        };
+        try {
+          const agentResult = await runAgentLoop({
+            client,
+            page,
+            scenario: syntheticScenario,
+            screenshotter,
+            model,
+            runId: `hybrid-${Date.now()}`,
+            maxTurns: step.maxTurns ?? 15
+          });
+          stepResults.push({
+            stepIndex: i,
+            type: step.type,
+            status: agentResult.status === "passed" ? "passed" : "failed",
+            durationMs: Date.now() - stepStart,
+            reasoning: agentResult.reasoning,
+            error: agentResult.status !== "passed" ? agentResult.reasoning : undefined
+          });
+          if (agentResult.status !== "passed") {
+            return {
+              scenarioName: scenario.name,
+              status: "failed",
+              stepResults,
+              durationMs: Date.now() - startTime,
+              error: `Step ${i + 1} (ai): ${agentResult.reasoning}`
+            };
+          }
+        } catch (err) {
+          const msg = err instanceof Error ? err.message : String(err);
+          stepResults.push({ stepIndex: i, type: step.type, status: "failed", durationMs: Date.now() - stepStart, error: msg });
+          return { scenarioName: scenario.name, status: "failed", stepResults, durationMs: Date.now() - startTime, error: msg };
+        }
+      } else {
+        const result = await runDeterministicStep(page, step, baseUrl);
+        stepResults.push({
+          stepIndex: i,
+          type: step.type,
+          status: result.ok ? "passed" : "failed",
+          durationMs: Date.now() - stepStart,
+          error: result.error
+        });
+        if (!result.ok) {
+          return {
+            scenarioName: scenario.name,
+            status: "failed",
+            stepResults,
+            durationMs: Date.now() - startTime,
+            error: `Step ${i + 1} (${step.type}): ${result.error}`
+          };
+        }
+      }
+    }
+    return { scenarioName: scenario.name, status: "passed", stepResults, durationMs: Date.now() - startTime };
+  } catch (err) {
+    const msg = err instanceof Error ? err.message : String(err);
+    return { scenarioName: scenario.name, status: "error", stepResults, durationMs: Date.now() - startTime, error: msg };
+  } finally {
+    if (browser)
+      await closeBrowser(browser);
+  }
+}
+var init_hybrid_runner = __esm(() => {
+  init_browser();
+  init_screenshotter();
+  init_ai_client();
+  init_config();
+});
+
 // node_modules/commander/esm.mjs
 var import__ = __toESM(require_commander(), 1);
 var {
@@ -9660,7 +10951,7 @@ import chalk6 from "chalk";
 // package.json
 var package_default = {
   name: "@hasna/testers",
-  version: "0.0.15",
+  version: "0.0.16",
   description: "AI-powered QA testing CLI \u2014 spawns cheap AI agents to test web apps with headless browsers",
   type: "module",
   main: "dist/index.js",
@@ -9716,14 +11007,14 @@ var package_default = {
     bun: ">=1.0.0"
   },
   publishConfig: {
-    access: "public",
-    registry: "https://registry.npmjs.org/"
+    registry: "https://registry.npmjs.org",
+    access: "public"
   },
   repository: {
     type: "git",
     url: "https://github.com/hasna/open-testers.git"
   },
-  license: "MIT",
+  license: "Apache-2.0",
   keywords: [
     "testing",
     "qa",
@@ -9751,13 +11042,13 @@ import { render, Box, Text, useInput, useApp } from "ink";
 import React, { useState } from "react";
 import { readFileSync as readFileSync7, readdirSync, writeFileSync as writeFileSync3 } from "fs";
 import { createInterface } from "readline";
-import { join as join6, resolve } from "path";
+import { join as join7, resolve } from "path";
 
 // src/lib/init.ts
 init_scenarios();
-import { existsSync as existsSync5, readFileSync as readFileSync2, writeFileSync as writeFileSync2, mkdirSync as mkdirSync3 } from "fs";
-import { join as join5, basename } from "path";
-import { homedir as homedir5 } from "os";
+import { existsSync as existsSync5, readFileSync as readFileSync2, writeFileSync as writeFileSync2, mkdirSync as mkdirSync4 } from "fs";
+import { join as join6, basename } from "path";
+import { homedir as homedir6 } from "os";
 
 // src/db/projects.ts
 init_types();
@@ -9795,7 +11086,7 @@ function ensureProject(name, path) {
 
 // src/lib/init.ts
 function detectFramework(dir) {
-  const pkgPath = join5(dir, "package.json");
+  const pkgPath = join6(dir, "package.json");
   if (!existsSync5(pkgPath))
     return null;
   let pkg;
@@ -10015,10 +11306,10 @@ function initProject(options) {
   const project = ensureProject(name, projectPath);
   const starterInputs = getStarterScenarios(framework ?? { name: "Unknown", features: [] }, project.id);
   const scenarios = starterInputs.map((input) => createScenario(input));
-  const configDir = join5(homedir5(), ".testers");
-  const configPath = join5(configDir, "config.json");
+  const configDir = join6(homedir6(), ".testers");
+  const configPath = join6(configDir, "config.json");
   if (!existsSync5(configDir)) {
-    mkdirSync3(configDir, { recursive: true });
+    mkdirSync4(configDir, { recursive: true });
   }
   let config = {};
   if (existsSync5(configPath)) {
@@ -10258,7 +11549,7 @@ function formatSmokeReport(result) {
 init_runs();
 init_results();
 init_scenarios();
-import chalk2 from "chalk";
+import chalk3 from "chalk";
 function diffRuns(runId1, runId2) {
   const run1 = getRun(runId1);
   if (!run1) {
@@ -10348,630 +11639,398 @@ function formatDurationComparison(d1, d2) {
 function formatDiffTerminal(diff) {
   const lines = [];
   lines.push("");
-  lines.push(chalk2.bold("  Run Comparison"));
-  lines.push(`  Run 1: ${chalk2.dim(diff.run1.id.slice(0, 8))} (${diff.run1.status}) \u2014 ${diff.run1.startedAt}`);
-  lines.push(`  Run 2: ${chalk2.dim(diff.run2.id.slice(0, 8))} (${diff.run2.status}) \u2014 ${diff.run2.startedAt}`);
+  lines.push(chalk3.bold("  Run Comparison"));
+  lines.push(`  Run 1: ${chalk3.dim(diff.run1.id.slice(0, 8))} (${diff.run1.status}) \u2014 ${diff.run1.startedAt}`);
+  lines.push(`  Run 2: ${chalk3.dim(diff.run2.id.slice(0, 8))} (${diff.run2.status}) \u2014 ${diff.run2.startedAt}`);
   lines.push("");
   if (diff.regressions.length > 0) {
-    lines.push(chalk2.red.bold(`  Regressions (${diff.regressions.length}):`));
+    lines.push(chalk3.red.bold(`  Regressions (${diff.regressions.length}):`));
     for (const d of diff.regressions) {
       const label = formatScenarioLabel(d);
       const dur = formatDurationComparison(d.duration1, d.duration2);
-      lines.push(chalk2.red(`    \u2B07 ${label}  ${d.status1} -> ${d.status2}  ${chalk2.dim(dur)}`));
+      lines.push(chalk3.red(`    \u2B07 ${label}  ${d.status1} -> ${d.status2}  ${chalk3.dim(dur)}`));
     }
     lines.push("");
   }
   if (diff.fixes.length > 0) {
-    lines.push(chalk2.green.bold(`  Fixes (${diff.fixes.length}):`));
+    lines.push(chalk3.green.bold(`  Fixes (${diff.fixes.length}):`));
     for (const d of diff.fixes) {
       const label = formatScenarioLabel(d);
       const dur = formatDurationComparison(d.duration1, d.duration2);
-      lines.push(chalk2.green(`    \u2B06 ${label}  ${d.status1} -> ${d.status2}  ${chalk2.dim(dur)}`));
+      lines.push(chalk3.green(`    \u2B06 ${label}  ${d.status1} -> ${d.status2}  ${chalk3.dim(dur)}`));
     }
     lines.push("");
   }
   if (diff.unchanged.length > 0) {
-    lines.push(chalk2.dim(`  Unchanged (${diff.unchanged.length}):`));
+    lines.push(chalk3.dim(`  Unchanged (${diff.unchanged.length}):`));
     for (const d of diff.unchanged) {
       const label = formatScenarioLabel(d);
       const dur = formatDurationComparison(d.duration1, d.duration2);
-      lines.push(chalk2.dim(`    = ${label}  ${d.status2}  ${dur}`));
+      lines.push(chalk3.dim(`    = ${label}  ${d.status2}  ${dur}`));
     }
     lines.push("");
   }
   if (diff.newScenarios.length > 0) {
-    lines.push(chalk2.cyan(`  New in run 2 (${diff.newScenarios.length}):`));
+    lines.push(chalk3.cyan(`  New in run 2 (${diff.newScenarios.length}):`));
     for (const d of diff.newScenarios) {
       const label = formatScenarioLabel(d);
-      lines.push(chalk2.cyan(`    + ${label}  ${d.status2}`));
-    }
-    lines.push("");
-  }
-  if (diff.removedScenarios.length > 0) {
-    lines.push(chalk2.yellow(`  Removed from run 2 (${diff.removedScenarios.length}):`));
-    for (const d of diff.removedScenarios) {
-      const label = formatScenarioLabel(d);
-      lines.push(chalk2.yellow(`    - ${label}  was ${d.status1}`));
-    }
-    lines.push("");
-  }
-  lines.push(chalk2.bold(`  Summary: ${diff.regressions.length} regressions, ${diff.fixes.length} fixes, ${diff.unchanged.length} unchanged`));
-  lines.push("");
-  return lines.join(`
-`);
-}
-function formatDiffJSON(diff) {
-  return JSON.stringify(diff, null, 2);
-}
-
-// src/lib/visual-diff.ts
-init_screenshots();
-init_results();
-init_runs();
-init_scenarios();
-init_database();
-import { readFileSync as readFileSync3, existsSync as existsSync6 } from "fs";
-import chalk3 from "chalk";
-var DEFAULT_THRESHOLD = 0.1;
-function setBaseline(runId) {
-  const run = getRun(runId);
-  if (!run) {
-    throw new Error(`Run not found: ${runId}`);
-  }
-  const db2 = getDatabase();
-  if (run.projectId) {
-    db2.query("UPDATE runs SET is_baseline = 0 WHERE project_id = ? AND is_baseline = 1").run(run.projectId);
-  } else {
-    db2.query("UPDATE runs SET is_baseline = 0 WHERE project_id IS NULL AND is_baseline = 1").run();
-  }
-  updateRun(run.id, { is_baseline: 1 });
-}
-function compareImages(image1Path, image2Path) {
-  if (!existsSync6(image1Path)) {
-    throw new Error(`Baseline image not found: ${image1Path}`);
-  }
-  if (!existsSync6(image2Path)) {
-    throw new Error(`Current image not found: ${image2Path}`);
-  }
-  const buf1 = readFileSync3(image1Path);
-  const buf2 = readFileSync3(image2Path);
-  if (buf1.equals(buf2)) {
-    const estimatedPixels = Math.max(1, Math.floor(buf1.length / 4));
-    return { diffPercent: 0, diffPixels: 0, totalPixels: estimatedPixels };
-  }
-  if (buf1.length !== buf2.length) {
-    const maxLen = Math.max(buf1.length, buf2.length);
-    const estimatedPixels = Math.max(1, Math.floor(maxLen / 4));
-    return { diffPercent: 100, diffPixels: estimatedPixels, totalPixels: estimatedPixels };
-  }
-  let diffBytes = 0;
-  for (let i = 0;i < buf1.length; i++) {
-    if (buf1[i] !== buf2[i]) {
-      diffBytes++;
-    }
-  }
-  const totalPixels = Math.max(1, Math.floor(buf1.length / 4));
-  const diffPixels = Math.max(1, Math.floor(diffBytes / 4));
-  const diffPercent = parseFloat((diffBytes / buf1.length * 100).toFixed(4));
-  return { diffPercent, diffPixels, totalPixels };
-}
-function compareRunScreenshots(runId, baselineRunId, threshold = DEFAULT_THRESHOLD) {
-  const run = getRun(runId);
-  if (!run)
-    throw new Error(`Run not found: ${runId}`);
-  const baselineRun = getRun(baselineRunId);
-  if (!baselineRun)
-    throw new Error(`Baseline run not found: ${baselineRunId}`);
-  const currentResults = getResultsByRun(run.id);
-  const baselineResults = getResultsByRun(baselineRun.id);
-  const baselineMap = new Map;
-  for (const result of baselineResults) {
-    const screenshots = listScreenshots(result.id);
-    for (const ss of screenshots) {
-      const key = `${result.scenarioId}:${ss.stepNumber}`;
-      baselineMap.set(key, { path: ss.filePath, action: ss.action });
-    }
-  }
-  const results = [];
-  for (const result of currentResults) {
-    const screenshots = listScreenshots(result.id);
-    for (const ss of screenshots) {
-      const key = `${result.scenarioId}:${ss.stepNumber}`;
-      const baseline = baselineMap.get(key);
-      if (!baseline)
-        continue;
-      if (!existsSync6(baseline.path) || !existsSync6(ss.filePath))
-        continue;
-      try {
-        const comparison = compareImages(baseline.path, ss.filePath);
-        results.push({
-          scenarioId: result.scenarioId,
-          stepNumber: ss.stepNumber,
-          action: ss.action,
-          baselinePath: baseline.path,
-          currentPath: ss.filePath,
-          diffPercent: comparison.diffPercent,
-          isRegression: comparison.diffPercent > threshold
-        });
-      } catch {}
-    }
-  }
-  return results;
-}
-function formatVisualDiffTerminal(results, threshold = DEFAULT_THRESHOLD) {
-  if (results.length === 0) {
-    return chalk3.dim(`
-  No screenshot comparisons found.
-`);
-  }
-  const lines = [];
-  lines.push("");
-  lines.push(chalk3.bold("  Visual Regression Summary"));
-  lines.push("");
-  const regressions = results.filter((r) => r.diffPercent >= threshold);
-  const passed = results.filter((r) => r.diffPercent < threshold);
-  if (regressions.length > 0) {
-    lines.push(chalk3.red.bold(`  Regressions (${regressions.length}):`));
-    for (const r of regressions) {
-      const scenario = getScenario(r.scenarioId);
-      const label = scenario ? `${scenario.shortId}: ${scenario.name}` : r.scenarioId.slice(0, 8);
-      const pct = chalk3.red(`${r.diffPercent.toFixed(2)}%`);
-      lines.push(`    ${chalk3.red("!")} ${label} step ${r.stepNumber} (${r.action}) \u2014 ${pct} diff`);
+      lines.push(chalk3.cyan(`    + ${label}  ${d.status2}`));
     }
     lines.push("");
   }
-  if (passed.length > 0) {
-    lines.push(chalk3.green.bold(`  Passed (${passed.length}):`));
-    for (const r of passed) {
-      const scenario = getScenario(r.scenarioId);
-      const label = scenario ? `${scenario.shortId}: ${scenario.name}` : r.scenarioId.slice(0, 8);
-      const pct = chalk3.green(`${r.diffPercent.toFixed(2)}%`);
-      lines.push(`    ${chalk3.green("\u2713")} ${label} step ${r.stepNumber} (${r.action}) \u2014 ${pct} diff`);
+  if (diff.removedScenarios.length > 0) {
+    lines.push(chalk3.yellow(`  Removed from run 2 (${diff.removedScenarios.length}):`));
+    for (const d of diff.removedScenarios) {
+      const label = formatScenarioLabel(d);
+      lines.push(chalk3.yellow(`    - ${label}  was ${d.status1}`));
     }
     lines.push("");
   }
-  lines.push(chalk3.bold(`  Visual Summary: ${regressions.length} regressions, ${passed.length} passed (threshold: ${threshold}%)`));
+  lines.push(chalk3.bold(`  Summary: ${diff.regressions.length} regressions, ${diff.fixes.length} fixes, ${diff.unchanged.length} unchanged`));
   lines.push("");
   return lines.join(`
 `);
 }
+function formatDiffJSON(diff) {
+  return JSON.stringify(diff, null, 2);
+}
 
-// src/lib/report.ts
-init_runs();
-init_results();
+// src/lib/visual-diff.ts
 init_screenshots();
+init_results();
+init_runs();
 init_scenarios();
-import { readFileSync as readFileSync4, existsSync as existsSync7 } from "fs";
-function imageToBase64(filePath) {
-  if (!filePath || !existsSync7(filePath))
-    return "";
-  try {
-    const buffer = readFileSync4(filePath);
-    const base64 = buffer.toString("base64");
-    return `data:image/png;base64,${base64}`;
-  } catch {
-    return "";
-  }
-}
-function escapeHtml(text) {
-  return text.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;").replace(/"/g, "&quot;").replace(/'/g, "&#039;");
-}
-function formatDuration2(ms) {
-  if (ms < 1000)
-    return `${ms}ms`;
-  if (ms < 60000)
-    return `${(ms / 1000).toFixed(1)}s`;
-  const mins = Math.floor(ms / 60000);
-  const secs = (ms % 60000 / 1000).toFixed(0);
-  return `${mins}m ${secs}s`;
-}
-function formatCost(cents) {
-  if (cents < 1)
-    return `$${(cents / 100).toFixed(4)}`;
-  return `$${(cents / 100).toFixed(2)}`;
-}
-function statusBadge(status) {
-  const colors = {
-    passed: { bg: "#22c55e", text: "#000" },
-    failed: { bg: "#ef4444", text: "#fff" },
-    error: { bg: "#eab308", text: "#000" },
-    skipped: { bg: "#6b7280", text: "#fff" }
-  };
-  const c = colors[status] ?? { bg: "#6b7280", text: "#fff" };
-  const label = status.toUpperCase();
-  return `<span style="display:inline-block;padding:2px 10px;border-radius:4px;font-size:12px;font-weight:700;background:${c.bg};color:${c.text};letter-spacing:0.5px;">${label}</span>`;
-}
-function renderScreenshots(screenshots) {
-  if (screenshots.length === 0)
-    return "";
-  let html = `<div style="display:flex;flex-wrap:wrap;gap:12px;margin-top:12px;">`;
-  for (let i = 0;i < screenshots.length; i++) {
-    const ss = screenshots[i];
-    const dataUri = imageToBase64(ss.filePath);
-    const checkId = `ss-${ss.id}`;
-    if (dataUri) {
-      html += `
-        <div style="flex:0 0 auto;">
-          <input type="checkbox" id="${checkId}" style="display:none;" />
-          <label for="${checkId}" style="cursor:pointer;">
-            <img src="${dataUri}" alt="Step ${ss.stepNumber}: ${escapeHtml(ss.action)}"
-              style="max-width:200px;max-height:150px;border-radius:6px;border:1px solid #262626;display:block;" />
-          </label>
-          <div style="position:fixed;top:0;left:0;width:100%;height:100%;background:rgba(0,0,0,0.9);z-index:1000;display:none;align-items:center;justify-content:center;">
-            <label for="${checkId}" style="position:absolute;top:0;left:0;width:100%;height:100%;cursor:pointer;"></label>
-            <img src="${dataUri}" alt="Step ${ss.stepNumber}: ${escapeHtml(ss.action)}"
-              style="max-width:600px;max-height:90vh;border-radius:8px;position:relative;z-index:1001;" />
-          </div>
-          <div style="font-size:11px;color:#888;margin-top:4px;max-width:200px;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;">
-            ${ss.stepNumber}. ${escapeHtml(ss.action)}
-          </div>
-        </div>`;
-    } else {
-      html += `
-        <div style="flex:0 0 auto;width:200px;height:150px;background:#1a1a1a;border:1px dashed #333;border-radius:6px;display:flex;align-items:center;justify-content:center;color:#555;font-size:12px;">
-          Screenshot not found
-          <div style="font-size:11px;color:#888;margin-top:4px;">${ss.stepNumber}. ${escapeHtml(ss.action)}</div>
-        </div>`;
-    }
-  }
-  html += `</div>`;
-  return html;
-}
-function generateHtmlReport(runId) {
+init_database();
+import { readFileSync as readFileSync3, existsSync as existsSync6 } from "fs";
+import chalk4 from "chalk";
+var DEFAULT_THRESHOLD = 0.1;
+function setBaseline(runId) {
   const run = getRun(runId);
-  if (!run)
+  if (!run) {
     throw new Error(`Run not found: ${runId}`);
-  const results = getResultsByRun(run.id);
-  const resultData = [];
-  for (const result of results) {
-    const screenshots = listScreenshots(result.id);
-    const scenario = getScenario(result.scenarioId);
-    resultData.push({
-      result,
-      scenarioName: scenario?.name ?? "Unknown Scenario",
-      scenarioShortId: scenario?.shortId ?? result.scenarioId.slice(0, 8),
-      screenshots
-    });
   }
-  const passedCount = results.filter((r) => r.status === "passed").length;
-  const failedCount = results.filter((r) => r.status === "failed").length;
-  const errorCount = results.filter((r) => r.status === "error").length;
-  const totalCount = results.length;
-  const totalTokens = results.reduce((sum, r) => sum + r.tokensUsed, 0);
-  const totalCostCents = results.reduce((sum, r) => sum + r.costCents, 0);
-  const totalDurationMs = run.finishedAt && run.startedAt ? new Date(run.finishedAt).getTime() - new Date(run.startedAt).getTime() : results.reduce((sum, r) => sum + r.durationMs, 0);
-  const generatedAt = new Date().toISOString();
-  let resultCards = "";
-  for (const { result, scenarioName, scenarioShortId, screenshots } of resultData) {
-    resultCards += `
-      <div style="background:#141414;border:1px solid #262626;border-radius:8px;padding:20px;margin-bottom:16px;">
-        <div style="display:flex;align-items:center;gap:12px;margin-bottom:12px;">
-          ${statusBadge(result.status)}
-          <span style="font-size:16px;font-weight:600;color:#e5e5e5;">${escapeHtml(scenarioName)}</span>
-          <span style="font-size:12px;color:#666;font-family:monospace;">${escapeHtml(scenarioShortId)}</span>
-        </div>
-
-        ${result.reasoning ? `<div style="color:#a3a3a3;font-size:14px;line-height:1.6;margin-bottom:12px;padding:12px;background:#0d0d0d;border-radius:6px;border-left:3px solid #333;">${escapeHtml(result.reasoning)}</div>` : ""}
-
-        ${result.error ? `<div style="color:#ef4444;font-size:13px;margin-bottom:12px;padding:12px;background:#1a0a0a;border-radius:6px;border-left:3px solid #ef4444;font-family:monospace;">${escapeHtml(result.error)}</div>` : ""}
-
-        <div style="display:flex;gap:24px;font-size:13px;color:#888;">
-          <span>Duration: <span style="color:#d4d4d4;">${formatDuration2(result.durationMs)}</span></span>
-          <span>Steps: <span style="color:#d4d4d4;">${result.stepsCompleted}/${result.stepsTotal}</span></span>
-          <span>Tokens: <span style="color:#d4d4d4;">${result.tokensUsed.toLocaleString()}</span></span>
-          <span>Cost: <span style="color:#d4d4d4;">${formatCost(result.costCents)}</span></span>
-          <span>Model: <span style="color:#d4d4d4;">${escapeHtml(result.model)}</span></span>
-        </div>
-
-        ${renderScreenshots(screenshots)}
-      </div>`;
+  const db2 = getDatabase();
+  if (run.projectId) {
+    db2.query("UPDATE runs SET is_baseline = 0 WHERE project_id = ? AND is_baseline = 1").run(run.projectId);
+  } else {
+    db2.query("UPDATE runs SET is_baseline = 0 WHERE project_id IS NULL AND is_baseline = 1").run();
   }
-  return `<!DOCTYPE html>
-<html lang="en">
-<head>
-  <meta charset="UTF-8" />
-  <meta name="viewport" content="width=device-width, initial-scale=1.0" />
-  <title>Test Report - ${escapeHtml(run.id.slice(0, 8))}</title>
-  <style>
-    * { margin: 0; padding: 0; box-sizing: border-box; }
-    body { background: #0a0a0a; color: #e5e5e5; font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif; padding: 40px 20px; }
-    .container { max-width: 960px; margin: 0 auto; }
-    input[type="checkbox"]:checked ~ div:last-of-type { display: flex !important; }
-  </style>
-</head>
-<body>
-  <div class="container">
-    <!-- Header -->
-    <div style="margin-bottom:32px;">
-      <h1 style="font-size:28px;font-weight:700;margin-bottom:8px;color:#fff;">Test Report</h1>
-      <div style="display:flex;flex-wrap:wrap;gap:24px;font-size:14px;color:#888;">
-        <span>Run: <span style="color:#d4d4d4;font-family:monospace;">${escapeHtml(run.id.slice(0, 8))}</span></span>
-        <span>URL: <a href="${escapeHtml(run.url)}" style="color:#60a5fa;text-decoration:none;">${escapeHtml(run.url)}</a></span>
-        <span>Model: <span style="color:#d4d4d4;">${escapeHtml(run.model)}</span></span>
-        <span>Date: <span style="color:#d4d4d4;">${escapeHtml(run.startedAt)}</span></span>
-        <span>Duration: <span style="color:#d4d4d4;">${formatDuration2(totalDurationMs)}</span></span>
-        <span>Status: ${statusBadge(run.status)}</span>
-      </div>
-    </div>
-
-    <!-- Summary Bar -->
-    <div style="display:flex;gap:16px;margin-bottom:32px;">
-      <div style="flex:1;background:#141414;border:1px solid #262626;border-radius:8px;padding:16px;text-align:center;">
-        <div style="font-size:28px;font-weight:700;color:#e5e5e5;">${totalCount}</div>
-        <div style="font-size:12px;color:#888;margin-top:4px;">TOTAL</div>
-      </div>
-      <div style="flex:1;background:#141414;border:1px solid #262626;border-radius:8px;padding:16px;text-align:center;">
-        <div style="font-size:28px;font-weight:700;color:#22c55e;">${passedCount}</div>
-        <div style="font-size:12px;color:#888;margin-top:4px;">PASSED</div>
-      </div>
-      <div style="flex:1;background:#141414;border:1px solid #262626;border-radius:8px;padding:16px;text-align:center;">
-        <div style="font-size:28px;font-weight:700;color:#ef4444;">${failedCount}</div>
-        <div style="font-size:12px;color:#888;margin-top:4px;">FAILED</div>
-      </div>
-      ${errorCount > 0 ? `
-      <div style="flex:1;background:#141414;border:1px solid #262626;border-radius:8px;padding:16px;text-align:center;">
-        <div style="font-size:28px;font-weight:700;color:#eab308;">${errorCount}</div>
-        <div style="font-size:12px;color:#888;margin-top:4px;">ERRORS</div>
-      </div>` : ""}
-    </div>
-
-    <!-- Results -->
-    ${resultCards}
-
-    <!-- Footer -->
-    <div style="margin-top:32px;padding-top:20px;border-top:1px solid #262626;display:flex;justify-content:space-between;font-size:13px;color:#666;">
-      <div>
-        Total tokens: ${totalTokens.toLocaleString()} | Total cost: ${formatCost(totalCostCents)}
-      </div>
-      <div>
-        Generated: ${escapeHtml(generatedAt)}
-      </div>
-    </div>
-  </div>
-</body>
-</html>`;
-}
-function generateLatestReport() {
-  const runs = listRuns({ limit: 1 });
-  if (runs.length === 0)
-    throw new Error("No runs found");
-  return generateHtmlReport(runs[0].id);
+  updateRun(run.id, { is_baseline: 1 });
 }
-
-// src/lib/costs.ts
-init_database();
-init_config();
-import chalk4 from "chalk";
-function getDateFilter(period) {
-  switch (period) {
-    case "day":
-      return "AND r.created_at >= date('now', 'start of day')";
-    case "week":
-      return "AND r.created_at >= date('now', '-7 days')";
-    case "month":
-      return "AND r.created_at >= date('now', '-30 days')";
-    case "all":
-      return "";
+function compareImages(image1Path, image2Path) {
+  if (!existsSync6(image1Path)) {
+    throw new Error(`Baseline image not found: ${image1Path}`);
   }
-}
-function getPeriodDays(period) {
-  switch (period) {
-    case "day":
-      return 1;
-    case "week":
-      return 7;
-    case "month":
-      return 30;
-    case "all":
-      return 30;
+  if (!existsSync6(image2Path)) {
+    throw new Error(`Current image not found: ${image2Path}`);
+  }
+  const buf1 = readFileSync3(image1Path);
+  const buf2 = readFileSync3(image2Path);
+  if (buf1.equals(buf2)) {
+    const estimatedPixels = Math.max(1, Math.floor(buf1.length / 4));
+    return { diffPercent: 0, diffPixels: 0, totalPixels: estimatedPixels };
+  }
+  if (buf1.length !== buf2.length) {
+    const maxLen = Math.max(buf1.length, buf2.length);
+    const estimatedPixels = Math.max(1, Math.floor(maxLen / 4));
+    return { diffPercent: 100, diffPixels: estimatedPixels, totalPixels: estimatedPixels };
   }
-}
-function loadBudgetConfig() {
-  const config = loadConfig();
-  const budget = config.budget;
-  return {
-    maxPerRunCents: budget?.maxPerRunCents ?? 50,
-    maxPerDayCents: budget?.maxPerDayCents ?? 500,
-    warnAtPercent: budget?.warnAtPercent ?? 0.8
-  };
-}
-function getCostSummary(options) {
-  const db2 = getDatabase();
-  const period = options?.period ?? "month";
-  const projectId = options?.projectId;
-  const dateFilter = getDateFilter(period);
-  const projectFilter = projectId ? "AND ru.project_id = ?" : "";
-  const projectParams = projectId ? [projectId] : [];
-  const totalsRow = db2.query(`SELECT
-        COALESCE(SUM(r.cost_cents), 0) as total_cost,
-        COALESCE(SUM(r.tokens_used), 0) as total_tokens,
-        COUNT(DISTINCT r.run_id) as run_count
-      FROM results r
-      JOIN runs ru ON r.run_id = ru.id
-      WHERE 1=1 ${dateFilter} ${projectFilter}`).get(...projectParams);
-  const modelRows = db2.query(`SELECT
-        r.model,
-        COALESCE(SUM(r.cost_cents), 0) as cost_cents,
-        COALESCE(SUM(r.tokens_used), 0) as tokens,
-        COUNT(DISTINCT r.run_id) as runs
-      FROM results r
-      JOIN runs ru ON r.run_id = ru.id
-      WHERE 1=1 ${dateFilter} ${projectFilter}
-      GROUP BY r.model
-      ORDER BY cost_cents DESC`).all(...projectParams);
-  const byModel = {};
-  for (const row of modelRows) {
-    byModel[row.model] = {
-      costCents: row.cost_cents,
-      tokens: row.tokens,
-      runs: row.runs
-    };
+  let diffBytes = 0;
+  for (let i = 0;i < buf1.length; i++) {
+    if (buf1[i] !== buf2[i]) {
+      diffBytes++;
+    }
   }
-  const scenarioRows = db2.query(`SELECT
-        r.scenario_id,
-        COALESCE(s.name, r.scenario_id) as name,
-        COALESCE(SUM(r.cost_cents), 0) as cost_cents,
-        COALESCE(SUM(r.tokens_used), 0) as tokens,
-        COUNT(DISTINCT r.run_id) as runs
-      FROM results r
-      JOIN runs ru ON r.run_id = ru.id
-      LEFT JOIN scenarios s ON r.scenario_id = s.id
-      WHERE 1=1 ${dateFilter} ${projectFilter}
-      GROUP BY r.scenario_id
-      ORDER BY cost_cents DESC
-      LIMIT 10`).all(...projectParams);
-  const byScenario = scenarioRows.map((row) => ({
-    scenarioId: row.scenario_id,
-    name: row.name,
-    costCents: row.cost_cents,
-    tokens: row.tokens,
-    runs: row.runs
-  }));
-  const runCount = totalsRow.run_count;
-  const avgCostPerRun = runCount > 0 ? totalsRow.total_cost / runCount : 0;
-  const periodDays = getPeriodDays(period);
-  const estimatedMonthlyCents = periodDays > 0 ? totalsRow.total_cost / periodDays * 30 : 0;
-  return {
-    period,
-    totalCostCents: totalsRow.total_cost,
-    totalTokens: totalsRow.total_tokens,
-    runCount,
-    byModel,
-    byScenario,
-    avgCostPerRun,
-    estimatedMonthlyCents
-  };
+  const totalPixels = Math.max(1, Math.floor(buf1.length / 4));
+  const diffPixels = Math.max(1, Math.floor(diffBytes / 4));
+  const diffPercent = parseFloat((diffBytes / buf1.length * 100).toFixed(4));
+  return { diffPercent, diffPixels, totalPixels };
 }
-function getCostsByScenario(options) {
-  const db2 = getDatabase();
-  const period = options?.period ?? "month";
-  const projectId = options?.projectId;
-  const dateFilter = getDateFilter(period);
-  const projectFilter = projectId ? "AND ru.project_id = ?" : "";
-  const projectParams = projectId ? [projectId] : [];
-  const rows = db2.query(`SELECT
-        r.scenario_id,
-        COALESCE(s.name, r.scenario_id) as name,
-        COUNT(DISTINCT r.run_id) as run_count,
-        COALESCE(SUM(r.cost_cents), 0) as total_cost_cents
-      FROM results r
-      JOIN runs ru ON r.run_id = ru.id
-      LEFT JOIN scenarios s ON r.scenario_id = s.id
-      WHERE 1=1 ${dateFilter} ${projectFilter}
-      GROUP BY r.scenario_id
-      ORDER BY total_cost_cents DESC`).all(...projectParams);
-  return rows.map((row) => ({
-    scenarioId: row.scenario_id,
-    name: row.name,
-    runCount: row.run_count,
-    totalCostCents: row.total_cost_cents,
-    avgCostPerRunCents: row.run_count > 0 ? row.total_cost_cents / row.run_count : 0
-  }));
+function compareRunScreenshots(runId, baselineRunId, threshold = DEFAULT_THRESHOLD) {
+  const run = getRun(runId);
+  if (!run)
+    throw new Error(`Run not found: ${runId}`);
+  const baselineRun = getRun(baselineRunId);
+  if (!baselineRun)
+    throw new Error(`Baseline run not found: ${baselineRunId}`);
+  const currentResults = getResultsByRun(run.id);
+  const baselineResults = getResultsByRun(baselineRun.id);
+  const baselineMap = new Map;
+  for (const result of baselineResults) {
+    const screenshots = listScreenshots(result.id);
+    for (const ss of screenshots) {
+      const key = `${result.scenarioId}:${ss.stepNumber}`;
+      baselineMap.set(key, { path: ss.filePath, action: ss.action });
+    }
+  }
+  const results = [];
+  for (const result of currentResults) {
+    const screenshots = listScreenshots(result.id);
+    for (const ss of screenshots) {
+      const key = `${result.scenarioId}:${ss.stepNumber}`;
+      const baseline = baselineMap.get(key);
+      if (!baseline)
+        continue;
+      if (!existsSync6(baseline.path) || !existsSync6(ss.filePath))
+        continue;
+      try {
+        const comparison = compareImages(baseline.path, ss.filePath);
+        results.push({
+          scenarioId: result.scenarioId,
+          stepNumber: ss.stepNumber,
+          action: ss.action,
+          baselinePath: baseline.path,
+          currentPath: ss.filePath,
+          diffPercent: comparison.diffPercent,
+          isRegression: comparison.diffPercent > threshold
+        });
+      } catch {}
+    }
+  }
+  return results;
 }
-function formatCostsByScenarioTerminal(rows, period) {
+function formatVisualDiffTerminal(results, threshold = DEFAULT_THRESHOLD) {
+  if (results.length === 0) {
+    return chalk4.dim(`
+  No screenshot comparisons found.
+`);
+  }
   const lines = [];
   lines.push("");
-  lines.push(chalk4.bold(`  Cost by Scenario (${period})`));
+  lines.push(chalk4.bold("  Visual Regression Summary"));
   lines.push("");
-  if (rows.length === 0) {
-    lines.push(chalk4.dim("  No cost data found."));
+  const regressions = results.filter((r) => r.diffPercent >= threshold);
+  const passed = results.filter((r) => r.diffPercent < threshold);
+  if (regressions.length > 0) {
+    lines.push(chalk4.red.bold(`  Regressions (${regressions.length}):`));
+    for (const r of regressions) {
+      const scenario = getScenario(r.scenarioId);
+      const label = scenario ? `${scenario.shortId}: ${scenario.name}` : r.scenarioId.slice(0, 8);
+      const pct = chalk4.red(`${r.diffPercent.toFixed(2)}%`);
+      lines.push(`    ${chalk4.red("!")} ${label} step ${r.stepNumber} (${r.action}) \u2014 ${pct} diff`);
+    }
     lines.push("");
-    return lines.join(`
-`);
   }
-  lines.push(`  ${"Scenario".padEnd(40)} ${"Runs".padEnd(8)} ${"Total Cost".padEnd(14)} Avg/Run`);
-  lines.push(`  ${"\u2500".repeat(40)} ${"\u2500".repeat(8)} ${"\u2500".repeat(14)} ${"\u2500".repeat(10)}`);
-  for (const row of rows) {
-    const label = row.name.length > 38 ? row.name.slice(0, 35) + "..." : row.name;
-    lines.push(`  ${label.padEnd(40)} ${String(row.runCount).padEnd(8)} ${formatDollars(row.totalCostCents).padEnd(14)} ${formatDollars(row.avgCostPerRunCents)}`);
+  if (passed.length > 0) {
+    lines.push(chalk4.green.bold(`  Passed (${passed.length}):`));
+    for (const r of passed) {
+      const scenario = getScenario(r.scenarioId);
+      const label = scenario ? `${scenario.shortId}: ${scenario.name}` : r.scenarioId.slice(0, 8);
+      const pct = chalk4.green(`${r.diffPercent.toFixed(2)}%`);
+      lines.push(`    ${chalk4.green("\u2713")} ${label} step ${r.stepNumber} (${r.action}) \u2014 ${pct} diff`);
+    }
+    lines.push("");
   }
+  lines.push(chalk4.bold(`  Visual Summary: ${regressions.length} regressions, ${passed.length} passed (threshold: ${threshold}%)`));
   lines.push("");
   return lines.join(`
 `);
 }
-function checkBudget(estimatedCostCents) {
-  const budget = loadBudgetConfig();
-  if (estimatedCostCents > budget.maxPerRunCents) {
-    return {
-      allowed: false,
-      warning: `Estimated cost (${formatDollars(estimatedCostCents)}) exceeds per-run limit (${formatDollars(budget.maxPerRunCents)})`
-    };
-  }
-  const todaySummary = getCostSummary({ period: "day" });
-  const projectedDaily = todaySummary.totalCostCents + estimatedCostCents;
-  if (projectedDaily > budget.maxPerDayCents) {
-    return {
-      allowed: false,
-      warning: `Daily spending (${formatDollars(todaySummary.totalCostCents)}) + this run (${formatDollars(estimatedCostCents)}) would exceed daily limit (${formatDollars(budget.maxPerDayCents)})`
-    };
+
+// src/lib/report.ts
+init_runs();
+init_results();
+init_screenshots();
+init_scenarios();
+import { readFileSync as readFileSync4, existsSync as existsSync7 } from "fs";
+function imageToBase64(filePath) {
+  if (!filePath || !existsSync7(filePath))
+    return "";
+  try {
+    const buffer = readFileSync4(filePath);
+    const base64 = buffer.toString("base64");
+    return `data:image/png;base64,${base64}`;
+  } catch {
+    return "";
   }
-  if (projectedDaily > budget.maxPerDayCents * budget.warnAtPercent) {
-    return {
-      allowed: true,
-      warning: `Approaching daily limit: ${formatDollars(projectedDaily)} of ${formatDollars(budget.maxPerDayCents)} (${Math.round(projectedDaily / budget.maxPerDayCents * 100)}%)`
-    };
+}
+function escapeHtml(text) {
+  return text.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;").replace(/"/g, "&quot;").replace(/'/g, "&#039;");
+}
+function formatDuration2(ms) {
+  if (ms < 1000)
+    return `${ms}ms`;
+  if (ms < 60000)
+    return `${(ms / 1000).toFixed(1)}s`;
+  const mins = Math.floor(ms / 60000);
+  const secs = (ms % 60000 / 1000).toFixed(0);
+  return `${mins}m ${secs}s`;
+}
+function formatCost(cents) {
+  if (cents < 1)
+    return `$${(cents / 100).toFixed(4)}`;
+  return `$${(cents / 100).toFixed(2)}`;
+}
+function statusBadge(status) {
+  const colors = {
+    passed: { bg: "#22c55e", text: "#000" },
+    failed: { bg: "#ef4444", text: "#fff" },
+    error: { bg: "#eab308", text: "#000" },
+    skipped: { bg: "#6b7280", text: "#fff" }
+  };
+  const c = colors[status] ?? { bg: "#6b7280", text: "#fff" };
+  const label = status.toUpperCase();
+  return `<span style="display:inline-block;padding:2px 10px;border-radius:4px;font-size:12px;font-weight:700;background:${c.bg};color:${c.text};letter-spacing:0.5px;">${label}</span>`;
+}
+function renderScreenshots(screenshots) {
+  if (screenshots.length === 0)
+    return "";
+  let html = `<div style="display:flex;flex-wrap:wrap;gap:12px;margin-top:12px;">`;
+  for (let i = 0;i < screenshots.length; i++) {
+    const ss = screenshots[i];
+    const dataUri = imageToBase64(ss.filePath);
+    const checkId = `ss-${ss.id}`;
+    if (dataUri) {
+      html += `
+        <div style="flex:0 0 auto;">
+          <input type="checkbox" id="${checkId}" style="display:none;" />
+          <label for="${checkId}" style="cursor:pointer;">
+            <img src="${dataUri}" alt="Step ${ss.stepNumber}: ${escapeHtml(ss.action)}"
+              style="max-width:200px;max-height:150px;border-radius:6px;border:1px solid #262626;display:block;" />
+          </label>
+          <div style="position:fixed;top:0;left:0;width:100%;height:100%;background:rgba(0,0,0,0.9);z-index:1000;display:none;align-items:center;justify-content:center;">
+            <label for="${checkId}" style="position:absolute;top:0;left:0;width:100%;height:100%;cursor:pointer;"></label>
+            <img src="${dataUri}" alt="Step ${ss.stepNumber}: ${escapeHtml(ss.action)}"
+              style="max-width:600px;max-height:90vh;border-radius:8px;position:relative;z-index:1001;" />
+          </div>
+          <div style="font-size:11px;color:#888;margin-top:4px;max-width:200px;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;">
+            ${ss.stepNumber}. ${escapeHtml(ss.action)}
+          </div>
+        </div>`;
+    } else {
+      html += `
+        <div style="flex:0 0 auto;width:200px;height:150px;background:#1a1a1a;border:1px dashed #333;border-radius:6px;display:flex;align-items:center;justify-content:center;color:#555;font-size:12px;">
+          Screenshot not found
+          <div style="font-size:11px;color:#888;margin-top:4px;">${ss.stepNumber}. ${escapeHtml(ss.action)}</div>
+        </div>`;
+    }
   }
-  return { allowed: true };
-}
-function formatDollars(cents) {
-  return `$${(cents / 100).toFixed(2)}`;
-}
-function formatTokens(tokens) {
-  if (tokens >= 1e6)
-    return `${(tokens / 1e6).toFixed(1)}M`;
-  if (tokens >= 1000)
-    return `${(tokens / 1000).toFixed(1)}K`;
-  return String(tokens);
+  html += `</div>`;
+  return html;
 }
-function formatCostsTerminal(summary) {
-  const lines = [];
-  lines.push("");
-  lines.push(chalk4.bold(`  Cost Summary (${summary.period})`));
-  lines.push("");
-  lines.push(`  Total:     ${chalk4.yellow(formatDollars(summary.totalCostCents))} (${formatTokens(summary.totalTokens)} tokens across ${summary.runCount} runs)`);
-  lines.push(`  Avg/run:   ${chalk4.yellow(formatDollars(summary.avgCostPerRun))}`);
-  lines.push(`  Est/month: ${chalk4.yellow(formatDollars(summary.estimatedMonthlyCents))}`);
-  const modelEntries = Object.entries(summary.byModel);
-  if (modelEntries.length > 0) {
-    lines.push("");
-    lines.push(chalk4.bold("  By Model"));
-    lines.push(`  ${"Model".padEnd(40)} ${"Cost".padEnd(12)} ${"Tokens".padEnd(12)} Runs`);
-    lines.push(`  ${"\u2500".repeat(40)} ${"\u2500".repeat(12)} ${"\u2500".repeat(12)} ${"\u2500".repeat(6)}`);
-    for (const [model, data] of modelEntries) {
-      lines.push(`  ${model.padEnd(40)} ${formatDollars(data.costCents).padEnd(12)} ${formatTokens(data.tokens).padEnd(12)} ${data.runs}`);
-    }
+function generateHtmlReport(runId) {
+  const run = getRun(runId);
+  if (!run)
+    throw new Error(`Run not found: ${runId}`);
+  const results = getResultsByRun(run.id);
+  const resultData = [];
+  for (const result of results) {
+    const screenshots = listScreenshots(result.id);
+    const scenario = getScenario(result.scenarioId);
+    resultData.push({
+      result,
+      scenarioName: scenario?.name ?? "Unknown Scenario",
+      scenarioShortId: scenario?.shortId ?? result.scenarioId.slice(0, 8),
+      screenshots
+    });
   }
-  if (summary.byScenario.length > 0) {
-    lines.push("");
-    lines.push(chalk4.bold("  Scenarios by Cost (most expensive first)"));
-    lines.push(`  ${"Scenario".padEnd(40)} ${"Total Cost".padEnd(12)} ${"Avg/Run".padEnd(12)} ${"Runs".padEnd(6)} Tokens`);
-    lines.push(`  ${"\u2500".repeat(40)} ${"\u2500".repeat(12)} ${"\u2500".repeat(12)} ${"\u2500".repeat(6)} ${"\u2500".repeat(10)}`);
-    for (const s of summary.byScenario) {
-      const label = s.name.length > 38 ? s.name.slice(0, 35) + "..." : s.name;
-      const avgPerRun = s.runs > 0 ? s.costCents / s.runs : 0;
-      lines.push(`  ${label.padEnd(40)} ${formatDollars(s.costCents).padEnd(12)} ${formatDollars(avgPerRun).padEnd(12)} ${String(s.runs).padEnd(6)} ${formatTokens(s.tokens)}`);
-    }
+  const passedCount = results.filter((r) => r.status === "passed").length;
+  const failedCount = results.filter((r) => r.status === "failed").length;
+  const errorCount = results.filter((r) => r.status === "error").length;
+  const totalCount = results.length;
+  const totalTokens = results.reduce((sum, r) => sum + r.tokensUsed, 0);
+  const totalCostCents = results.reduce((sum, r) => sum + r.costCents, 0);
+  const totalDurationMs = run.finishedAt && run.startedAt ? new Date(run.finishedAt).getTime() - new Date(run.startedAt).getTime() : results.reduce((sum, r) => sum + r.durationMs, 0);
+  const generatedAt = new Date().toISOString();
+  let resultCards = "";
+  for (const { result, scenarioName, scenarioShortId, screenshots } of resultData) {
+    resultCards += `
+      <div style="background:#141414;border:1px solid #262626;border-radius:8px;padding:20px;margin-bottom:16px;">
+        <div style="display:flex;align-items:center;gap:12px;margin-bottom:12px;">
+          ${statusBadge(result.status)}
+          <span style="font-size:16px;font-weight:600;color:#e5e5e5;">${escapeHtml(scenarioName)}</span>
+          <span style="font-size:12px;color:#666;font-family:monospace;">${escapeHtml(scenarioShortId)}</span>
+        </div>
+
+        ${result.reasoning ? `<div style="color:#a3a3a3;font-size:14px;line-height:1.6;margin-bottom:12px;padding:12px;background:#0d0d0d;border-radius:6px;border-left:3px solid #333;">${escapeHtml(result.reasoning)}</div>` : ""}
+
+        ${result.error ? `<div style="color:#ef4444;font-size:13px;margin-bottom:12px;padding:12px;background:#1a0a0a;border-radius:6px;border-left:3px solid #ef4444;font-family:monospace;">${escapeHtml(result.error)}</div>` : ""}
+
+        <div style="display:flex;gap:24px;font-size:13px;color:#888;">
+          <span>Duration: <span style="color:#d4d4d4;">${formatDuration2(result.durationMs)}</span></span>
+          <span>Steps: <span style="color:#d4d4d4;">${result.stepsCompleted}/${result.stepsTotal}</span></span>
+          <span>Tokens: <span style="color:#d4d4d4;">${result.tokensUsed.toLocaleString()}</span></span>
+          <span>Cost: <span style="color:#d4d4d4;">${formatCost(result.costCents)}</span></span>
+          <span>Model: <span style="color:#d4d4d4;">${escapeHtml(result.model)}</span></span>
+        </div>
+
+        ${renderScreenshots(screenshots)}
+      </div>`;
   }
-  lines.push("");
-  return lines.join(`
-`);
-}
-function formatCostsJSON(summary) {
-  return JSON.stringify(summary, null, 2);
+  return `<!DOCTYPE html>
+<html lang="en">
+<head>
+  <meta charset="UTF-8" />
+  <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+  <title>Test Report - ${escapeHtml(run.id.slice(0, 8))}</title>
+  <style>
+    * { margin: 0; padding: 0; box-sizing: border-box; }
+    body { background: #0a0a0a; color: #e5e5e5; font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif; padding: 40px 20px; }
+    .container { max-width: 960px; margin: 0 auto; }
+    input[type="checkbox"]:checked ~ div:last-of-type { display: flex !important; }
+  </style>
+</head>
+<body>
+  <div class="container">
+    <!-- Header -->
+    <div style="margin-bottom:32px;">
+      <h1 style="font-size:28px;font-weight:700;margin-bottom:8px;color:#fff;">Test Report</h1>
+      <div style="display:flex;flex-wrap:wrap;gap:24px;font-size:14px;color:#888;">
+        <span>Run: <span style="color:#d4d4d4;font-family:monospace;">${escapeHtml(run.id.slice(0, 8))}</span></span>
+        <span>URL: <a href="${escapeHtml(run.url)}" style="color:#60a5fa;text-decoration:none;">${escapeHtml(run.url)}</a></span>
+        <span>Model: <span style="color:#d4d4d4;">${escapeHtml(run.model)}</span></span>
+        <span>Date: <span style="color:#d4d4d4;">${escapeHtml(run.startedAt)}</span></span>
+        <span>Duration: <span style="color:#d4d4d4;">${formatDuration2(totalDurationMs)}</span></span>
+        <span>Status: ${statusBadge(run.status)}</span>
+      </div>
+    </div>
+
+    <!-- Summary Bar -->
+    <div style="display:flex;gap:16px;margin-bottom:32px;">
+      <div style="flex:1;background:#141414;border:1px solid #262626;border-radius:8px;padding:16px;text-align:center;">
+        <div style="font-size:28px;font-weight:700;color:#e5e5e5;">${totalCount}</div>
+        <div style="font-size:12px;color:#888;margin-top:4px;">TOTAL</div>
+      </div>
+      <div style="flex:1;background:#141414;border:1px solid #262626;border-radius:8px;padding:16px;text-align:center;">
+        <div style="font-size:28px;font-weight:700;color:#22c55e;">${passedCount}</div>
+        <div style="font-size:12px;color:#888;margin-top:4px;">PASSED</div>
+      </div>
+      <div style="flex:1;background:#141414;border:1px solid #262626;border-radius:8px;padding:16px;text-align:center;">
+        <div style="font-size:28px;font-weight:700;color:#ef4444;">${failedCount}</div>
+        <div style="font-size:12px;color:#888;margin-top:4px;">FAILED</div>
+      </div>
+      ${errorCount > 0 ? `
+      <div style="flex:1;background:#141414;border:1px solid #262626;border-radius:8px;padding:16px;text-align:center;">
+        <div style="font-size:28px;font-weight:700;color:#eab308;">${errorCount}</div>
+        <div style="font-size:12px;color:#888;margin-top:4px;">ERRORS</div>
+      </div>` : ""}
+    </div>
+
+    <!-- Results -->
+    ${resultCards}
+
+    <!-- Footer -->
+    <div style="margin-top:32px;padding-top:20px;border-top:1px solid #262626;display:flex;justify-content:space-between;font-size:13px;color:#666;">
+      <div>
+        Total tokens: ${totalTokens.toLocaleString()} | Total cost: ${formatCost(totalCostCents)}
+      </div>
+      <div>
+        Generated: ${escapeHtml(generatedAt)}
+      </div>
+    </div>
+  </div>
+</body>
+</html>`;
 }
-function formatCostsCsv(summary) {
-  const lines = [];
-  lines.push("scenario,runs,total_cost_cents,avg_cost_cents,tokens");
-  for (const s of summary.byScenario) {
-    const avgCostCents = s.runs > 0 ? s.costCents / s.runs : 0;
-    const name = s.name.includes(",") ? `"${s.name.replace(/"/g, '""')}"` : s.name;
-    lines.push(`${name},${s.runs},${s.costCents},${avgCostCents.toFixed(2)},${s.tokens}`);
-  }
-  return lines.join(`
-`);
+function generateLatestReport() {
+  const runs = listRuns({ limit: 1 });
+  if (runs.length === 0)
+    throw new Error("No runs found");
+  return generateHtmlReport(runs[0].id);
 }
 
 // src/cli/index.tsx
+init_costs();
 init_personas();
 init_api_checks();
 
@@ -11436,7 +12495,7 @@ function parseAssertionString(str) {
 }
 
 // src/cli/index.tsx
-import { existsSync as existsSync8, mkdirSync as mkdirSync4 } from "fs";
+import { existsSync as existsSync8, mkdirSync as mkdirSync5 } from "fs";
 import { jsxDEV } from "react/jsx-dev-runtime";
 var PRIORITIES = ["low", "medium", "high", "critical"];
 function AddForm({ onComplete }) {
@@ -11695,8 +12754,8 @@ function logError(...args) {
   console.error(...args);
 }
 program2.name("testers").version(package_default.version).description("AI-powered browser testing CLI").option("-q, --quiet", "Suppress all output", false).option("--no-color", "Disable color output");
-var CONFIG_DIR2 = join6(process.env["HOME"] ?? "~", ".testers");
-var CONFIG_PATH2 = join6(CONFIG_DIR2, "config.json");
+var CONFIG_DIR2 = join7(process.env["HOME"] ?? "~", ".testers");
+var CONFIG_PATH2 = join7(CONFIG_DIR2, "config.json");
 function getActiveProject() {
   try {
     if (existsSync8(CONFIG_PATH2)) {
@@ -11964,7 +13023,7 @@ program2.command("remove <id>").alias("uninstall").description("Remove a scenari
 program2.command("run [url] [description]").alias("test").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").option("-b, --background", "Start run in background and return immediately", false).option("--browser <engine>", "Browser engine: playwright or lightpanda", "playwright").option("--env <name>", "Use a named environment for the URL").option("--dry-run", "Print what would run without launching browser", false).option("--retry <n>", "Retry failed scenarios up to n times", "0").option("--samples <n>", "Run each scenario N times and report flakiness (pass rate)", "1").option("--flakiness-threshold <n>", "Pass rate threshold below which a scenario is marked flaky (0-1)", "0.95").option("--a11y [level]", "Run axe-core WCAG accessibility scan after each navigation (level: A, AA, AAA \u2014 default AA)").option("--self-heal", "Enable AI-powered selector repair when elements can't be found (requires judgeModel or ANTHROPIC_API_KEY)", false).option("--verbose", "Show per-step timing and full tool results", false).option("--watch-results", "When used with --background, poll and display live results table until run completes", false).option("--failed-only", "Only show failed/error scenarios in output (passed count shown as summary)", false).option("--smoke", "Run only smoke-tagged scenarios (fast validation suite, <2 min)", false).option("--github-comment", "Post pass/fail summary as a GitHub PR comment (requires GITHUB_TOKEN env var)", false).option("--pr <number>", "GitHub PR number (auto-detected from GITHUB_REF if not provided)").option("--persona <id>", "Override persona for this run (comma-separated IDs for divergence testing)").action(async (urlArg, 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).option("--browser <engine>", "Browser engine: playwright (default), lightpanda (9x faster, no screenshots), or bun (native WKWebView, 11x faster, Bun canary required)", "playwright").option("--env <name>", "Use a named environment for the URL").option("--dry-run", "Print what would run without launching browser", false).option("--retry <n>", "Retry failed scenarios up to n times", "0").option("--samples <n>", "Run each scenario N times and report flakiness (pass rate)", "1").option("--flakiness-threshold <n>", "Pass rate threshold below which a scenario is marked flaky (0-1)", "0.95").option("--a11y [level]", "Run axe-core WCAG accessibility scan after each navigation (level: A, AA, AAA \u2014 default AA)").option("--self-heal", "Enable AI-powered selector repair when elements can't be found (requires judgeModel or ANTHROPIC_API_KEY)", false).option("--verbose", "Show per-step timing and full tool results", false).option("--watch-results", "When used with --background, poll and display live results table until run completes", false).option("--failed-only", "Only show failed/error scenarios in output (passed count shown as summary)", false).option("--smoke", "Run only smoke-tagged scenarios (fast validation suite, <2 min)", false).option("--minimal", "Fastest possible run: cheapest model, max parallelism, min turns (ideal for CI)", false).option("--github-comment", "Post pass/fail summary as a GitHub PR comment (requires GITHUB_TOKEN env var)", false).option("--pr <number>", "GitHub PR number (auto-detected from GITHUB_REF if not provided)").option("--persona <id>", "Override persona for this run (comma-separated IDs for divergence testing)").option("--max-cost <dollars>", "Hard budget cap in dollars \u2014 abort if estimated cost exceeds this (e.g. 0.50 for 50 cents)").option("--cache-max-age <seconds>", "Skip scenarios that passed at the same URL within this many seconds (0 = disabled)", "0").option("--diff", "Auto-detect changed files from git diff and run only relevant scenarios", false).action(async (urlArg, description, opts) => {
   try {
     const projectId = resolveProject(opts.project);
     let url = urlArg;
@@ -12224,11 +13283,38 @@ program2.command("run [url] [description]").alias("test").description("Run test
       log(chalk6.bold(`  Running all ${allScenarios.length} scenarios...`));
       log("");
     }
+    let diffScenarioIds;
+    if (opts.diff) {
+      try {
+        const { execSync: execSync3 } = await import("child_process");
+        const staged = execSync3("git diff --cached --name-only", { cwd: process.cwd(), encoding: "utf-8" }).trim();
+        const unstaged = execSync3("git diff --name-only HEAD", { cwd: process.cwd(), encoding: "utf-8" }).trim();
+        const diffOutput = [staged, unstaged].filter(Boolean).join(`
+`);
+        if (!diffOutput.trim()) {
+          log(chalk6.yellow("  --diff: No changed files detected. Running all scenarios."));
+        } else {
+          const filePaths = [...new Set(diffOutput.split(`
+`).filter(Boolean))];
+          const { matchFilesToScenarios: matchFilesToScenarios2 } = await Promise.resolve().then(() => exports_affected);
+          const allScenarios = listScenarios({ projectId });
+          const matched = matchFilesToScenarios2(filePaths, allScenarios, []);
+          if (matched.length === 0) {
+            log(chalk6.yellow(`  --diff: No scenarios match changed files (${filePaths.length} files changed). Exiting.`));
+            process.exit(0);
+          }
+          diffScenarioIds = matched.map((s) => s.id);
+          log(chalk6.dim(`  --diff: ${filePaths.length} files changed \u2192 ${matched.length} matching scenario(s)`));
+        }
+      } catch {
+        log(chalk6.yellow("  --diff: git diff failed. Running all scenarios."));
+      }
+    }
     const personaIdList = opts.persona ? opts.persona.split(",").map((s) => s.trim()).filter(Boolean) : undefined;
     const { run, results } = await runByFilter({
       url,
       tags: opts.tag.length > 0 ? opts.tag : undefined,
-      scenarioIds: opts.scenario ? [opts.scenario] : undefined,
+      scenarioIds: diffScenarioIds ?? (opts.scenario ? [opts.scenario] : undefined),
       priority: opts.priority,
       model: opts.model,
       headed: opts.headed,
@@ -12242,7 +13328,10 @@ program2.command("run [url] [description]").alias("test").description("Run test
       a11y: opts.a11y ? typeof opts.a11y === "string" ? { level: opts.a11y } : true : undefined,
       selfHeal: opts.selfHeal || undefined,
       personaId: personaIdList?.[0],
-      personaIds: personaIdList && personaIdList.length > 1 ? personaIdList : undefined
+      personaIds: personaIdList && personaIdList.length > 1 ? personaIdList : undefined,
+      maxCostCents: opts.maxCost ? Math.round(parseFloat(opts.maxCost) * 100) : undefined,
+      cacheMaxAgeMs: opts.cacheMaxAge ? parseInt(opts.cacheMaxAge, 10) * 1000 : undefined,
+      minimal: opts.minimal || undefined
     });
     if (opts.json || opts.output) {
       const jsonOutput = formatJSON(run, results);
@@ -12355,7 +13444,7 @@ program2.command("import <dir>").description("Import markdown test files as scen
     }
     let imported = 0;
     for (const file of files) {
-      const content = readFileSync7(join6(absDir, file), "utf-8");
+      const content = readFileSync7(join7(absDir, file), "utf-8");
       const lines = content.split(`
 `);
       let name = file.replace(/\.md$/, "");
@@ -12416,7 +13505,7 @@ program2.command("export [format]").description("Export scenarios as JSON (defau
     }
     const outputDir = opts.output ?? ".";
     if (!existsSync8(outputDir)) {
-      mkdirSync4(outputDir, { recursive: true });
+      mkdirSync5(outputDir, { recursive: true });
     }
     for (const s of scenarios) {
       const lines = [];
@@ -12443,7 +13532,7 @@ program2.command("export [format]").description("Export scenarios as JSON (defau
         lines.push("");
       }
       const safeFilename = s.name.toLowerCase().replace(/[^a-z0-9]+/g, "-").replace(/^-|-$/g, "").slice(0, 80);
-      const filePath = join6(outputDir, `${s.shortId}-${safeFilename}.md`);
+      const filePath = join7(outputDir, `${s.shortId}-${safeFilename}.md`);
       writeFileSync3(filePath, lines.join(`
 `), "utf-8");
       log(chalk6.dim(`  ${s.shortId}: ${s.name} \u2192 ${filePath}`));
@@ -12468,7 +13557,7 @@ program2.command("status").description("Show database and auth status").action((
   try {
     const config = loadConfig();
     const hasApiKey = !!config.anthropicApiKey || !!process.env["ANTHROPIC_API_KEY"];
-    const dbPath = join6(process.env["HOME"] ?? "~", ".testers", "testers.db");
+    const dbPath = join7(process.env["HOME"] ?? "~", ".testers", "testers.db");
     log("");
     log(chalk6.bold("  Open Testers Status"));
     log("");
@@ -12558,7 +13647,7 @@ projectCmd.command("use <name>").description("Set active project (find or create
   try {
     const project = ensureProject(name, process.cwd());
     if (!existsSync8(CONFIG_DIR2)) {
-      mkdirSync4(CONFIG_DIR2, { recursive: true });
+      mkdirSync5(CONFIG_DIR2, { recursive: true });
     }
     let config = {};
     if (existsSync8(CONFIG_PATH2)) {
@@ -12812,11 +13901,11 @@ program2.command("init").description("Initialize a new testing project").option(
       log(`    ${chalk6.dim(s.shortId)} ${s.name} ${chalk6.dim(`[${s.tags.join(", ")}]`)}`);
     }
     if (opts.ci === "github") {
-      const workflowDir = join6(process.cwd(), ".github", "workflows");
+      const workflowDir = join7(process.cwd(), ".github", "workflows");
       if (!existsSync8(workflowDir)) {
-        mkdirSync4(workflowDir, { recursive: true });
+        mkdirSync5(workflowDir, { recursive: true });
       }
-      const workflowPath = join6(workflowDir, "testers.yml");
+      const workflowPath = join7(workflowDir, "testers.yml");
       writeFileSync3(workflowPath, generateGitHubActionsWorkflow(), "utf-8");
       log(`  CI:         ${chalk6.green("GitHub Actions workflow written to .github/workflows/testers.yml")}`);
     } else if (opts.ci) {
@@ -13816,7 +14905,7 @@ program2.command("doctor").description("Check system setup and configuration").a
     log(chalk6.red("\u2717") + " ANTHROPIC_API_KEY is not set (required for AI-powered tests)");
     allPassed = false;
   }
-  const dbPath = join6(process.env["HOME"] ?? "~", ".testers", "testers.db");
+  const dbPath = join7(process.env["HOME"] ?? "~", ".testers", "testers.db");
   try {
     const { Database: Database3 } = await import("bun:sqlite");
     const db2 = new Database3(dbPath, { create: true });
@@ -13843,6 +14932,23 @@ program2.command("doctor").description("Check system setup and configuration").a
   const { isLightpandaAvailable: isLightpandaAvailable2 } = await Promise.resolve().then(() => (init_browser_lightpanda(), exports_browser_lightpanda));
   const lightpandaAvailable = isLightpandaAvailable2();
   log((lightpandaAvailable ? chalk6.green("\u2713") : chalk6.dim("\u25CB")) + ` Lightpanda: ${lightpandaAvailable ? "installed" : "not installed (optional)"}`);
+  const { isBunWebViewAvailable: isBunWebViewAvailable2 } = await Promise.resolve().then(() => (init_browser_bun(), exports_browser_bun));
+  const bunAvailable = isBunWebViewAvailable2();
+  log((bunAvailable ? chalk6.green("\u2713") : chalk6.dim("\u25CB")) + ` Bun.WebView: ${bunAvailable ? "available (native, ~11x faster)" : "not available \u2014 upgrade to Bun canary: bun upgrade --canary (optional)"}`);
+  log("");
+  log(chalk6.dim("  AI Providers:"));
+  const anthropicKey = !!process.env["ANTHROPIC_API_KEY"];
+  const openaiKey = !!process.env["OPENAI_API_KEY"];
+  const googleKey = !!process.env["GOOGLE_API_KEY"];
+  const cerebrasKey = !!process.env["CEREBRAS_API_KEY"];
+  log((anthropicKey ? chalk6.green("  \u2713") : chalk6.red("  \u2717")) + ` Anthropic (ANTHROPIC_API_KEY)${!anthropicKey ? " \u2014 required for default model" : ""}`);
+  log((openaiKey ? chalk6.green("  \u2713") : chalk6.dim("  \u25CB")) + ` OpenAI (OPENAI_API_KEY) \u2014 optional, enables gpt-* models`);
+  log((googleKey ? chalk6.green("  \u2713") : chalk6.dim("  \u25CB")) + ` Google Gemini (GOOGLE_API_KEY) \u2014 optional, enables gemini-* models`);
+  log((cerebrasKey ? chalk6.green("  \u2713") : chalk6.dim("  \u25CB")) + ` Cerebras (CEREBRAS_API_KEY) \u2014 optional, enables llama-*/qwen-* at ~20x faster inference`);
+  if (!anthropicKey && !openaiKey && !googleKey && !cerebrasKey) {
+    log(chalk6.red("  \u2717") + " No AI provider API keys found \u2014 at least one is required");
+    allPassed = false;
+  }
   if (!allPassed) {
     process.exit(1);
   }
@@ -13851,7 +14957,7 @@ program2.command("serve").description("Start the Open Testers web dashboard").op
   try {
     const port = parseInt(opts.port, 10);
     const url = `http://localhost:${port}`;
-    const serverBin = join6(resolve(process.execPath, ".."), "..", "dist", "server", "index.js");
+    const serverBin = join7(resolve(process.execPath, ".."), "..", "dist", "server", "index.js");
     const { join: pathJoin, resolve: pathResolve, dirname: dirname2 } = await import("path");
     const { fileURLToPath } = await import("url");
     const serverPath = pathJoin(dirname2(fileURLToPath(import.meta.url)), "..", "server", "index.js");
@@ -14403,6 +15509,24 @@ personaCmd.command("attach <persona-id> <scenario-id>").description("Attach a pe
     process.exit(1);
   }
 });
+personaCmd.command("seed").description("Seed the 7 default global personas (idempotent)").option("--json", "Output as JSON", false).action((seedOpts) => {
+  try {
+    const { seedDefaultPersonas: seedDefaultPersonas2 } = (init_seed_personas(), __toCommonJS(exports_seed_personas));
+    const result = seedDefaultPersonas2();
+    if (seedOpts.json) {
+      log(JSON.stringify(result, null, 2));
+    } else {
+      if (result.seeded > 0) {
+        log(chalk6.green(`Seeded ${result.seeded} default personas.`));
+      } else {
+        log(chalk6.dim(`Default personas already present (${result.skipped} skipped).`));
+      }
+    }
+  } catch (error) {
+    logError(chalk6.red(`Error: ${error instanceof Error ? error.message : String(error)}`));
+    process.exit(1);
+  }
+});
 personaCmd.command("detach <scenario-id>").description("Detach persona from a scenario").action(async (scenarioId) => {
   try {
     const scenario = getScenario(scenarioId) ?? getScenarioByShortId(scenarioId);
@@ -14677,6 +15801,107 @@ goldenCmd.command("run <base-url>").description("Run all golden answer checks an
     process.exit(1);
   }
 });
+program2.command("run-many <url>").description("Run scenarios \xD7 personas matrix \u2014 test each scenario under multiple personas").option("--personas <ids>", "Comma-separated persona IDs, or 'all' for all global personas", "all").option("--scenarios <ids>", "Comma-separated scenario IDs, or 'all'", "all").option("--parallel <n>", "Parallel workers per run", "2").option("--model <model>", "AI model to use").option("--project <id>", "Filter by project ID").option("--json", "Output as JSON", false).action(async (url, opts) => {
+  try {
+    const projectId = resolveProject(opts.project);
+    let personas;
+    if (opts.personas === "all") {
+      personas = listPersonas({ globalOnly: true, enabled: true });
+    } else {
+      const ids = opts.personas.split(",").map((s) => s.trim()).filter(Boolean);
+      personas = ids.map((id) => getPersona(id)).filter(Boolean);
+    }
+    if (personas.length === 0) {
+      logError(chalk6.red("No personas found. Run: testers persona seed"));
+      process.exit(1);
+    }
+    let scenarios;
+    if (opts.scenarios === "all") {
+      scenarios = listScenarios({ projectId, limit: 20 });
+    } else {
+      const ids = opts.scenarios.split(",").map((s) => s.trim()).filter(Boolean);
+      const all = listScenarios({ projectId });
+      scenarios = all.filter((s) => ids.includes(s.id) || ids.includes(s.shortId));
+    }
+    if (scenarios.length === 0) {
+      logError(chalk6.red("No scenarios found."));
+      process.exit(1);
+    }
+    log("");
+    log(chalk6.bold(`  Running ${scenarios.length} scenarios \xD7 ${personas.length} personas (${scenarios.length * personas.length} total runs)`));
+    log("");
+    const matrixResults = [];
+    for (const persona of personas) {
+      if (!persona)
+        continue;
+      log(chalk6.dim(`  Starting run for persona: ${persona.name} ...`));
+      const { run, results } = await runByFilter({
+        url,
+        scenarioIds: scenarios.map((s) => s.id),
+        model: opts.model,
+        parallel: parseInt(opts.parallel, 10),
+        projectId,
+        personaId: persona.id
+      });
+      matrixResults.push({ personaName: persona.name, runId: run.id, run });
+      const status = run.status === "passed" ? chalk6.green("PASS") : chalk6.red("FAIL");
+      log(`  ${status}  ${persona.name.padEnd(24)} ${run.passed}/${run.total} passed`);
+    }
+    if (opts.json) {
+      log(JSON.stringify(matrixResults.map((r) => ({ personaName: r.personaName, runId: r.runId, run: r.run })), null, 2));
+    } else {
+      log("");
+      log(chalk6.bold("  Summary"));
+      let allPassed = true;
+      for (const r of matrixResults) {
+        if (r.run && r.run.failed > 0)
+          allPassed = false;
+      }
+      log(allPassed ? chalk6.green("  All personas passed!") : chalk6.yellow("  Some personas had failures \u2014 review per-persona results above."));
+      log("");
+    }
+    const anyFailed = matrixResults.some((r) => r.run && r.run.failed > 0);
+    process.exit(anyFailed ? 1 : 0);
+  } catch (error) {
+    logError(chalk6.red(`Error: ${error instanceof Error ? error.message : String(error)}`));
+    process.exit(1);
+  }
+});
+program2.command("run-script <file>").description("Run a hybrid test script (.ts) that exports an array of HybridScenario objects").option("--url <url>", "Base URL to run against").option("--json", "Output as JSON", false).action(async (file, opts) => {
+  try {
+    const { resolve: resolve2 } = await import("path");
+    const { runHybridScenario: runHybridScenario2 } = await Promise.resolve().then(() => (init_hybrid_runner(), exports_hybrid_runner));
+    const scriptPath = resolve2(process.cwd(), file);
+    const mod = await import(scriptPath);
+    const scenarios = mod.scenarios ?? mod.default ?? [];
+    if (!Array.isArray(scenarios) || scenarios.length === 0) {
+      logError(chalk6.red(`No scenarios exported from ${file}. Export an array as 'export const scenarios = [...]'`));
+      process.exit(1);
+    }
+    const results = [];
+    for (const scenario of scenarios) {
+      log(chalk6.dim(`Running: ${scenario.name} ...`));
+      const result = await runHybridScenario2(scenario, { baseUrl: opts.url });
+      results.push(result);
+      const icon = result.status === "passed" ? chalk6.green("PASS") : chalk6.red("FAIL");
+      log(`${icon}  ${result.name ?? scenario.name} (${result.durationMs}ms)`);
+      if (result.status !== "passed" && result.error) {
+        log(chalk6.dim(`     ${result.error}`));
+      }
+    }
+    if (opts.json) {
+      log(JSON.stringify(results, null, 2));
+    }
+    const passed = results.filter((r) => r.status === "passed").length;
+    const failed = results.length - passed;
+    log("");
+    log(chalk6.bold(`Results: ${passed}/${results.length} passed${failed > 0 ? `, ${failed} failed` : ""}`));
+    process.exit(failed > 0 ? 1 : 0);
+  } catch (error) {
+    logError(chalk6.red(`Error: ${error instanceof Error ? error.message : String(error)}`));
+    process.exit(1);
+  }
+});
 program2.hook("preAction", () => {
   const opts = program2.opts();
   QUIET = opts.quiet === true;