explorbot 0.0.1 → 0.1.0

package/src/ai/captain/test-mode.ts

@@ -0,0 +1,262 @@
+import { tool } from 'ai';
+import dedent from 'dedent';
+import { z } from 'zod';
+import type { WebPageState } from '../../state-manager.ts';
+import { toolExecutionLabel } from '../conversation.ts';
+import { type Constructor, type ModeContext, debugLog } from './mixin.ts';
+
+export function WithTestMode<T extends Constructor>(Base: T) {
+  return class extends Base {
+    testModeTools(ctx: ModeContext): Record<string, any> {
+      return {
+        test: tool({
+          description: dedent`
+            Inspect test execution data. Use flags to control output depth.
+            Tests are identified by sessionName (e.g. "brave-fox123").
+
+            Examples:
+              test()                                        — summary of all tests in plan (with sessionNames)
+              test --session brave-fox123                   — overview with structured timeline (step, action, success, state, url)
+              test --session brave-fox123 --step 5          — drill into step 5: full commands, code, attempts, error, suggestion
+              test --session brave-fox123 --log             — full chronological execution log
+              test --session brave-fox123 --code            — generated CodeceptJS code
+              test --session brave-fox123 --pilot           — Pilot's last analysis
+          `,
+          inputSchema: z.object({
+            session: z.string().optional().describe('Test sessionName (e.g. "brave-fox123"). Omit for plan summary.'),
+            step: z.number().optional().describe('Drill into step N from timeline — shows full commands, code, attempts, suggestions'),
+            log: z.boolean().optional().describe('Include full execution log'),
+            code: z.boolean().optional().describe('Include generated CodeceptJS code'),
+            pilot: z.boolean().optional().describe('Include Pilot analysis'),
+          }),
+          execute: async ({ session, step, log, code, pilot }) => {
+            const plan = ctx.explorBot.getCurrentPlan();
+            if (!plan || plan.tests.length === 0) return { success: false, message: 'No plan or tests available' };
+
+            if (!session) {
+              return {
+                success: true,
+                planTitle: plan.title,
+                tests: plan.tests.map((t) => ({
+                  session: t.sessionName,
+                  scenario: t.scenario,
+                  status: t.status,
+                  result: t.result,
+                  priority: t.priority,
+                })),
+              };
+            }
+
+            const t = plan.tests.find((test) => test.sessionName === session);
+            if (!t) return { success: false, message: `Test "${session}" not found` };
+
+            const testerConv = ctx.explorBot.agentTester().getConversation();
+
+            if (step !== undefined) {
+              if (!testerConv) return { success: false, message: 'No tester conversation available' };
+              const execs = testerConv.getToolExecutions();
+              if (step < 0 || step >= execs.length) return { success: false, message: `Step ${step} not found (${execs.length} steps available)` };
+              const e = execs[step];
+              return {
+                success: true,
+                step,
+                action: e.toolName,
+                explanation: toolExecutionLabel(e.input),
+                commands: e.input?.commands || null,
+                code: e.output?.code || null,
+                attempts: e.output?.attempts || null,
+                suggestion: e.output?.suggestion || null,
+                wasSuccessful: e.wasSuccessful,
+                error: e.wasSuccessful ? null : e.output?.message || null,
+                ariaChanges: e.output?.pageDiff?.ariaChanges || null,
+                url: e.output?.pageDiff?.currentUrl || null,
+              };
+            }
+
+            const result: Record<string, any> = {
+              success: true,
+              session: t.sessionName,
+              scenario: t.scenario,
+              status: t.status,
+              result: t.result,
+              priority: t.priority,
+              startUrl: t.startUrl,
+              expected: t.expected,
+              notes: t.notesToString(),
+            };
+
+            if (testerConv) {
+              const execs = testerConv.getToolExecutions();
+              let currentStateIdx = -1;
+              let currentUrl = t.startUrl;
+
+              result.timeline = execs.map((e, i) => {
+                const newUrl = e.output?.pageDiff?.currentUrl;
+                if (newUrl && newUrl !== currentUrl) {
+                  currentUrl = newUrl;
+                  currentStateIdx = t.states.findIndex((s, idx) => idx > currentStateIdx && s.url === newUrl);
+                  if (currentStateIdx === -1) currentStateIdx = t.states.length - 1;
+                }
+                const entry: Record<string, any> = {
+                  step: i,
+                  action: e.toolName,
+                  explanation: toolExecutionLabel(e.input),
+                  success: e.wasSuccessful,
+                  url: currentUrl,
+                };
+                if (currentStateIdx >= 0) entry.state = currentStateIdx;
+                if (!e.wasSuccessful) entry.error = e.output?.message || null;
+                if (e.output?.pageDiff?.ariaChanges) entry.ariaChanges = e.output.pageDiff.ariaChanges;
+                return entry;
+              });
+            }
+
+            if (log) result.log = t.getLogString();
+
+            if (code) result.generatedCode = t.generatedCode || 'Not generated yet';
+
+            if (pilot) {
+              result.pilotAnalysis = ctx.explorBot.agentPilot().getLastAnalysis() || 'No analysis available';
+            }
+
+            return result;
+          },
+        }),
+
+        inspectState: tool({
+          description: dedent`
+            Inspect previously visited page states. Shows URL, title, headings, associated files, and changes from previous state.
+            Use bash('cat <file>') to read HTML/log files. Use navigate() to go to the state's URL.
+
+            inspectState()                              — list all states from current page history
+            inspectState --session brave-fox123         — list all states from test session
+            inspectState --session brave-fox123 --index 2 — details of state 2: headings, ARIA, files, diff from state 1
+          `,
+          inputSchema: z.object({
+            session: z.string().optional().describe('Test session name. Uses current page history if omitted.'),
+            index: z.number().optional().describe('State index for detailed inspection'),
+          }),
+          execute: async ({ session, index }) => {
+            let states: WebPageState[];
+
+            if (session) {
+              const plan = ctx.explorBot.getCurrentPlan();
+              if (!plan) return { success: false, message: 'No plan available' };
+              const t = plan.tests.find((test) => test.sessionName === session);
+              if (!t) return { success: false, message: `Test "${session}" not found` };
+              states = t.states;
+            } else {
+              const stateManager = ctx.explorBot.getExplorer().getStateManager();
+              const history = stateManager.getStateHistory();
+              const seen = new Set<string>();
+              states = history
+                .map((t) => t.toState)
+                .filter((s) => {
+                  const key = s.hash || s.url;
+                  if (seen.has(key)) return false;
+                  seen.add(key);
+                  return true;
+                });
+            }
+
+            if (states.length === 0) return { success: true, states: [], message: 'No states recorded' };
+
+            if (index === undefined) {
+              return {
+                success: true,
+                states: states.map((s, i) => ({
+                  index: i,
+                  url: s.url,
+                  title: s.title,
+                  h1: s.h1,
+                })),
+              };
+            }
+
+            if (index < 0 || index >= states.length) {
+              return { success: false, message: `State ${index} not found (${states.length} states available)` };
+            }
+
+            const state = states[index];
+            const prev = index > 0 ? states[index - 1] : null;
+
+            const changes: string[] = [];
+            if (prev) {
+              if (prev.url !== state.url) changes.push(`URL changed from ${prev.url} to ${state.url}`);
+              if (prev.h1 !== state.h1 && state.h1) changes.push(`new H1: ${state.h1}`);
+              if (prev.title !== state.title && state.title) changes.push(`title changed to: ${state.title}`);
+            }
+
+            const files: Record<string, string | null> = {
+              html: state.htmlFile || null,
+              screenshot: state.screenshotFile || null,
+              log: state.logFile || null,
+              aria: state.ariaSnapshotFile || null,
+            };
+
+            const suggestions: string[] = [];
+            if (state.htmlFile) suggestions.push(`Use bash('cat ${state.htmlFile}') to see full HTML.`);
+            if (state.url) suggestions.push(`Use navigate('${state.url}') to go to this page.`);
+
+            debugLog('inspectState index=%d url=%s', index, state.url);
+
+            return {
+              success: true,
+              index,
+              url: state.url,
+              title: state.title,
+              h1: state.h1,
+              h2: state.h2,
+              h3: state.h3,
+              h4: state.h4,
+              ariaSnapshot: state.ariaSnapshot,
+              files,
+              changesFromPrevious: changes.length > 0 ? changes.join(', ') : null,
+              suggestion: suggestions.join(' '),
+            };
+          },
+        }),
+      };
+    }
+
+    testModePrompt(): string {
+      return dedent`
+        <test_capabilities>
+        - Test inspection: test() — timeline, steps, log, code, pilot analysis
+        - State inspection: inspectState() — traverse previously visited page states
+        - Test control: runCommand("/test stop"), runCommand("/test skip")
+        </test_capabilities>
+
+        <debugging_skill>
+        When user asks "why did X fail?" or about test problems:
+
+        1. IDENTIFY — test() without flags to list all tests. Name the failed test.
+        2. OVERVIEW — test --session <name> to read the timeline. Find the step where behavior diverged.
+        3. DRILL — test --session <name> --step N to inspect the failing step details.
+        4. INSPECT STATE — inspectState --session <name> --index N to see page ARIA at that point.
+        5. CONTEXT — Use bash('cat <file>') to read HTML/log files referenced by inspectState.
+        6. REPORT — Explain with specifics: test name, which step failed, what was attempted, your hypothesis.
+
+        Always name the test session and scenario when discussing results.
+        Do NOT call done() after just reading data — investigate first.
+        </debugging_skill>
+      `;
+    }
+
+    testModeRules(): string {
+      return dedent`
+        - Use test() to inspect what the tester is doing
+        - Use inspectState() to explore page states — use bash('cat <file>') for HTML/logs
+        - When debugging test failures, follow <debugging_skill> — do not summarize without investigating
+        - Use test() with minimal flags first, then drill into timeline steps and states as needed
+        - Prefer ARIA over HTML — avoid full HTML reads
+      `;
+    }
+  };
+}
+
+export interface TestModeMethods {
+  testModeTools(ctx: ModeContext): Record<string, any>;
+  testModePrompt(): string;
+  testModeRules(): string;
+}

package/src/ai/captain/web-mode.ts

@@ -0,0 +1,136 @@
+import { tool } from 'ai';
+import dedent from 'dedent';
+import { z } from 'zod';
+import { actionRule, locatorRule, sectionContextRule } from '../rules.ts';
+import { createAgentTools, createCodeceptJSTools } from '../tools.ts';
+import { type Constructor, type ModeContext, debugLog } from './mixin.ts';
+
+export function WithWebMode<T extends Constructor>(Base: T) {
+  return class extends Base {
+    webModeTools(ctx: ModeContext): Record<string, any> {
+      const explorer = ctx.explorBot.getExplorer();
+      const codeceptTools = createCodeceptJSTools(explorer, ctx.task);
+      const agentTools = createAgentTools({
+        explorer,
+        researcher: ctx.explorBot.agentResearcher(),
+        navigator: ctx.explorBot.agentNavigator(),
+      });
+      const { see, context, visualClick } = agentTools;
+
+      return {
+        navigate: tool({
+          description: 'Navigate to a URL or page description using AI-powered navigation.',
+          inputSchema: z.object({
+            destination: z.string().describe('URL path or page description, e.g. "/login", "go to user settings"'),
+          }),
+          execute: async ({ destination }) => {
+            try {
+              debugLog('navigate', destination);
+              await ctx.explorBot.agentNavigator().visit(destination);
+              const stateManager = ctx.explorBot.getExplorer().getStateManager();
+              const state = stateManager.getCurrentState();
+              return { success: true, url: state?.url, title: state?.title };
+            } catch (error: any) {
+              return { success: false, message: `Navigation failed: ${error.message}` };
+            }
+          },
+        }),
+
+        browser: tool({
+          description: dedent`
+            Direct browser access via Playwright. Use for diagnostics and browser management.
+            Actions:
+            - evaluate: Run JavaScript in browser context (localStorage, cookies, DOM, console)
+            - closeTabs: Close all browser tabs except the current one
+            - screenshot: Take a screenshot of current page
+            - reload: Reload the current page
+          `,
+          inputSchema: z.object({
+            action: z.enum(['evaluate', 'closeTabs', 'screenshot', 'reload']).describe('Browser action to perform'),
+            code: z.string().optional().describe('JavaScript code for evaluate action'),
+          }),
+          execute: async ({ action, code }) => {
+            const page = ctx.explorBot.getExplorer().playwrightHelper?.page;
+            if (!page) return { success: false, message: 'No browser page available' };
+
+            if (action === 'evaluate') {
+              if (!code) return { success: false, message: 'Code required for evaluate action' };
+              try {
+                const result = await page.evaluate(code);
+                const serialized = typeof result === 'object' ? JSON.stringify(result, null, 2) : String(result ?? '');
+                return { success: true, result: serialized.length > 2000 ? `${serialized.slice(0, 2000)}\n...truncated` : serialized };
+              } catch (error: any) {
+                return { success: false, message: `Evaluate failed: ${error.message}` };
+              }
+            }
+
+            if (action === 'closeTabs') {
+              const browserContext = page.context();
+              const pages = browserContext.pages();
+              let closed = 0;
+              for (const p of pages) {
+                if (p !== page) {
+                  await p.close();
+                  closed++;
+                }
+              }
+              return { success: true, message: `Closed ${closed} tab(s), ${pages.length - closed} remaining` };
+            }
+
+            if (action === 'screenshot') {
+              const buffer = await page.screenshot({ type: 'png' });
+              const base64 = buffer.toString('base64');
+              return { success: true, screenshot: base64.length > 5000 ? `${base64.slice(0, 5000)}...truncated` : base64 };
+            }
+
+            if (action === 'reload') {
+              await page.reload();
+              return { success: true, message: 'Page reloaded' };
+            }
+
+            return { success: false, message: `Unknown action: ${action}` };
+          },
+        }),
+
+        ...codeceptTools,
+        see,
+        context,
+        visualClick,
+      };
+    }
+
+    webModePrompt(): string {
+      return dedent`
+        <web_capabilities>
+        - Page actions: click, pressKey, form (CodeceptJS tools)
+        - Navigation: navigate() — AI-powered navigation to URLs or page descriptions
+        - Browser diagnostics: browser() — evaluate JS, close tabs, screenshot, reload
+        - Visual analysis: see() — screenshot-based page verification
+        - Context refresh: context() — get fresh HTML/ARIA snapshot
+        - Visual fallback: visualClick() — coordinate-based click when locators fail
+        </web_capabilities>
+
+        ${locatorRule}
+
+        ${actionRule}
+
+        ${sectionContextRule}
+      `;
+    }
+
+    webModeRules(): string {
+      return dedent`
+        - Follow <locator_priority> rules when selecting locators for all tools
+        - click() accepts array of commands to try in order — include ARIA, CSS, XPath variants
+        - If click() fails with all provided commands, use visualClick() tool as fallback
+        - When an action fails or result is unexpected, investigate or inform the user instead of retrying silently
+      `;
+    }
+  };
+}
+
+export interface WebModeMethods {
+  webModeTools(ctx: ModeContext): Record<string, any>;
+  webModePrompt(): string;
+  webModeRules(): string;
+}