explorbot 0.1.15 → 0.1.17

package/bin/explorbot-cli.ts

@@ -122,7 +122,15 @@ addCommonOptions(program.command('start [path]').description('Start web explorat
   await startTUI(explorBot);
 });
 
-addCommonOptions(program.command('explore <path>').description('Explore a page autonomously and run invented scenarios').option('--max-tests <count>', 'Maximum number of tests to run').option('--focus <feature>', 'Focus area for exploration')).action(async (explorePath, options) => {
+addCommonOptions(
+  program
+    .command('explore <path>')
+    .description('Explore a page autonomously and run invented scenarios')
+    .option('--max-tests <count>', 'Maximum number of tests to run')
+    .option('--focus <feature>', 'Focus area for exploration')
+    .option('--configure <spec>', 'Reuse spec: keys new|from|style|subpages|pick_by|priority, e.g. "new:25%;pick_by=random;priority=critical,high"')
+    .option('--dry-run', 'Mark picked tests as skipped without executing or generating new ones')
+).action(async (explorePath, options) => {
   try {
     const explorBot = new ExplorBot(buildExplorBotOptions(explorePath, options));
     await explorBot.start();
@@ -130,8 +138,11 @@ addCommonOptions(program.command('explore <path>').description('Explore a page a
     const { ExploreCommand } = await import('../src/commands/explore-command.js');
     const cmd = new ExploreCommand(explorBot);
     if (options.maxTests) cmd.maxTests = Number.parseInt(options.maxTests, 10);
+    if (options.dryRun) cmd.dryRun = true;
     const execArgs: string[] = [];
     if (options.focus) execArgs.push('--focus', `"${options.focus}"`);
+    if (options.configure) execArgs.push('--configure', `"${options.configure}"`);
+    if (options.dryRun) execArgs.push('--dry-run');
     await cmd.execute(execArgs.join(' '));
     await explorBot.stop();
     await showStatsAndExit(0);

package/dist/bin/explorbot-cli.js

@@ -93,7 +93,13 @@ addCommonOptions(program.command('start [path]').description('Start web explorat
     await explorBot.start();
     await startTUI(explorBot);
 });
-addCommonOptions(program.command('explore <path>').description('Explore a page autonomously and run invented scenarios').option('--max-tests <count>', 'Maximum number of tests to run').option('--focus <feature>', 'Focus area for exploration')).action(async (explorePath, options) => {
+addCommonOptions(program
+    .command('explore <path>')
+    .description('Explore a page autonomously and run invented scenarios')
+    .option('--max-tests <count>', 'Maximum number of tests to run')
+    .option('--focus <feature>', 'Focus area for exploration')
+    .option('--configure <spec>', 'Reuse spec: keys new|from|style|subpages|pick_by|priority, e.g. "new:25%;pick_by=random;priority=critical,high"')
+    .option('--dry-run', 'Mark picked tests as skipped without executing or generating new ones')).action(async (explorePath, options) => {
     try {
         const explorBot = new ExplorBot(buildExplorBotOptions(explorePath, options));
         await explorBot.start();
@@ -102,9 +108,15 @@ addCommonOptions(program.command('explore <path>').description('Explore a page a
         const cmd = new ExploreCommand(explorBot);
         if (options.maxTests)
             cmd.maxTests = Number.parseInt(options.maxTests, 10);
+        if (options.dryRun)
+            cmd.dryRun = true;
         const execArgs = [];
         if (options.focus)
             execArgs.push('--focus', `"${options.focus}"`);
+        if (options.configure)
+            execArgs.push('--configure', `"${options.configure}"`);
+        if (options.dryRun)
+            execArgs.push('--dry-run');
         await cmd.execute(execArgs.join(' '));
         await explorBot.stop();
         await showStatsAndExit(0);

package/dist/package.json

@@ -1,6 +1,6 @@
 {
   "name": "explorbot",
-  "version": "0.1.15",
+  "version": "0.1.17",
   "description": "CLI app built with React Ink, CodeceptJS, and Playwright",
   "license": "Elastic-2.0",
   "type": "module",

package/dist/src/ai/pilot.js

@@ -277,14 +277,9 @@ export class Pilot {
       - "Edit X" → updated value must be persisted (visible in list/detail). Opening edit is NOT enough; redirect after save with the new value visible IS enough.
       - Negative tests ("without a name", "invalid", "duplicate", "unauthorized") → success means the system PREVENTED the action with validation/error.
 
-      PROVENANCE for create/edit scenarios: the task prompt instructs the tester to inject the
-      session marker "${task.sessionName ?? ''}" into newly created or edited free-text values.
-      When that marker COULD be injected, the entity used as proof MUST contain it. A record
-      matching the goal by text alone but missing the marker is a stale leftover from a prior
-      run — it is NOT evidence the current scenario produced anything. Vote \`fail\`, not \`pass\`.
-      This does not apply when the field is restricted (numeric only, enum, etc.) or when the
-      session_log shows no fillField/type/select actions were attempted at all (in that case
-      the scenario clearly didn't run — also vote \`fail\`).
+      PROVENANCE: the entity you cite as proof must appear by name in <notes> or
+      <session_log> tool inputs for THIS run. Name absent from tester activity = stale
+      coincidence, vote \`fail\`. Same if no fillField/type/select/click on a target ran.
 
       Expected results are MILESTONES, not the goal. Never fail because a milestone (toast, icon, styling)
       didn't match if the scenario goal IS accomplished.

package/dist/src/ai/researcher/focus.js

@@ -1,20 +1,61 @@
-import { detectFocusArea } from "../../utils/aria.js";
 import { mdq } from "../../utils/markdown-query.js";
 export const FOCUSED_MARKER = '> **Focused**';
 const FOCUS_SKIP_SECTIONS = new Set(['navigation', 'menu']);
 export function hasFocusedSection(text) {
     return text.includes(FOCUSED_MARKER);
 }
-export function detectFocusFromAria(ariaSnapshot, sections) {
-    const focusArea = detectFocusArea(ariaSnapshot);
-    if (!focusArea.detected)
-        return null;
-    if (focusArea.type === 'dialog' || focusArea.type === 'modal') {
-        const dialogSection = sections.find((s) => s.containerCss && (s.containerCss.includes('[role="dialog"]') || s.containerCss.includes('[role="alertdialog"]') || s.containerCss.includes('[aria-modal')));
-        if (dialogSection)
-            return dialogSection.name;
+export async function detectFocusedSection(page, sections) {
+    const candidates = [];
+    for (const section of sections) {
+        if (!section.containerCss)
+            continue;
+        const key = section.name.toLowerCase().replace(/^section:\s*/, '');
+        if (FOCUS_SKIP_SECTIONS.has(key))
+            continue;
+        try {
+            const locator = page.locator(section.containerCss).first();
+            if (!(await locator.isVisible()))
+                continue;
+            const probe = await locator.evaluate((el) => {
+                const dialogSelector = '[role="dialog"], [role="alertdialog"], [aria-modal="true"]';
+                const isDialog = el.matches(dialogSelector) || !!el.querySelector(dialogSelector);
+                let cur = el;
+                let maxZ = 0;
+                while (cur && cur !== document.body) {
+                    const cs = window.getComputedStyle(cur);
+                    if (cs.position !== 'static') {
+                        const z = Number.parseInt(cs.zIndex, 10);
+                        if (!Number.isNaN(z) && z > maxZ)
+                            maxZ = z;
+                    }
+                    cur = cur.parentElement;
+                }
+                const shadow = window.getComputedStyle(el).boxShadow;
+                const hasShadow = !!shadow && shadow !== 'none';
+                return { isDialog, zIndex: maxZ, hasShadow };
+            });
+            candidates.push({ name: section.name, ...probe });
+        }
+        catch { }
     }
-    return null;
+    if (candidates.length === 0)
+        return null;
+    const dialogs = candidates.filter((c) => c.isDialog);
+    const pool = dialogs.length > 0 ? dialogs : candidates;
+    const winner = pool.reduce((best, c) => {
+        if (!best)
+            return c;
+        if (c.zIndex !== best.zIndex)
+            return c.zIndex > best.zIndex ? c : best;
+        if (c.hasShadow !== best.hasShadow)
+            return c.hasShadow ? c : best;
+        return best;
+    }, null);
+    if (!winner)
+        return null;
+    if (dialogs.length === 0 && winner.zIndex === 0 && !winner.hasShadow)
+        return null;
+    return winner.name;
 }
 export function markSectionAsFocused(result, sectionName) {
     if (hasFocusedSection(result.text))

package/dist/src/ai/researcher/sections.js

@@ -3,6 +3,8 @@ import { executionController } from "../../execution-controller.js";
 import { tag } from '../../utils/logger.js';
 import { RulesLoader } from "../../utils/rules-loader.js";
 import { locatorRule as generalLocatorRuleText } from '../rules.js';
+import { markSectionAsFocused } from "./focus.js";
+import { ResearchResult } from "./research-result.js";
 export function WithSections(Base) {
     return class extends Base {
         async researchBySections() {
@@ -40,10 +42,12 @@ export function WithSections(Base) {
             if (parts.length === 0) {
                 throw new Error('Per-section research produced no sections — AI responses all empty or NOT_PRESENT');
             }
-            let merged = parts.join('\n\n');
-            if (focusCss)
-                merged += '\n\n> Focused: Focus';
-            return merged;
+            const merged = parts.join('\n\n');
+            if (!focusCss)
+                return merged;
+            const focused = new ResearchResult(merged, this.actionResult?.url || '');
+            markSectionAsFocused(focused, 'Focus');
+            return focused.text;
         }
         async _detectFocusCss() {
             const focusSections = this.explorer.getConfig().ai?.agents?.researcher?.focusSections;

package/dist/src/ai/researcher.js

@@ -16,7 +16,7 @@ import { ContextLengthError } from './provider.js';
 import { findSimilarResearch, getCachedResearch, saveResearch } from "./researcher/cache.js";
 import { WithCoordinates } from "./researcher/coordinates.js";
 import { WithDeepAnalysis } from "./researcher/deep-analysis.js";
-import { detectFocusFromAria, hasFocusedSection, markSectionAsFocused, pickDefaultFocusedSection } from "./researcher/focus.js";
+import { detectFocusedSection, hasFocusedSection, markSectionAsFocused, pickDefaultFocusedSection } from "./researcher/focus.js";
 import { WithLocators } from "./researcher/locators.js";
 import { extractValidContainers, formatResearchSummary, parseResearchSections } from "./researcher/parser.js";
 import { ResearchResult } from "./researcher/research-result.js";
@@ -186,18 +186,13 @@ export class Researcher extends ResearcherBase {
             if (!interrupted() && fix && result.locators.some((l) => l.valid === false)) {
                 await this.fixBrokenSections(result, activeConversation);
             }
-            // Focused section: parse AI declaration, then ARIA fallback
-            const focusMatch = result.text.match(/^>\s*Focused:\s*(.+)/m);
-            if (focusMatch) {
-                result.text = result.text.replace(focusMatch[0], '');
-                markSectionAsFocused(result, focusMatch[1].trim());
-            }
-            if (!hasFocusedSection(result.text)) {
+            // Focused section: unified Playwright probe (HTML+CSS+visibility).
+            // Must run BEFORE visuallyAnnotateContainers — annotation overlays inject z-index 99998+ which would pollute the scoring.
+            if (!interrupted() && this.hasScreenshotToAnalyze) {
                 const sections = parseResearchSections(result.text);
-                const ariaSnapshot = this.actionResult?.getCompactARIA() || '';
-                const focusedName = detectFocusFromAria(ariaSnapshot, sections);
-                if (focusedName)
-                    markSectionAsFocused(result, focusedName);
+                const focused = await detectFocusedSection(this.explorer.playwrightHelper.page, sections);
+                if (focused)
+                    markSectionAsFocused(result, focused);
             }
             // Stage 4: Visual analysis
             if (!interrupted() && this.hasScreenshotToAnalyze) {
@@ -232,8 +227,8 @@ export class Researcher extends ResearcherBase {
                 await this.backfillCoordinates(result);
                 await this.backfillBrokenLocators(result);
             }
-            // Focused section: final fallback
-            if (!hasFocusedSection(result.text)) {
+            // Focused section: final fallback (vision-only — without a screenshot we don't infer focus)
+            if (this.hasScreenshotToAnalyze && !hasFocusedSection(result.text)) {
                 const sections = parseResearchSections(result.text);
                 const fallback = pickDefaultFocusedSection(sections);
                 if (fallback)
@@ -388,16 +383,6 @@ export class Researcher extends ResearcherBase {
 
       | Element | ARIA | CSS | eidx |
       </section_format>
-
-      <focused_section>
-      At the end of your output, declare the primary focus area on a single line:
-
-      > Focused: <exact section name>
-
-      - If a dialog/modal/drawer/overlay exists, it is focused.
-      - Otherwise pick the section where the main business action happens (list for catalog, detail for item page, content for article).
-      - Navigation and menu/toolbar are never focused.
-      </focused_section>
     `;
     }
     async buildResearchPrompt() {

package/dist/src/ai/tester.js

@@ -329,10 +329,15 @@ export class Tester extends TaskAgent {
                     : undefined,
                 catch: async ({ error, stop }) => {
                     tag('error').log(`Test execution error: ${error}`);
+                    const message = error instanceof Error ? error.message : String(error);
                     if (!task.hasFinished) {
-                        task.addNote(`Execution error: ${error instanceof Error ? error.message : String(error)}`);
+                        task.addNote(`Execution error: ${message}`);
                     }
-                    stop();
+                    if (error instanceof Error && error.name === 'AbortError') {
+                        stop();
+                        return;
+                    }
+                    conversation.addUserText(`Previous AI call failed: ${message}. Take a different approach on the next step.`);
                 },
             });
             if (task.hasFinished)
@@ -643,6 +648,7 @@ export class Tester extends TaskAgent {
     - Use pressKey() for pressing special keys (Enter, Escape, Tab, Arrow keys) or key combinations with modifiers (Ctrl+A, Shift+Delete, etc.)
     - Use container CSS locators from <page_ui_map> to interact with elements inside sections
     - Systematically use record({ notes: ["..."] }) to write your findings, planned actions, observations, etc.
+    - When creating/editing/deleting a named entity, include its identifier verbatim in the note — Pilot uses it to confirm provenance.
     - Call record({ notes: ["..."], status: "success" }) when you see success/info message on a page or when expected outcome is achieved
     - Call record({ notes: ["..."], status: "fail" }) when an expected outcome cannot be achieved or has failed or you see error/alert/warning message on a page
     - NEVER call record(status: "success") if your last verify() or see() call FAILED. A failed check means the outcome is NOT confirmed — use record(status: "fail") instead, or retry with a different approach.