explorbot 0.1.25 → 0.1.27

package/src/ai/pilot.ts

@@ -104,7 +104,7 @@ export class Pilot implements Agent {
 
     const schema = z.object({
       decision: z.enum(['pass', 'fail', 'continue', 'skipped']).describe('pass = test succeeded, fail = test failed, continue = tester should keep going, skipped = scenario is irrelevant OR systematic execution failures prevented testing'),
-      reason: z.string().describe('What happened and why (1-2 sentences). Do NOT repeat the decision status (e.g. "scenario goal achieved/not achieved") — just explain the evidence. For continue: explain why rejected and suggest alternatives.'),
+      reason: z.string().describe('Concise user-facing reason, maximum 1 short sentence and 120 characters. Do NOT repeat the decision status; explain only the evidence. For continue: explain why rejected and suggest alternatives.'),
       guidance: z.string().nullable().describe('Required for "continue": specific actionable instruction for the tester — what exactly to verify, retry differently, or complete next. Be concrete.'),
       requestVerification: z
         .string()
@@ -177,7 +177,7 @@ export class Pilot implements Agent {
         }
       }
 
-      tag('info').log(`Pilot: ${result.decision} — ${result.reason}`);
+      tag('info').log(`Pilot: ${result.decision} - ${result.reason}`);
       task.summary = result.reason;
 
       const verdictState = screenshotState || currentState;
@@ -221,7 +221,7 @@ export class Pilot implements Agent {
 
     const schema = z.object({
       decision: z.enum(['allow', 'fail', 'continue', 'skipped']).describe('allow = reset proceeds, fail = test failed (stop looping), continue = veto reset, tester should act on current page instead, skipped = scenario is irrelevant or cannot be executed'),
-      reason: z.string().describe('What evidence justifies this decision (1-2 sentences). Do not restate the decision.'),
+      reason: z.string().describe('Concise evidence-only reason, maximum 1 short sentence and 120 characters. Do not restate the decision.'),
       guidance: z.string().nullable().describe('Required for "continue": concrete instruction for what the tester should do instead of resetting (e.g. which tool to call, what to verify).'),
     });
 
@@ -388,8 +388,9 @@ export class Pilot implements Agent {
       - "continue": tester hasn't completed the goal; provide concrete guidance (which tool, what to check).
         If a verify() asserted a state that was ALREADY TRUE before the test, it proves nothing — reject.
 
-      reason field: do NOT restate the decision ("scenario goal achieved/not achieved"). State what happened —
-      what was verified, what failed, what evidence was found.
+      reason field: one short sentence, maximum 120 characters. Do NOT restate the decision
+      ("scenario goal achieved/not achieved"). State what happened: what was verified, what failed,
+      or what evidence was found.
     `;
   }
 
@@ -1017,6 +1018,8 @@ export class Pilot implements Agent {
       Response format:
       PROGRESS: <1 sentence assessment>
       NEXT: <specific actionable instruction for Tester>
+
+      Keep user-facing reasons concise: one short sentence, maximum 120 characters, evidence only, no repeated verdict wording.
     `;
   }
 }

package/src/ai/planner.ts

@@ -36,7 +36,7 @@ const TasksSchema = z.object({
         scenario: z.string().describe('A single sentence describing what to test'),
         priority: z.enum(['critical', 'important', 'high', 'normal', 'low']).describe('Priority of the task based on business importance'),
         startUrl: z.string().nullable().describe('Start URL for the test if different from plan URL (only for tests on visited subpages)'),
-        steps: z.array(z.string()).describe('List of steps to perform for this scenario. Each step should be a specific action (e.g., "Click on Login button", "Enter username in email field", "Submit the form"). Keep steps atomic and actionable.'),
+        steps: z.array(z.string()).describe('List of steps to perform for this scenario. Each step should be a specific action (e.g., "Open the form", "Enter required data", "Submit the form"). Keep steps atomic and actionable.'),
         expectedOutcomes: z
           .array(z.string())
           .describe('List of expected outcomes that can be verified. Each outcome should be simple, specific, and easy to check (e.g., "Success message appears", "URL changes to /dashboard", "Form field shows error"). Keep outcomes atomic - do not combine multiple checks into one.'),
@@ -226,9 +226,7 @@ export class Planner extends PlannerBase implements Agent {
       }
     }
 
-    const availableStyles = Object.keys(getStyles()).join(', ');
     tag('success').log(`Planning complete! ${this.currentPlan.tests.length} tests in plan: ${this.currentPlan.title}`);
-    tag('info').log(`Planning style: ${this.lastStyleName} (available: ${availableStyles})`);
 
     if (state.url) registerPlan(state.url, this.currentPlan, feature, state.hash);
 

package/src/ai/provider.ts

@@ -294,8 +294,11 @@ export class Provider {
         }
         throw new ContextLengthError(error.message || error.toString());
       }
-      tag('error').log(error.message || error.toString());
-      throw new AiError(error.message || error.toString());
+      const message = error.message || error.toString();
+      if (message !== 'No response text from AI') {
+        tag('error').log(message);
+      }
+      throw new AiError(message);
     }
   }
 
@@ -376,7 +379,6 @@ export class Provider {
     } catch (error: any) {
       clearActivity();
       if (error?.message?.includes('Tool choice is required')) {
-        tag('warning').log('Model completed without calling a tool, returning empty result');
         return { text: '', toolCalls: [], toolResults: [], response: { messages: [] }, usage: null };
       }
       if (error?.name === 'AbortError') throw error;

package/src/ai/researcher/cache.ts

@@ -49,6 +49,13 @@ export function getCachedResearch(hash: string): string {
   return cached;
 }
 
+export function getPreviousResearch(hash: string): string {
+  if (!hash) return '';
+  const researchFile = outputPath('research', `${hash}.md`);
+  if (!existsSync(researchFile)) return '';
+  return readFileSync(researchFile, 'utf8');
+}
+
 export function saveResearch(hash: string, text: string, combinedHtml?: string): string {
   const researchDir = outputPath('research');
   const researchFile = join(researchDir, `${hash}.md`);

package/src/ai/researcher/deep-analysis.ts

@@ -5,10 +5,11 @@ import type Explorer from '../../explorer.ts';
 import type { StateManager } from '../../state-manager.js';
 import { WebPageState } from '../../state-manager.js';
 import { detectFocusArea, diffAriaSnapshots } from '../../utils/aria.ts';
+import { extractCodeBlocks } from '../../utils/code-extractor.ts';
 import { tag } from '../../utils/logger.js';
 import { mdq } from '../../utils/markdown-query.ts';
 import type { Provider } from '../provider.js';
-import { getCachedResearch, saveResearch } from './cache.ts';
+import { getCachedResearch, getPreviousResearch, saveResearch } from './cache.ts';
 import { type Constructor, debugLog } from './mixin.ts';
 import { type ResearchElement, parseResearchSections } from './parser.ts';
 import type { ResearchResult } from './research-result.ts';
@@ -26,14 +27,30 @@ export function WithDeepAnalysis<T extends Constructor>(Base: T) {
       tag('info').log('Starting deep analysis of expandable elements');
       await (this as any).navigateTo(state.fullUrl || state.url);
 
-      let expandables = await this._discoverExpandables(result.text);
-      if (expandables.length === 0) {
-        tag('info').log('No expandable elements identified by AI');
-        return;
+      const maxClicks = (this.explorer.getConfig().ai?.agents?.researcher as any)?.maxExpandableClicks ?? DEFAULT_MAX_EXPANDABLE_CLICKS;
+
+      const expandedSections: string[] = [];
+      const navigationLinks: Array<{ code: string; url: string }> = [];
+      let verifiedCodes: string[] = [];
+      let missing: PreviousSection[] = [];
+
+      const previousSections = this._loadPreviousExtendedSections(state.hash || '');
+      if (previousSections.length > 0) {
+        tag('substep').log(`Replaying ${previousSections.length} previously discovered sections`);
+        const replay = await this._replayPreviousSections(state, previousSections, maxClicks);
+        expandedSections.push(...replay.verified);
+        verifiedCodes = replay.verifiedCodes;
+        missing = replay.missing;
+        tag('info').log(`Reused ${replay.verified.length}/${previousSections.length} previous sections, ${missing.length} to re-discover`);
+
+        if (missing.length === 0 && replay.verified.length >= maxClicks) {
+          tag('info').log('Page appears unchanged, reusing previous sections and skipping discovery');
+          this._appendExtendedResearch(result, expandedSections, navigationLinks);
+          return;
+        }
       }
-      tag('substep').log(`Identified ${expandables.length} expandable elements`);
 
-      const maxClicks = (this.explorer.getConfig().ai?.agents?.researcher as any)?.maxExpandableClicks ?? DEFAULT_MAX_EXPANDABLE_CLICKS;
+      let expandables = await this._discoverExpandables(result.text, missing, verifiedCodes);
       if (expandables.length > maxClicks) {
         expandables = await this._selectExpandables(expandables, state.fullUrl || state.url, maxClicks);
         tag('substep').log(`Selected ${expandables.length} expandables to click (max: ${maxClicks})`);
@@ -44,10 +61,12 @@ export function WithDeepAnalysis<T extends Constructor>(Base: T) {
           commands: this._buildClickCommands(el),
           description: el.name,
         }))
-        .filter((el) => el.commands.length > 0);
+        .filter((el) => el.commands.length > 0)
+        .filter((el) => !el.commands.some((cmd) => verifiedCodes.includes(cmd)));
 
       if (elements.length === 0) {
-        tag('info').log('No expandables with valid locators');
+        tag('info').log('No new expandable elements to click');
+        this._appendExtendedResearch(result, expandedSections, navigationLinks);
         return;
       }
 
@@ -57,25 +76,11 @@ export function WithDeepAnalysis<T extends Constructor>(Base: T) {
       for (const el of elements) debugLog(`Expandable: ${el.description} → ${el.commands[0]}`);
       tag('substep').log(`Clicking ${elements.length} expandable elements`);
 
-      const expandedSections: string[] = [];
-      const navigationLinks: Array<{ code: string; url: string }> = [];
-
       await this._clickExpandableElements(elements, state, expandedSections, navigationLinks);
 
       tag('info').log(`Deep analysis complete. Sections: ${expandedSections.length}, navigation links: ${navigationLinks.length}`);
 
-      const dedupedSections = this._deduplicateExpandedSections(expandedSections);
-      if (dedupedSections.length !== expandedSections.length) {
-        tag('substep').log(`Deduplicated ${expandedSections.length} → ${dedupedSections.length} extended sections`);
-      }
-
-      if (dedupedSections.length > 0) {
-        result.text += `\n\n# Extended Research\n\n${dedupedSections.join('\n\n---\n\n')}`;
-      }
-      if (navigationLinks.length > 0) {
-        const links = navigationLinks.map((l) => `- \`${l.code}\` opens ${l.url}`).join('\n');
-        result.text += `\n\n## Navigation Links\n\n${links}`;
-      }
+      this._appendExtendedResearch(result, expandedSections, navigationLinks);
     }
 
     async researchOverlay(current: ActionResult, previous: ActionResult, pageStateHash: string): Promise<string | null> {
@@ -127,7 +132,70 @@ export function WithDeepAnalysis<T extends Constructor>(Base: T) {
       return sectionMarkdown;
     }
 
-    private async _discoverExpandables(researchText: string): Promise<ExpandableElement[]> {
+    private _loadPreviousExtendedSections(hash: string): PreviousSection[] {
+      if (!hash) return [];
+      const previous = getPreviousResearch(hash);
+      if (!previous) return [];
+
+      const sections: PreviousSection[] = [];
+      for (const section of parseResearchSections(previous)) {
+        if (!section.isExtended) continue;
+        const code = extractCodeBlocks(section.rawMarkdown)[0];
+        if (!code) continue;
+        sections.push({ name: section.name, code });
+      }
+      return sections;
+    }
+
+    private async _replayPreviousSections(state: WebPageState, prevSections: PreviousSection[], maxClicks: number): Promise<{ verified: string[]; verifiedCodes: string[]; missing: PreviousSection[] }> {
+      const originalAria = state.ariaSnapshot || '';
+      const verified: string[] = [];
+      const verifiedCodes: string[] = [];
+      const missing: PreviousSection[] = [];
+
+      for (const section of prevSections.slice(0, maxClicks)) {
+        if (executionController.isInterrupted()) break;
+
+        let outcome: ExpansionOutcome;
+        try {
+          outcome = await this._executeAndAnalyze([section.code], section.name, state, originalAria, this._summarizeExpanded(verified));
+        } catch (err) {
+          tag('warning').log(`Replay failed for "${section.name}": ${err instanceof Error ? err.message : err}`);
+          await this._restorePageState(state.url, originalAria).catch(() => {});
+          missing.push(section);
+          continue;
+        }
+
+        if (outcome.status === 'revealed') {
+          verified.push(outcome.sectionMarkdown);
+          verifiedCodes.push(section.code);
+          debugLog(`Replayed and verified section: ${section.name}`);
+          continue;
+        }
+
+        debugLog(`Could not replay previous section: ${section.name}`);
+        missing.push(section);
+      }
+
+      return { verified, verifiedCodes, missing };
+    }
+
+    private _appendExtendedResearch(result: ResearchResult, expandedSections: string[], navigationLinks: Array<{ code: string; url: string }>): void {
+      const dedupedSections = this._deduplicateExpandedSections(expandedSections);
+      if (dedupedSections.length !== expandedSections.length) {
+        tag('substep').log(`Deduplicated ${expandedSections.length} → ${dedupedSections.length} extended sections`);
+      }
+
+      if (dedupedSections.length > 0) {
+        result.text += `\n\n# Extended Research\n\n${dedupedSections.join('\n\n---\n\n')}`;
+      }
+      if (navigationLinks.length > 0) {
+        const links = navigationLinks.map((l) => `- \`${l.code}\` opens ${l.url}`).join('\n');
+        result.text += `\n\n## Navigation Links\n\n${links}`;
+      }
+    }
+
+    private async _discoverExpandables(researchText: string, missing: PreviousSection[] = [], verifiedCodes: string[] = []): Promise<ExpandableElement[]> {
       const allElements = new Map<string, ExpandableElement>();
       for (const section of parseResearchSections(researchText)) {
         for (const el of section.elements) {
@@ -138,6 +206,16 @@ export function WithDeepAnalysis<T extends Constructor>(Base: T) {
 
       const eidxList = [...allElements.keys()].join(', ');
 
+      let missingHint = '';
+      if (missing.length > 0) {
+        const list = missing.map((s) => `- "${s.name}" (previously revealed via ${s.code})`).join('\n');
+        missingHint = dedent`
+
+          These sections were present on a previous visit but their trigger could not be replayed now — the element may have moved or been renamed. Prioritize finding the element that now reveals each:
+          ${list}
+        `;
+      }
+
       const textPrompt = dedent`
         From this UI research, identify elements that could reveal hidden UI when clicked
         (dropdown menus, popups, expandable panels, accordion sections, overflow menus, tab switches).
@@ -145,6 +223,7 @@ export function WithDeepAnalysis<T extends Constructor>(Base: T) {
         Available eidx refs: ${eidxList}
 
         ${researchText}
+        ${missingHint}
 
         Rules:
         - Only pick elements that HIDE content until clicked (menus, dropdowns, accordions, tabs)
@@ -168,6 +247,7 @@ export function WithDeepAnalysis<T extends Constructor>(Base: T) {
 
           Look for: overflow/ellipsis menus, chevron dropdowns, hamburger menus,
           gear/settings buttons, accordion toggles, tab switches, filter buttons.
+          ${missingHint}
 
           Rules:
           - For repeated icons (same icon on every list row), pick only the FIRST one
@@ -302,53 +382,15 @@ export function WithDeepAnalysis<T extends Constructor>(Base: T) {
             }
           }
 
-          let clickCode: string | null = null;
-          const action = this.explorer.createAction();
-          for (const cmd of el.commands) {
-            if (await action.attempt(cmd, undefined, false)) {
-              clickCode = cmd;
-              break;
-            }
-          }
-          if (!clickCode) {
-            debugLog(`Click failed: ${el.description.slice(0, 80)}`);
-            continue;
-          }
-
-          await new Promise((r) => setTimeout(r, 500));
-
-          let diff: Diff;
-          try {
-            await this.explorer.createAction().capturePageState();
-            const currAR = ActionResult.fromState(this.stateManager.getCurrentState()!);
-            diff = await currAR.diff(previousState);
-            await diff.calculate();
-          } catch (err) {
-            tag('warning').log(`State capture failed after click: ${err instanceof Error ? err.message : err}`);
-            await this._restorePageState(state.url, originalAria);
-            continue;
-          }
-
-          if (diff.urlHasChanged()) {
-            debugLog(`Click navigated to ${this.stateManager.getCurrentState()?.url}`);
-            navigationLinks.push({ code: clickCode, url: this.stateManager.getCurrentState()?.url || '' });
-            await (this as any).navigateTo(state.url);
-            continue;
-          }
-
-          const clickHtmlSize = diff.htmlParts.reduce((sum, p) => sum + p.subtree.length, 0);
-          if (!diff.ariaChanged && clickHtmlSize <= 150) {
-            debugLog(`No changes from: ${el.description.slice(0, 80)}`);
+          const outcome = await this._executeAndAnalyze(el.commands, el.description, state, originalAria, this._summarizeExpanded(expandedSections));
+          if (outcome.status === 'navigated') {
+            navigationLinks.push({ code: outcome.code, url: outcome.url });
             continue;
           }
-
-          const sectionMarkdown = await this._analyzeExpandedAction(clickCode, el.description, diff, this._summarizeExpanded(expandedSections));
-          if (sectionMarkdown) {
-            expandedSections.push(sectionMarkdown);
+          if (outcome.status === 'revealed') {
+            expandedSections.push(outcome.sectionMarkdown);
             debugLog(`Captured section from: ${el.description.slice(0, 80)}`);
           }
-
-          await this._restorePageState(state.url, originalAria);
         } catch (err) {
           tag('warning').log(`Expandable click failed for "${el.description.slice(0, 80)}": ${err instanceof Error ? err.message : err}`);
           try {
@@ -358,6 +400,55 @@ export function WithDeepAnalysis<T extends Constructor>(Base: T) {
       }
     }
 
+    private async _executeAndAnalyze(commands: string[], description: string, state: WebPageState, originalAria: string, alreadyExpanded: string[]): Promise<ExpansionOutcome> {
+      const previousState = ActionResult.fromState(this.stateManager.getCurrentState()!);
+
+      let clickCode: string | null = null;
+      const action = this.explorer.createAction();
+      for (const cmd of commands) {
+        if (await action.attempt(cmd, undefined, false)) {
+          clickCode = cmd;
+          break;
+        }
+      }
+      if (!clickCode) {
+        debugLog(`Click failed: ${description.slice(0, 80)}`);
+        return { status: 'failed' };
+      }
+
+      await new Promise((r) => setTimeout(r, 500));
+
+      let diff: Diff;
+      try {
+        await this.explorer.createAction().capturePageState();
+        const currAR = ActionResult.fromState(this.stateManager.getCurrentState()!);
+        diff = await currAR.diff(previousState);
+        await diff.calculate();
+      } catch (err) {
+        tag('warning').log(`State capture failed after click: ${err instanceof Error ? err.message : err}`);
+        await this._restorePageState(state.url, originalAria);
+        return { status: 'failed' };
+      }
+
+      if (diff.urlHasChanged()) {
+        const url = this.stateManager.getCurrentState()?.url || '';
+        debugLog(`Click navigated to ${url}`);
+        await (this as any).navigateTo(state.url);
+        return { status: 'navigated', code: clickCode, url };
+      }
+
+      const clickHtmlSize = diff.htmlParts.reduce((sum, p) => sum + p.subtree.length, 0);
+      if (!diff.ariaChanged && clickHtmlSize <= 150) {
+        debugLog(`No changes from: ${description.slice(0, 80)}`);
+        return { status: 'none', code: clickCode };
+      }
+
+      const sectionMarkdown = await this._analyzeExpandedAction(clickCode, description, diff, alreadyExpanded);
+      await this._restorePageState(state.url, originalAria);
+      if (!sectionMarkdown) return { status: 'none', code: clickCode };
+      return { status: 'revealed', code: clickCode, sectionMarkdown };
+    }
+
     private async _restorePageState(url: string, originalAria: string): Promise<void> {
       try {
         await (this as any).cancelInUi();
@@ -484,6 +575,13 @@ interface ExpandableElement extends ResearchElement {
   container: string | null;
 }
 
+interface PreviousSection {
+  name: string;
+  code: string;
+}
+
+type ExpansionOutcome = { status: 'revealed'; code: string; sectionMarkdown: string } | { status: 'navigated'; code: string; url: string } | { status: 'none'; code: string } | { status: 'failed' };
+
 export interface DeepAnalysisMethods {
   performDeepAnalysis(state: WebPageState, result: ResearchResult): Promise<void>;
   researchOverlay(current: ActionResult, previous: ActionResult, pageStateHash: string): Promise<string | null>;

package/src/ai/researcher/locators.ts

@@ -80,7 +80,7 @@ export function WithLocators<T extends Constructor>(Base: T) {
         }
       }
 
-      tag('substep').log(`Validated ${locators.length} locators: ${locators.length - broken} valid, ${broken} broken`);
+      tag('operation').log(`Validated ${locators.length} locators: ${locators.length - broken} valid, ${broken} broken`);
     }
 
     async fixBrokenSections(result: ResearchResult, conversation: Conversation): Promise<void> {

package/src/ai/researcher.ts

@@ -151,7 +151,7 @@ export class Researcher extends ResearcherBase implements Agent {
       if (!deep && !force) {
         const similar = await findSimilarResearch(combinedHtml);
         if (similar) {
-          tag('substep').log('Similar research found, reusing cached result');
+          tag('operation').log('Similar research found, reusing cached result');
           if (stateHash) saveResearch(stateHash, similar, combinedHtml);
           tag('multiline').log(formatResearchSummary(similar));
           tag('success').log('Research complete (reused)');
@@ -316,10 +316,10 @@ export class Researcher extends ResearcherBase implements Agent {
 
       tag('multiline').log(formatResearchSummary(result.text, { visionUsed: this.hasScreenshotToAnalyze }));
       tag('success').log('Research complete');
-      if (researchFile) tag('substep').log(`Research file saved to: ${researchFile}`);
+      if (researchFile) tag('operation').log(`Research file saved to: ${researchFile}`);
       if (this.actionResult?.screenshotFile) {
         const screenshotPath = outputPath('states', this.actionResult.screenshotFile);
-        tag('substep').log(`UI screenshot: file://${screenshotPath}`);
+        tag('operation').log(`UI screenshot: file://${screenshotPath}`);
       }
 
       await this.hooksRunner.runAfterHook('researcher', state.url);
@@ -467,7 +467,7 @@ export class Researcher extends ResearcherBase implements Agent {
         .filter((k) => !!k)
         .join('\n\n');
 
-      tag('substep').log(`Found ${knowledgeFiles.length} relevant knowledge ${pluralize(knowledgeFiles.length, 'file')} for: ${this.actionResult.url}`);
+      tag('operation').log(`Found ${knowledgeFiles.length} relevant knowledge ${pluralize(knowledgeFiles.length, 'file')} for: ${this.actionResult.url}`);
       knowledge = `
         <hint>
         Here is relevant knowledge for this page:

package/src/ai/session-analyst.ts

@@ -41,6 +41,9 @@ export class SessionAnalyst implements Agent {
 
       Crucial distinction: "the app misbehaved" vs "the automation could not interact with the app". ONLY the first is a Defect. If the automation gives up before the app responds — timeout, retries exhausted, dead loop / loop detected, could not click or find an element — that is an Execution issue regardless of what the log calls it. Failure inside the automation ≠ failure inside the product.
 
+      The action log is more authoritative than the scenario title. If the actual submitted data, page state, or action sequence does not match the scenario title, classify it as Execution issue and do not list that scenario under What works. Do NOT infer a product Defect or UX issue from behavior caused by incorrect test data or an automation mismatch.
+      Negative test data is valid when it matches a negative scenario. Do not call intentionally invalid input wrong data when the scenario expects rejection or validation feedback.
+
       A solitary failure where adjacent tests on the same feature passed → Execution, not Defect.
 
       ## Severity (defects only)
@@ -76,7 +79,7 @@ export class SessionAnalyst implements Agent {
 
       ## Brevity rules
 
-      - Headline: 2 sentences MAX. About the FEATURE, not the run. No counts, no "N tests", no "this session". Banned words: "exercised", "comprehensive", "notably", "this session", "module", "targeted", "covered creation".
+      - Headline: 2 sentences MAX. About the FEATURE, not the run. No counts, no "N tests", no "this session". Never use these words: "exercised", "comprehensive", "notably", "this session", "module", "targeted", "covered creation".
       - What works: feature name + test refs. NO parentheticals, NO caveats. If there's a caveat, the entry doesn't belong here.
       - Defect title is the BUG ("Search returns non-matching results"), never the scenario name.
       - Reproduce steps are imperative one-liners drawn from the log.

package/src/ai/task-agent.ts

@@ -44,7 +44,7 @@ export abstract class TaskAgent {
       .filter((k) => !!k)
       .join('\n\n');
 
-    tag('substep').log(`Found ${knowledgeFiles.length} relevant knowledge ${pluralize(knowledgeFiles.length, 'file')}`);
+    tag('operation').log(`Found ${knowledgeFiles.length} relevant knowledge ${pluralize(knowledgeFiles.length, 'file')}`);
     return dedent`
       <knowledge>
       Here is relevant knowledge for this page:
@@ -61,7 +61,7 @@ export abstract class TaskAgent {
 
     const totalSections = toc.reduce((sum, entry) => sum + entry.sections.length, 0);
     debugLog(`injecting experience TOC (${toc.length} files, ${totalSections} sections)`);
-    tag('substep').log(`Found ${toc.length} experience ${pluralize(toc.length, 'file')} (${totalSections} sections)`);
+    tag('operation').log(`Found ${toc.length} experience ${pluralize(toc.length, 'file')} (${totalSections} sections)`);
     return renderExperienceToc(toc);
   }
 

package/src/commands/context-aria-command.ts

@@ -17,6 +17,6 @@ export class ContextAriaCommand extends BaseCommand {
       throw new Error('No ARIA snapshot available for current page');
     }
 
-    tag('multiline').log(`ARIA Snapshot:\n\n${ariaSnapshot}`);
+    tag('multiline').log(`ARIA Snapshot:\n\n${ariaSnapshot}`, { maxLines: 10 });
   }
 }