fix: restore preview image in README

.gitignore

@@ -27,5 +27,4 @@ Thumbs.db
 coverage/
 
 # Claude Code
-CLAUDE.md
 .claude/

CLAUDE.md

@@ -0,0 +1,102 @@
+# Open Alice
+
+File-driven AI trading agent. All state (sessions, config, logs) stored as files — no database.
+
+## Quick Start
+
+```bash
+pnpm install
+pnpm dev        # Dev mode (tsx watch, port 3002)
+pnpm build      # Production build (backend + UI)
+pnpm test       # Vitest
+```
+
+## Project Structure
+
+```
+src/
+├── main.ts                    # Composition root
+├── core/
+│   ├── agent-center.ts        # Top-level AI orchestration, owns GenerateRouter
+│   ├── ai-provider-manager.ts # GenerateRouter + StreamableResult + AskOptions
+│   ├── tool-center.ts         # Centralized tool registry (Vercel + MCP export)
+│   ├── session.ts             # JSONL session store
+│   ├── compaction.ts          # Auto-summarize long context windows
+│   ├── config.ts              # Zod-validated config loader
+│   ├── ai-config.ts           # Runtime AI provider selection
+│   ├── event-log.ts           # Append-only JSONL event log
+│   ├── connector-center.ts    # ConnectorCenter — push delivery + last-interacted tracking
+│   ├── async-channel.ts       # AsyncChannel for streaming provider events to SSE
+│   ├── model-factory.ts       # Model instance factory for Vercel AI SDK
+│   ├── media.ts               # MediaAttachment extraction
+│   ├── media-store.ts         # Media file persistence
+│   └── types.ts               # Plugin, EngineContext interfaces
+├── ai-providers/
+│   ├── claude-code/           # Claude Code CLI subprocess
+│   ├── vercel-ai-sdk/         # Vercel AI SDK ToolLoopAgent
+│   └── agent-sdk/             # Agent SDK (@anthropic-ai/claude-agent-sdk)
+├── extension/
+│   ├── analysis-kit/          # Indicators, market data tools, sandbox
+│   ├── equity/                # Equity fundamentals
+│   ├── market/                # Unified symbol search
+│   ├── news/                  # OpenBB news tools
+│   ├── news-collector/        # RSS collector + archive search
+│   ├── trading/               # Unified multi-account trading, guard pipeline, git-like commits
+│   ├── thinking-kit/          # Reasoning and calculation tools
+│   ├── brain/                 # Cognitive state (memory, emotion)
+│   └── browser/               # Browser automation bridge (OpenClaw)
+├── openbb/                    # In-process data SDK (equity, crypto, currency, commodity, economy, news)
+├── connectors/
+│   ├── web/                   # Web UI (Hono, SSE streaming, sub-channels)
+│   ├── telegram/              # Telegram bot (grammY)
+│   └── mcp-ask/               # MCP Ask connector
+├── plugins/
+│   └── mcp.ts                 # MCP protocol server
+├── task/
+│   ├── cron/                  # Cron scheduling
+│   └── heartbeat/             # Periodic heartbeat
+├── skills/                    # Agent skill definitions
+└── openclaw/                  # ⚠️ Frozen — DO NOT MODIFY
+```
+
+## Key Architecture
+
+### AgentCenter → GenerateRouter → GenerateProvider
+
+Two layers (Engine was removed):
+
+1. **AgentCenter** (`core/agent-center.ts`) — top-level orchestration. Manages sessions, compaction, and routes calls through GenerateRouter. Exposes `ask()` (stateless) and `askWithSession()` (with history).
+
+2. **GenerateRouter** (`core/ai-provider-manager.ts`) — reads `ai-provider.json` on each call, resolves to active provider. Three backends:
+   - Claude Code CLI (`inputKind: 'text'`)
+   - Vercel AI SDK (`inputKind: 'messages'`)
+   - Agent SDK (`inputKind: 'text'`)
+
+**AIProvider interface**: `ask(prompt)` for one-shot, `generate(input, opts)` for streaming `ProviderEvent` (tool_use / tool_result / text / done). Optional `compact()` for provider-native compaction.
+
+**StreamableResult**: dual interface — `PromiseLike` (await for result) + `AsyncIterable` (for-await for streaming). Multiple consumers each get independent cursors.
+
+Per-request provider and model overrides via `AskOptions.provider` and `AskOptions.vercelAiSdk` / `AskOptions.agentSdk`.
+
+### ConnectorCenter
+
+`connector-center.ts` manages push channels (Web, Telegram, MCP Ask). Tracks last-interacted channel for delivery routing.
+
+### ToolCenter
+
+Centralized registry. Extensions register tools, exports in Vercel and MCP formats. Decoupled from AgentCenter.
+
+## Conventions
+
+- ESM only (`.js` extensions in imports), path alias `@/*` → `./src/*`
+- Strict TypeScript, ES2023 target
+- Zod for config, TypeBox for tool parameter schemas
+- `decimal.js` for financial math
+- Pino logger → `logs/engine.log`
+
+## Git Workflow
+
+- `origin` = `TraderAlice/OpenAlice` (production)
+- `dev` branch for all development, `master` only via PR
+- **Never** force push master, **never** push `archive/dev` (contains old API keys)
+- CLAUDE.md is **committed to the repo and publicly visible** — never put API keys, personal paths, or sensitive information in it

package.json

@@ -39,6 +39,7 @@
     "@hono/node-server": "^1.19.11",
     "@modelcontextprotocol/sdk": "^1.26.0",
     "@sinclair/typebox": "0.34.48",
+    "@traderalice/opentypebb": "workspace:*",
     "ai": "^6.0.86",
     "ajv": "^8.18.0",
     "ccxt": "^4.5.38",
@@ -49,7 +50,6 @@
     "grammy": "^1.40.0",
     "hono": "^4.12.7",
     "json5": "^2.2.3",
-    "@traderalice/opentypebb": "workspace:*",
     "pino": "^10.3.1",
     "playwright-core": "1.58.2",
     "sharp": "^0.34.5",
@@ -63,6 +63,7 @@
     "@types/express": "^5.0.6",
     "@types/node": "^25.2.3",
     "@types/ws": "^8.18.1",
+    "@vitest/coverage-v8": "^4.0.18",
     "tsup": "^8.5.1",
     "tsx": "^4.21.0",
     "typescript": "^5.9.3",

src/ai-providers/agent-sdk/agent-sdk-provider.ts

@@ -10,15 +10,14 @@
 
 import { resolve } from 'node:path'
 import type { Tool } from 'ai'
-import type { ProviderResult, ProviderEvent, GenerateProvider, GenerateInput, GenerateOpts } from '../types.js'
+import type { ProviderResult, ProviderEvent, AIProvider, GenerateInput, GenerateOpts } from '../types.js'
 import type { AgentSdkConfig, AgentSdkOverride } from './query.js'
 import { readAgentConfig } from '../../core/config.js'
-import { extractMediaFromToolResultContent } from '../../core/media.js'
 import { createChannel } from '../../core/async-channel.js'
 import { askAgentSdk } from './query.js'
 import { buildAgentSdkMcpServer } from './tool-bridge.js'
 
-export class AgentSdkProvider implements GenerateProvider {
+export class AgentSdkProvider implements AIProvider {
   readonly inputKind = 'text' as const
   readonly providerTag = 'agent-sdk' as const
 
@@ -66,7 +65,6 @@ export class AgentSdkProvider implements GenerateProvider {
     const mcpServer = await this.buildMcpServer(opts?.disabledTools)
 
     const channel = createChannel<ProviderEvent>()
-    const media: import('../../core/types.js').MediaAttachment[] = []
 
     const resultPromise = askAgentSdk(
       input.prompt,
@@ -76,7 +74,6 @@ export class AgentSdkProvider implements GenerateProvider {
           channel.push({ type: 'tool_use', id, name, input: toolInput })
         },
         onToolResult: ({ toolUseId, content }) => {
-          media.push(...extractMediaFromToolResultContent(content))
           channel.push({ type: 'tool_result', tool_use_id: toolUseId, content })
         },
         onText: (text) => {
@@ -92,7 +89,7 @@ export class AgentSdkProvider implements GenerateProvider {
 
     const result = await resultPromise
     const prefix = result.ok ? '' : '[error] '
-    yield { type: 'done', result: { text: prefix + result.text, media } }
+    yield { type: 'done', result: { text: prefix + result.text, media: [] } }
   }
 
 }

src/ai-providers/agent-sdk/query.ts

@@ -10,7 +10,7 @@ import type { McpSdkServerConfigWithInstance } from '@anthropic-ai/claude-agent-
 import { pino } from 'pino'
 import type { ContentBlock } from '../../core/session.js'
 import { readAIProviderConfig } from '../../core/config.js'
-import { logToolCall } from '../log-tool-call.js'
+import { logToolCall } from '../utils.js'
 
 const logger = pino({
   transport: { target: 'pino/file', options: { destination: 'logs/agent-sdk.log', mkdir: true } },