@tangle-network/browser-agent-driver 0.23.0 → 0.24.1

package/README.md

@@ -1,18 +1,40 @@
 # @tangle-network/browser-agent-driver
 
-LLM-driven browser automation. Reads page state via accessibility tree, decides actions via LLM, executes in a loop until the goal is done.
-
-90% pass rate on WebBench-50. Default model: `gpt-5.4`.
+General-purpose agentic browser automation. Completes real user outcomes on any website — search, extract, fill forms, compare prices, navigate complex UIs.
+
+**91.3% on WebVoyager (590 tasks, 15 sites) at $0.09/task.** 100% on held-out competitive bench. 95.7% on WebbBench-50 (excl. DataDome sites). Default model: `gpt-5.4`.
+
+## Table of Contents
+
+- [Install](#install)
+- [Quick Start](#quick-start)
+- [How It Works](#how-it-works)
+- [Features](#features)
+  - [Vision + DOM Hybrid](#vision--dom-hybrid)
+  - [Stealth & Anti-Bot](#stealth--anti-bot)
+  - [CAPTCHA Solving](#captcha-solving)
+  - [Parallel Tab Execution](#parallel-tab-execution)
+  - [Multi-Model Orchestration](#multi-model-orchestration)
+  - [Design Audit](#design-audit)
+  - [Wallet & DeFi Testing](#wallet--defi-testing)
+- [Configuration](#configuration)
+- [CLI Reference](#cli-reference)
+- [SDK / Library Usage](#sdk--library-usage)
+- [Test Suites](#test-suites)
+- [Drivers](#drivers)
+- [Benchmarks](#benchmarks)
+- [GitHub Action](#github-action)
+- [Development](#development)
 
 ## Install
 
-### CLI
+### CLI (recommended)
 
 ```bash
 curl -fsSL https://raw.githubusercontent.com/tangle-network/browser-agent-driver/main/scripts/install.sh | sh
 ```
 
-Installs the `bad` command to `~/.local/bin`, downloads Playwright Chromium, and adds PATH instructions. Requires Node.js 20+.
+Installs the `bad` command, downloads Chromium, adds PATH. Requires Node.js 20+.
 
 Or via npm:
 
@@ -30,167 +52,389 @@ pnpm add -D playwright
 
 ## Quick Start
 
-### Programmatic
+### CLI
+
+```bash
+# Run a task
+bad run --goal "Find the cheapest flight from NYC to London on Jan 15" \
+  --url https://www.google.com/travel/flights
+
+# With vision (screenshot-based decisions)
+bad run --goal "Compare MacBook Air prices" --url https://apple.com \
+  --observation-mode hybrid
+
+# Test suite from case file
+bad run --cases ./my-tests.json --concurrency 4
+
+# Authenticated session
+bad run --goal "Check account settings" --url https://app.example.com \
+  --storage-state .auth/session.json
+
+# With proxy (residential or SOCKS5)
+bad run --goal "Search hotels in Tokyo" --url https://booking.com \
+  --proxy http://user:pass@proxy.example.com:port
+```
+
+### SDK
 
 ```typescript
 import { chromium } from 'playwright'
 import { PlaywrightDriver, BrowserAgent } from '@tangle-network/browser-agent-driver'
 
-const browser = await chromium.launch()
+const browser = await chromium.launch({ channel: 'chrome' })
 const page = await browser.newPage()
 const driver = new PlaywrightDriver(page)
 
-const runner = new BrowserAgent({
+const agent = new BrowserAgent({
   driver,
-  config: { model: 'gpt-5.4' },
+  config: {
+    model: 'gpt-5.4',
+    observationMode: 'hybrid',    // vision + DOM
+    plannerEnabled: true,          // plan-then-execute
+  },
 })
 
-const result = await runner.run({
-  goal: 'Sign in and navigate to settings',
-  startUrl: 'https://app.example.com',
-  maxTurns: 30,
+const result = await agent.run({
+  goal: 'Find the top 3 trending repositories on GitHub',
+  startUrl: 'https://github.com/trending',
 })
 
-console.log(result.success, `${result.turns.length} turns`)
+console.log(result.success, result.reason)
 await browser.close()
 ```
 
-### CLI
-
-```bash
-# single task
-bad run --goal "Sign up for account" --url http://localhost:3000
-
-# test suite from case file
-bad run --cases ./cases.json
-
-# authenticated session
-bad run --goal "Open settings" --url https://app.example.com \
-  --storage-state ./.auth/session.json
-
-# speed-optimized mode
-bad run --cases ./cases.json --mode fast-explore
-
-# evidence-rich mode for signoff
-bad run --cases ./cases.json --mode full-evidence
-```
-
-## Config File
+### Config File
 
-Create `browser-agent-driver.config.ts` in your project root:
+Create `bad.config.ts` (or `.js`, `.mjs`) in your project root:
 
 ```typescript
 import { defineConfig } from '@tangle-network/browser-agent-driver'
 
 export default defineConfig({
+  provider: 'openai',
   model: 'gpt-5.4',
+  observationMode: 'hybrid',
+  plannerEnabled: true,
   headless: true,
   concurrency: 4,
   maxTurns: 30,
-  timeoutMs: 300_000,
   outputDir: './test-results',
-  reporters: ['junit', 'html'],
+
+  // Per-role model routing (Gen 28)
+  models: {
+    planner:    { model: 'claude-opus-4-6', provider: 'anthropic' },
+    executor:   { model: 'gpt-4.1-mini' },
+    verifier:   { model: 'gpt-4.1-mini' },
+    supervisor: { model: 'gpt-5.4' },
+  },
+
+  // Parallel tabs for compound goals (Gen 21)
+  parallelTabs: { enabled: true, maxTabs: 3 },
+
+  // Proxy for anti-bot bypass
+  proxy: 'http://user:pass@proxy:port',  // or set BAD_PROXY_URL env
+
+  // Supervisor for stuck detection
+  supervisor: { enabled: true, useVision: true },
 })
 ```
 
-Auto-detected by CLI and programmatic API. CLI flags override config values. Supports `.ts`, `.js`, `.mjs`.
+Auto-detected by CLI and SDK. CLI flags override config values.
 
-## Test Suites
+## How It Works
+
+```
+Goal → DOM Planner (1 LLM call with screenshot)
+  → N deterministic steps (click, type, fill, navigate)
+  → If deviation: vision per-action loop
+    → Observe (a11y tree + screenshot + SoM labels)
+    → LLM decides action (ref-based, coordinate, or label)
+    → Execute + verify expected effect
+    → Recovery on failure (strategy shift, form reset retry, search fallback)
+  → Goal verification (LLM or fast-path with script evidence)
+  → Complete with structured result
+```
+
+Recovery is automatic: cookie consent, modal blockers, A-B-A-B oscillation loops, form field resets, date picker stalls, and CAPTCHA challenges are handled before the agent continues.
+
+## Features
+
+### Vision + DOM Hybrid
+
+The agent sees BOTH a screenshot and the accessibility tree. Screenshots show visual layout, SoM labels mark interactive elements with numbered badges, and the a11y tree provides precise `@ref` IDs for targeting.
+
+Three observation modes:
+- **`dom`** — a11y tree only (fastest, cheapest)
+- **`vision`** — screenshot + coordinates only
+- **`hybrid`** — both (default for benchmarks, most reliable)
+
+### Stealth & Anti-Bot
+
+Built-in evasion for Cloudflare, Akamai, and most WAF systems:
+
+- **System Chrome** (`channel: 'chrome'`) — real TLS/JA3/HTTP2 fingerprint, not bundled Chromium
+- **Patchright** — Playwright fork that patches CDP protocol leaks
+- **Mouse humanization** — Bezier curve movement (8-15 points), gaussian click offset
+- **Browser fingerprint** — navigator.webdriver, plugins, languages, WebGL, canvas noise, screenX/Y fix
+- **GPU rendering** — `--use-gl=desktop` for real WebGL fingerprint
+- **Resource blocking** — 99+ analytics/tracking domains blocked
+
+Unblocks 9/13 previously-blocked sites on WebbBench-50 with zero configuration.
+
+### CAPTCHA Solving
+
+Automatic CAPTCHA detection and solving during runs:
+
+- **reCAPTCHA v2** — checkbox click + image grid solver (LLM vision-based)
+- **Cloudflare Turnstile** — checkbox behavioral click
+- **Google "unusual traffic"** — detected and solver attempted
 
 ```typescript
-import { TestRunner } from '@tangle-network/browser-agent-driver'
+// Enabled by default. To configure:
+{ captcha: { enabled: true, maxAttempts: 5 } }
+```
 
-const suite = await runner.runSuite([
-  {
-    id: 'login',
-    name: 'User login flow',
-    startUrl: 'https://app.example.com/login',
-    goal: 'Log in with test credentials',
-    successCriteria: [
-      { type: 'url-contains', value: '/dashboard' },
-      { type: 'element-visible', selector: '[data-testid="user-menu"]' },
-    ],
-  },
-])
+### Parallel Tab Execution
+
+For compound goals ("compare X vs Y", "find 5 items matching criteria"), the agent decomposes the goal and runs sub-tasks in parallel browser tabs.
+
+```typescript
+{
+  parallelTabs: { enabled: true, maxTabs: 3 },
+}
 ```
 
-## Actions
+The GoalDecomposer (1 cheap LLM call) classifies goals as simple or compound. Simple goals run as before. Compound goals get split into sub-goals, each running in its own tab via `Promise.all`, with results merged by the EvidenceMerger.
 
-The LLM can perform: `click`, `type`, `press`, `hover`, `select`, `scroll`, `navigate`, `wait`, `evaluate`, `verifyPreview`, `complete`, `abort`.
+### Multi-Model Orchestration
 
-## How It Works
+Different agent roles can use different models for optimal cost/quality:
 
-Each turn: observe page (a11y tree + optional screenshot) → LLM decides action → execute → verify effect → repeat.
+```typescript
+{
+  models: {
+    planner:    { model: 'claude-opus-4-6', provider: 'anthropic' },  // best reasoning
+    executor:   { model: 'gpt-4.1-mini' },                            // cheap, follows plans
+    verifier:   { model: 'gpt-4.1-mini' },                            // structured yes/no
+    supervisor: { model: 'gpt-5.4' },                                 // strategic recovery
+  },
+}
+```
 
-Recovery is automatic: cookie consent, modal blockers, stuck loops (A-B-A-B oscillation), and selector failures are handled before the agent continues.
+Each role falls back to the main `model` when not configured. The planner needs top-tier reasoning; the executor just follows instructions.
 
-## Design Audit
+### Design Audit
 
-`bad design-audit` is a vision-powered design quality analyzer with a closed-loop improvement mode. It auto-classifies the page, runs ground-truth measurements (axe-core + WCAG contrast math), then evaluates visual quality with a composable rubric — and ranks the top fixes by ROI.
+Vision-powered design quality analysis with closed-loop improvement:
 
 ```bash
-# Audit any URL — auto-classifies, no profile needed
+# Audit any URL
 bad design-audit --url https://your-app.com
 
 # Multi-page crawl with cross-page systemic detection
 bad design-audit --url https://your-app.com --pages 10
 
-# Closed-loop fix: dispatch findings to a coding agent that edits source files
-bad design-audit --url http://localhost:3000 \
-  --evolve claude-code \
-  --project-dir ~/my-app
+# Auto-fix: dispatch findings to a coding agent
+bad design-audit --url http://localhost:3000 --evolve claude-code --project-dir ~/my-app
+```
+
+Reports rank **Top Fixes by ROI** — the highest-leverage changes scored by `(impact * blast / effort)`. Multi-page systemic findings collapse automatically.
 
-# Other evolve modes: codex, opencode, css (browser injection), or any custom CLI
-bad design-audit --url http://localhost:3000 --evolve "aider --message"
+### Wallet & DeFi Testing
 
-# Pure DOM token extraction (no LLM)
-bad design-audit --url https://your-app.com --extract-tokens
+Built-in MetaMask integration for DeFi app testing:
+
+```bash
+pnpm wallet:setup      # download MetaMask
+pnpm wallet:onboard    # automate first-run wizard
+pnpm wallet:anvil      # start Anvil mainnet fork (100 ETH + 10 WETH + 10k USDC)
+pnpm wallet:validate   # run wallet test suite
 ```
 
-Reports open with **Top Fixes (by ROI)** — the 5 highest-leverage fixes ranked by `(impact × blast / effort)`. Findings appearing on multiple pages collapse into systemic findings. Verified end-to-end: a deliberately-bad fixture went 3.0 → 5.0 (+2.0) over 2 evolve rounds with claude-code rewriting actual source files.
+Supports connect, swap, supply workflows across Uniswap, Aave, SushiSwap, 1inch.
 
-See [Design Audit Guide](./docs/guides/design-audit.md) for the full pipeline, custom rubric fragments, and starter-foundry integration.
+## Configuration
 
-## Session Viewer
+| Option | Type | Default | Description |
+|--------|------|---------|-------------|
+| `provider` | string | `'openai'` | LLM provider |
+| `model` | string | `'gpt-5.4'` | LLM model |
+| `observationMode` | string | `'dom'` | `'dom'`, `'vision'`, or `'hybrid'` |
+| `plannerEnabled` | boolean | `false` | Plan-then-execute mode |
+| `headless` | boolean | `true` | Run browser headless |
+| `maxTurns` | number | `30` | Max turns per task |
+| `proxy` | string | — | Proxy URL (also `BAD_PROXY_URL` env) |
+| `models` | object | — | Per-role model overrides |
+| `parallelTabs` | object | — | `{ enabled, maxTabs }` |
+| `supervisor` | object | — | `{ enabled, useVision, model }` |
+| `captcha` | object | — | `{ enabled, maxAttempts }` |
+| `goalVerification` | boolean | `true` | Verify goal before accepting |
+| `vision` | boolean | `true` | Send screenshots to LLM |
+| `concurrency` | number | `1` | Parallel test cases |
+
+See [Configuration Reference](./docs/guides/configuration.md) for all options.
 
-`bad view` opens any run in a polished web UI:
+## CLI Reference
 
 ```bash
-bad view audit-results/stripe.com-1775502457141
+bad run [options]              # Run tasks
+bad design-audit [options]     # Design quality analysis
+bad view <run-dir>             # Open session viewer
+bad competitive [options]      # Head-to-head framework comparison
 ```
 
-- Sidebar lists every page (or turn) in the run
-- Top Fixes section opens by default for design audits, ranked by ROI
-- Per-page screenshots, design system breakdown, findings table, classification
-- Per-turn action JSON, reasoning, expected effect, result for agent runs
-- Self-contained — no build pipeline, single static HTML, no external dependencies (the viewer is served by a local loopback HTTP server on port 7777)
+### Run options
+
+| Flag | Description |
+|------|-------------|
+| `--goal "..."` | Natural language goal |
+| `--url URL` | Starting URL |
+| `--cases file.json` | Test case file |
+| `--mode fast-explore\|full-evidence` | Run mode |
+| `--model MODEL` | LLM model |
+| `--provider openai\|anthropic\|google` | LLM provider |
+| `--observation-mode dom\|vision\|hybrid` | Observation mode |
+| `--headless` | Run headless (default) |
+| `--proxy URL` | Residential/SOCKS5/HTTP proxy |
+| `--profile default\|stealth\|benchmark-webvoyager` | Launch profile |
+| `--concurrency N` | Parallel cases |
+| `--show-cursor` | Animated cursor overlay in screenshots |
+| `--live` | Real-time SSE viewer |
+
+## SDK / Library Usage
+
+```typescript
+import {
+  // Core
+  BrowserAgent,
+  PlaywrightDriver,
+  SteelDriver,
+  TestRunner,
+
+  // Config
+  defineConfig,
+
+  // Types
+  type AgentConfig,
+  type Scenario,
+  type AgentResult,
+  type Turn,
+
+  // Multi-actor (parallel users)
+  MultiActorSession,
+
+  // Design audit
+  runDesignAudit,
+
+  // CAPTCHA
+  detectCaptcha,
+  solveCaptcha,
+} from '@tangle-network/browser-agent-driver'
+```
+
+### Multi-Actor Sessions
+
+```typescript
+import { MultiActorSession } from '@tangle-network/browser-agent-driver'
+
+const session = await MultiActorSession.create(browser, {
+  actors: {
+    admin:   { storageState: '.auth/admin.json' },
+    user:    {},
+  },
+  agentConfig: { model: 'gpt-5.4', observationMode: 'hybrid' },
+})
+
+// Sequential
+await session.actor('admin').run({ goal: 'Create project', startUrl: '/admin' })
 
-Pair with `--show-cursor` to record runs with an animated cursor + element highlights overlaid on every screenshot.
+// Parallel
+await session.parallel(
+  ['admin', { goal: 'Monitor dashboard', startUrl: '/admin' }],
+  ['user',  { goal: 'Submit form', startUrl: '/app' }],
+)
 
-## Drivers — local, remote, and managed
+await session.close()
+```
 
-bad's agent loop is decoupled from the browser layer via the `Driver` interface. The default is local Playwright, but you can run the same agent against managed cloud infra without any code changes:
+## Test Suites
 
 ```typescript
-import { BrowserAgent, SteelDriver } from '@tangle-network/browser-agent-driver'
+import { TestRunner } from '@tangle-network/browser-agent-driver'
+
+const runner = new TestRunner({
+  driver,
+  config: { model: 'gpt-5.4', observationMode: 'hybrid' },
+  concurrency: 4,
+})
 
-// Local Playwright (default) — see Quick Start above
+const results = await runner.runSuite([
+  {
+    id: 'login',
+    goal: 'Log in with test@example.com / password123',
+    startUrl: 'https://app.example.com/login',
+  },
+  {
+    id: 'create-project',
+    goal: 'Create a new project called "Test Project"',
+    startUrl: 'https://app.example.com/projects',
+  },
+])
+```
 
-// Steel cloud browser with anti-bot, residential proxies, CAPTCHA solving
+## Drivers
+
+The agent loop is decoupled from the browser via the `Driver` interface:
+
+```typescript
+// Local Playwright (default)
+const driver = new PlaywrightDriver(page)
+
+// Steel cloud browser (anti-bot, residential proxies, CAPTCHA)
 const driver = await SteelDriver.create({
   apiKey: process.env.STEEL_API_KEY,
   sessionOptions: { useProxy: true, solveCaptcha: true },
 })
-const agent = new BrowserAgent({ driver, config: { model: 'sonnet' } })
-await agent.run({ goal: '...', startUrl: '...' })
-await driver.close()
+
+// Any Driver implementation works
+const agent = new BrowserAgent({ driver, config })
 ```
 
-The same agent — design audit, evolve loops, wallet automation, knowledge memory — runs against any driver. Steel handles infra you don't want to build; bad handles the agent layer Steel doesn't.
+## Benchmarks
 
-## GitHub Action
+| Benchmark | Score | Cost | Notes |
+|-----------|-------|------|-------|
+| **WebVoyager** (590 tasks, 15 sites) | 91.3% | $0.09/task | Full run, Gen 25 |
+| **Competitive** (10 real-web sites) | 100% | $0.03/task | Held-out, never optimized |
+| **WebbBench-50** (50 diverse sites) | 88% raw, 95.7% excl. DataDome | — | Held-out generalization |
+
+### Competitive position
 
-Drop bad design-audit into any PR pipeline:
+| Agent | WebVoyager | Cost/task |
+|-------|-----------|-----------|
+| Surfer-2 | 97.1% | $1-5 |
+| Magnitude | 93.9% | ~$0.10 |
+| **bad** | **91.3%** | **$0.09** |
+| OpenAI Operator | 87% | ChatGPT Pro |
+
+### Running benchmarks
+
+```bash
+# WebVoyager full 590
+node scripts/run-scenario-track.mjs \
+  --cases bench/external/webvoyager/cases.json \
+  --config bench/scenarios/configs/vision-hybrid.mjs \
+  --model gpt-5.4 --modes fast-explore --concurrency 5 \
+  --out agent-results/webvoyager-full
+
+# Multi-rep validation (≥3 reps required for any claim)
+node scripts/run-multi-rep.mjs \
+  --cases bench/scenarios/cases/my-cases.json \
+  --config bench/scenarios/configs/vision-hybrid.mjs \
+  --reps 3 --out agent-results/my-validation
+```
+
+## GitHub Action
 
 ```yaml
 - uses: tangle-network/browser-agent-driver/.github/actions/design-audit@main
@@ -198,57 +442,44 @@ Drop bad design-audit into any PR pipeline:
     url: ${{ steps.deploy.outputs.preview_url }}
     pages: 5
     fail-on-score-below: '6.5'
-    evolve: claude-code   # optional auto-fix
+    evolve: claude-code
     openai-api-key: ${{ secrets.OPENAI_API_KEY }}
 ```
 
-The action posts the **Top Fixes (by ROI)** as a PR comment, uploads the full report as a workflow artifact, and optionally fails the build on score regressions or critical findings. See [`.github/actions/design-audit`](./.github/actions/design-audit/).
+Posts Top Fixes as a PR comment, uploads full report as artifact, optionally fails on score regressions.
 
-## Guides
-
-- [Configuration Reference](./docs/guides/configuration.md) — all config options
-- [CLI Reference](./docs/guides/cli.md) — commands, modes, profiles, auth
-- [Design Audit](./docs/guides/design-audit.md) — vision-powered design quality + ROI-ranked closed-loop improvement
-- [Memory System](./docs/guides/memory.md) — trajectory store, app knowledge, selector cache
-- [Benchmarks & Experiments](./docs/guides/benchmarks.md) — tiered gates, AB specs, research cycles
-- [Wallet & EVM Apps](./docs/guides/wallet.md) — MetaMask, DeFi testing, RPC interception, Anvil forks
-- [Providers](./docs/guides/providers.md) — OpenAI, Anthropic, Codex CLI, Claude Code, sandbox backend
-- [Reporters & Sinks](./docs/guides/reporters.md) — JUnit, HTML, webhooks, custom sinks
-- [Custom Drivers](./docs/guides/custom-drivers.md) — implement the `Driver` interface
-
-## Research
-
-- [Operating Roadmap](./docs/roadmap/browser-agent-ops.md)
-- [Competitive Analysis](./docs/research/competitor-analysis-2026-03.md)
-- [Reliability Runbook](./RELIABILITY.md)
-
-## Skills
-
-Ships Codex skills under `skills/` for test execution discipline and agent-friendly UX conventions.
+## Session Viewer
 
 ```bash
-npm run skills:install
+bad view audit-results/stripe.com-1775502457141
 ```
 
-## Publishing
-
-Versioning and releases are automated via [Changesets](https://github.com/changesets/changesets).
+Web UI with per-turn screenshots, action JSON, reasoning, element highlights. Pair with `--show-cursor` for animated cursor recordings.
 
-**Contributors:** add a changeset to your PR with `pnpm changeset` — pick patch / minor / major and write a one-line summary. The CLI creates a markdown file under `.changeset/` to commit alongside your code.
-
-**Maintainers:** when PRs with changesets merge to `main`, the [changesets workflow](./.github/workflows/changesets.yml) automatically opens (or updates) a "Release: version packages" PR that bumps `package.json` and writes `CHANGELOG.md`. Merging that PR pushes a `browser-agent-driver-vX.Y.Z` git tag, which fires the existing [`release.yml`](./.github/workflows/release.yml) and [`publish-npm.yml`](./.github/workflows/publish-npm.yml) workflows that create the GitHub release tarball and publish to npm with provenance.
+## Guides
 
-You stay in control of *when* releases ship; the bump math, changelog, tagging, and publishing are all automated. See [`.changeset/README.md`](./.changeset/README.md) for the full contributor flow.
+- [Configuration Reference](./docs/guides/configuration.md)
+- [CLI Reference](./docs/guides/cli.md)
+- [Design Audit](./docs/guides/design-audit.md)
+- [Memory System](./docs/guides/memory.md)
+- [Benchmarks & Experiments](./docs/guides/benchmarks.md)
+- [Wallet & EVM Apps](./docs/guides/wallet.md)
+- [Providers](./docs/guides/providers.md)
+- [Custom Drivers](./docs/guides/custom-drivers.md)
 
 ## Development
 
 ```bash
-pnpm build          # TypeScript → dist/
-pnpm test           # vitest
-pnpm lint           # type-check
-pnpm check:boundaries
+pnpm build              # TypeScript → dist/
+pnpm test               # 993 tests
+pnpm lint               # type-check
+pnpm check:boundaries   # architecture boundaries
 ```
 
+## Publishing
+
+Automated via [Changesets](https://github.com/changesets/changesets) + OIDC trusted publishing. Add a changeset with `pnpm changeset`, merge the auto-generated release PR, and npm publish fires automatically with provenance attestation.
+
 ## License
 
 Dual-licensed under MIT and Apache 2.0. See [LICENSE](./LICENSE).

package/dist/brain/index.d.ts

@@ -67,6 +67,12 @@ export declare class Brain {
     private scoutModelName?;
     private scoutProvider?;
     private scoutUseVision;
+    private plannerModel?;
+    private plannerProvider?;
+    private verifierModel?;
+    private verifierProvider?;
+    private supervisorModel?;
+    private supervisorProvider?;
     private sandboxBackendType?;
     private sandboxBackendProfile?;
     private sandboxBackendProvider?;

package/dist/brain/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/brain/index.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,KAAK,EAAE,YAAY,EAAE,aAAa,EAAsB,MAAM,IAAI,CAAC;AAC1E,OAAO,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,WAAW,EAAE,aAAa,EAAE,gBAAgB,EAAE,IAAI,EAAY,MAAM,aAAa,CAAC;AAuXnH,MAAM,WAAW,aAAa;IAC5B,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,EAAE,CAAC;IACvB,GAAG,EAAE,MAAM,CAAC;IACZ,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB;;;;;OAKG;IACH,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B,6EAA6E;IAC7E,wBAAwB,CAAC,EAAE,MAAM,CAAC;IAClC,kFAAkF;IAClF,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,iBAAiB;IAChC,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,GAAG,EAAE,MAAM,CAAC;IACZ,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,uBAAuB;IACtC,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,GAAG,EAAE,MAAM,CAAC;IACZ,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAKD,qBAAa,KAAK;IAChB,OAAO,CAAC,UAAU,CAAoC;IACtD,OAAO,CAAC,QAAQ,CAA0G;IAC1H,OAAO,CAAC,SAAS,CAAS;IAC1B,OAAO,CAAC,oBAAoB,CAAU;IACtC,OAAO,CAAC,YAAY,CAAC,CAAS;IAC9B,OAAO,CAAC,WAAW,CAAC,CAA0G;IAC9H,OAAO,CAAC,cAAc,CAAC,CAAS;IAChC,OAAO,CAAC,OAAO,CAAC,CAAS;IACzB,OAAO,CAAC,KAAK,CAAU;IACvB,OAAO,CAAC,OAAO,CAAsB;IACrC,OAAO,CAAC,eAAe,CAAS;IAChC,OAAO,CAAC,aAAa,CAAU;IAC/B,OAAO,CAAC,cAAc,CAA8B;IACpD,OAAO,CAAC,eAAe,CAA8B;IACrD,OAAO,CAAC,YAAY,CAAS;IAC7B,OAAO,CAAC,gBAAgB,CAAU;IAClC,OAAO,CAAC,eAAe,CAAC,CAAS;IACjC,OAAO,CAAC,YAAY,CAAS;IAC7B,OAAO,CAAC,cAAc,CAAC,CAAS;IAChC,OAAO,CAAC,aAAa,CAAC,CAA0G;IAChI,OAAO,CAAC,cAAc,CAAU;IAChC,OAAO,CAAC,kBAAkB,CAAC,CAAS;IACpC,OAAO,CAAC,qBAAqB,CAAC,CAAS;IACvC,OAAO,CAAC,sBAAsB,CAAC,CAAS;IAGxC,OAAO,CAAC,cAAc,CAAC,CAAgF;IACvG,OAAO,CAAC,oBAAoB,CAAC,CAA0C;gBAE3D,MAAM,GAAE,WAAgB;IAyBpC,OAAO,CAAC,gBAAgB;IASxB,OAAO,CAAC,qBAAqB;IAK7B,OAAO,CAAC,iBAAiB;IAoBzB,oGAAoG;IAC9F,gBAAgB,CAAC,SAAS,CAAC,EAAE;QAAE,QAAQ,CAAC,EAAE,QAAQ,GAAG,WAAW,GAAG,QAAQ,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,OAAO,CAAC,aAAa,CAAC;IAI5H,oEAAoE;YACtD,QAAQ;YAsJR,QAAQ;IA2GtB;;;;;;;;;OASG;IACH,OAAO,CAAC,wBAAwB;IAgBhC;;;;OAIG;IACH,OAAO,CAAC,iBAAiB;IAIzB;;;;;;;;;OASG;IACH,OAAO,CAAC,wBAAwB;IAuChC;;;OAGG;IACH,OAAO,CAAC,gBAAgB;IAiBxB;;;OAGG;IACH,iBAAiB,CACf,YAAY,CAAC,EAAE;QAAE,MAAM,CAAC,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAC;QAAC,cAAc,CAAC,EAAE,MAAM,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAA;KAAE,EAC5F,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE;QAAE,UAAU,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,GACpD,IAAI;IAKP;;;;;;;;;;OAUG;IACH,OAAO,CAAC,oBAAoB;IA0B5B,0DAA0D;IAC1D,KAAK,IAAI,IAAI;IAIb;;;;;;;;;;;;OAYG;IACG,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC;IAwB7B,uCAAuC;IACvC,UAAU,IAAI,YAAY,EAAE;IAI5B,0DAA0D;IAC1D,cAAc,CAAC,QAAQ,EAAE,MAAM,GAAG,IAAI;IAItC;;;OAGG;IACH,OAAO,CAAC,gBAAgB;IAmBxB;;;;;;;OAOG;IACH,OAAO,CAAC,cAAc;IAmEtB,OAAO,CAAC,eAAe;IAgBvB,OAAO,CAAC,oBAAoB;IAkB5B;;OAEG;IACH,OAAO,CAAC,4BAA4B;IAoBpC;;OAEG;IACH,OAAO,CAAC,iBAAiB;IAgBnB,MAAM,CACV,IAAI,EAAE,MAAM,EACZ,KAAK,EAAE,SAAS,EAChB,YAAY,CAAC,EAAE,MAAM,EACrB,QAAQ,CAAC,EAAE;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,GAAG,EAAE,MAAM,CAAA;KAAE,EAC3C,OAAO,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,OAAO,CAAA;KAAE,GAClC,OAAO,CAAC,aAAa,CAAC;IA4LzB;;;;;OAKG;YACW,YAAY;IA2I1B;;;;;;;;;;;;;;;;OAgBG;IACG,IAAI,CACR,IAAI,EAAE,MAAM,EACZ,KAAK,EAAE,SAAS,EAChB,OAAO,CAAC,EAAE;QAAE,QAAQ,CAAC,EAAE,MAAM,CAAC;QAAC,YAAY,CAAC,EAAE,MAAM,CAAA;KAAE,GACrD,OAAO,CAAC;QACT,IAAI,EAAE,IAAI,GAAG,IAAI,CAAA;QACjB,GAAG,EAAE,MAAM,CAAA;QACX,UAAU,EAAE,MAAM,CAAA;QAClB,UAAU,CAAC,EAAE,MAAM,CAAA;QACnB,WAAW,CAAC,EAAE,MAAM,CAAA;QACpB,YAAY,CAAC,EAAE,MAAM,CAAA;QACrB,oBAAoB,CAAC,EAAE,MAAM,CAAA;QAC7B,wBAAwB,CAAC,EAAE,MAAM,CAAA;QACjC,UAAU,CAAC,EAAE,MAAM,CAAA;KACpB,CAAC;IA6OF;;;;OAIG;IACG,QAAQ,CAAC,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,iBAAiB,CAAC;IAsDpE,sBAAsB,CAC1B,IAAI,EAAE,MAAM,EACZ,KAAK,EAAE,SAAS,EAChB,UAAU,EAAE,KAAK,CAAC;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC,EAC/D,YAAY,CAAC,EAAE,MAAM,GACpB,OAAO,CAAC,uBAAuB,CAAC;IAsEnC;;;;OAIG;IACG,oBAAoB,CACxB,KAAK,EAAE,SAAS,EAChB,IAAI,EAAE,MAAM,EACZ,aAAa,EAAE,MAAM,GACpB,OAAO,CAAC,gBAAgB,CAAC;IAmE5B;;;;OAIG;IACG,WAAW,CACf,KAAK,EAAE,SAAS,EAChB,IAAI,EAAE,MAAM,EACZ,WAAW,EAAE,MAAM,EAAE,EACrB,YAAY,CAAC,EAAE,MAAM,GACpB,OAAO,CAAC;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,aAAa,EAAE,CAAC;QAAC,GAAG,EAAE,MAAM,CAAC;QAAC,UAAU,CAAC,EAAE,MAAM,CAAC;QAAC,iBAAiB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;KAAE,CAAC;IAiGvI;;;;OAIG;IACG,gBAAgB,CACpB,cAAc,EAAE,MAAM,EACtB,MAAM,EAAE,MAAM,GACb,OAAO,CAAC,KAAK,CAAC;QAAE,IAAI,EAAE,QAAQ,GAAG,UAAU,GAAG,SAAS,GAAG,OAAO,CAAC;QAAC,GAAG,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAqDpG,OAAO,CAAC,KAAK;CAqDd;AAsHD;;;;;;;;;;;GAWG;AACH,wBAAgB,cAAc,CAAC,QAAQ,EAAE,MAAM,EAAE,QAAQ,SAAS,GAAG,MAAM,CAgE1E"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/brain/index.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,KAAK,EAAE,YAAY,EAAE,aAAa,EAAsB,MAAM,IAAI,CAAC;AAC1E,OAAO,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,WAAW,EAAE,aAAa,EAAE,gBAAgB,EAAE,IAAI,EAAY,MAAM,aAAa,CAAC;AAuXnH,MAAM,WAAW,aAAa;IAC5B,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,EAAE,CAAC;IACvB,GAAG,EAAE,MAAM,CAAC;IACZ,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB;;;;;OAKG;IACH,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B,6EAA6E;IAC7E,wBAAwB,CAAC,EAAE,MAAM,CAAC;IAClC,kFAAkF;IAClF,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,iBAAiB;IAChC,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,GAAG,EAAE,MAAM,CAAC;IACZ,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,uBAAuB;IACtC,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,GAAG,EAAE,MAAM,CAAC;IACZ,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAKD,qBAAa,KAAK;IAChB,OAAO,CAAC,UAAU,CAAoC;IACtD,OAAO,CAAC,QAAQ,CAA0G;IAC1H,OAAO,CAAC,SAAS,CAAS;IAC1B,OAAO,CAAC,oBAAoB,CAAU;IACtC,OAAO,CAAC,YAAY,CAAC,CAAS;IAC9B,OAAO,CAAC,WAAW,CAAC,CAA0G;IAC9H,OAAO,CAAC,cAAc,CAAC,CAAS;IAChC,OAAO,CAAC,OAAO,CAAC,CAAS;IACzB,OAAO,CAAC,KAAK,CAAU;IACvB,OAAO,CAAC,OAAO,CAAsB;IACrC,OAAO,CAAC,eAAe,CAAS;IAChC,OAAO,CAAC,aAAa,CAAU;IAC/B,OAAO,CAAC,cAAc,CAA8B;IACpD,OAAO,CAAC,eAAe,CAA8B;IACrD,OAAO,CAAC,YAAY,CAAS;IAC7B,OAAO,CAAC,gBAAgB,CAAU;IAClC,OAAO,CAAC,eAAe,CAAC,CAAS;IACjC,OAAO,CAAC,YAAY,CAAS;IAC7B,OAAO,CAAC,cAAc,CAAC,CAAS;IAChC,OAAO,CAAC,aAAa,CAAC,CAA0G;IAChI,OAAO,CAAC,cAAc,CAAU;IAEhC,OAAO,CAAC,YAAY,CAAC,CAAS;IAC9B,OAAO,CAAC,eAAe,CAAC,CAAS;IACjC,OAAO,CAAC,aAAa,CAAC,CAAS;IAC/B,OAAO,CAAC,gBAAgB,CAAC,CAAS;IAClC,OAAO,CAAC,eAAe,CAAC,CAAS;IACjC,OAAO,CAAC,kBAAkB,CAAC,CAAS;IACpC,OAAO,CAAC,kBAAkB,CAAC,CAAS;IACpC,OAAO,CAAC,qBAAqB,CAAC,CAAS;IACvC,OAAO,CAAC,sBAAsB,CAAC,CAAS;IAGxC,OAAO,CAAC,cAAc,CAAC,CAAgF;IACvG,OAAO,CAAC,oBAAoB,CAAC,CAA0C;gBAE3D,MAAM,GAAE,WAAgB;IAsCpC,OAAO,CAAC,gBAAgB;IASxB,OAAO,CAAC,qBAAqB;IAK7B,OAAO,CAAC,iBAAiB;IAoBzB,oGAAoG;IAC9F,gBAAgB,CAAC,SAAS,CAAC,EAAE;QAAE,QAAQ,CAAC,EAAE,QAAQ,GAAG,WAAW,GAAG,QAAQ,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,OAAO,CAAC,aAAa,CAAC;IAI5H,oEAAoE;YACtD,QAAQ;YAsJR,QAAQ;IA2GtB;;;;;;;;;OASG;IACH,OAAO,CAAC,wBAAwB;IAgBhC;;;;OAIG;IACH,OAAO,CAAC,iBAAiB;IAIzB;;;;;;;;;OASG;IACH,OAAO,CAAC,wBAAwB;IAuChC;;;OAGG;IACH,OAAO,CAAC,gBAAgB;IAiBxB;;;OAGG;IACH,iBAAiB,CACf,YAAY,CAAC,EAAE;QAAE,MAAM,CAAC,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAC;QAAC,cAAc,CAAC,EAAE,MAAM,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAA;KAAE,EAC5F,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE;QAAE,UAAU,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,GACpD,IAAI;IAKP;;;;;;;;;;OAUG;IACH,OAAO,CAAC,oBAAoB;IA0B5B,0DAA0D;IAC1D,KAAK,IAAI,IAAI;IAIb;;;;;;;;;;;;OAYG;IACG,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC;IAwB7B,uCAAuC;IACvC,UAAU,IAAI,YAAY,EAAE;IAI5B,0DAA0D;IAC1D,cAAc,CAAC,QAAQ,EAAE,MAAM,GAAG,IAAI;IAItC;;;OAGG;IACH,OAAO,CAAC,gBAAgB;IAmBxB;;;;;;;OAOG;IACH,OAAO,CAAC,cAAc;IA4EtB,OAAO,CAAC,eAAe;IAgBvB,OAAO,CAAC,oBAAoB;IAkB5B;;OAEG;IACH,OAAO,CAAC,4BAA4B;IAoBpC;;OAEG;IACH,OAAO,CAAC,iBAAiB;IAgBnB,MAAM,CACV,IAAI,EAAE,MAAM,EACZ,KAAK,EAAE,SAAS,EAChB,YAAY,CAAC,EAAE,MAAM,EACrB,QAAQ,CAAC,EAAE;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,GAAG,EAAE,MAAM,CAAA;KAAE,EAC3C,OAAO,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,OAAO,CAAA;KAAE,GAClC,OAAO,CAAC,aAAa,CAAC;IA4LzB;;;;;OAKG;YACW,YAAY;IA2I1B;;;;;;;;;;;;;;;;OAgBG;IACG,IAAI,CACR,IAAI,EAAE,MAAM,EACZ,KAAK,EAAE,SAAS,EAChB,OAAO,CAAC,EAAE;QAAE,QAAQ,CAAC,EAAE,MAAM,CAAC;QAAC,YAAY,CAAC,EAAE,MAAM,CAAA;KAAE,GACrD,OAAO,CAAC;QACT,IAAI,EAAE,IAAI,GAAG,IAAI,CAAA;QACjB,GAAG,EAAE,MAAM,CAAA;QACX,UAAU,EAAE,MAAM,CAAA;QAClB,UAAU,CAAC,EAAE,MAAM,CAAA;QACnB,WAAW,CAAC,EAAE,MAAM,CAAA;QACpB,YAAY,CAAC,EAAE,MAAM,CAAA;QACrB,oBAAoB,CAAC,EAAE,MAAM,CAAA;QAC7B,wBAAwB,CAAC,EAAE,MAAM,CAAA;QACjC,UAAU,CAAC,EAAE,MAAM,CAAA;KACpB,CAAC;IAiPF;;;;OAIG;IACG,QAAQ,CAAC,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,iBAAiB,CAAC;IAsDpE,sBAAsB,CAC1B,IAAI,EAAE,MAAM,EACZ,KAAK,EAAE,SAAS,EAChB,UAAU,EAAE,KAAK,CAAC;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC,EAC/D,YAAY,CAAC,EAAE,MAAM,GACpB,OAAO,CAAC,uBAAuB,CAAC;IAsEnC;;;;OAIG;IACG,oBAAoB,CACxB,KAAK,EAAE,SAAS,EAChB,IAAI,EAAE,MAAM,EACZ,aAAa,EAAE,MAAM,GACpB,OAAO,CAAC,gBAAgB,CAAC;IAkE5B;;;;OAIG;IACG,WAAW,CACf,KAAK,EAAE,SAAS,EAChB,IAAI,EAAE,MAAM,EACZ,WAAW,EAAE,MAAM,EAAE,EACrB,YAAY,CAAC,EAAE,MAAM,GACpB,OAAO,CAAC;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,aAAa,EAAE,CAAC;QAAC,GAAG,EAAE,MAAM,CAAC;QAAC,UAAU,CAAC,EAAE,MAAM,CAAC;QAAC,iBAAiB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;KAAE,CAAC;IAiGvI;;;;OAIG;IACG,gBAAgB,CACpB,cAAc,EAAE,MAAM,EACtB,MAAM,EAAE,MAAM,GACb,OAAO,CAAC,KAAK,CAAC;QAAE,IAAI,EAAE,QAAQ,GAAG,UAAU,GAAG,SAAS,GAAG,OAAO,CAAC;QAAC,GAAG,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAqDpG,OAAO,CAAC,KAAK;CAqDd;AAsHD;;;;;;;;;;;GAWG;AACH,wBAAgB,cAAc,CAAC,QAAQ,EAAE,MAAM,EAAE,QAAQ,SAAS,GAAG,MAAM,CAgE1E"}