chore: remove legacy backend and npm-app

Author: brandonkachen

.agents/LESSONS.md

@@ -1,6 +1,6 @@
 # Agent Lessons
 
-Lessons accumulated across buffbench runs. Each lesson identifies what went wrong (Issue) and what should have been done instead (Fix).
+Lessons accumulated across buffbench runs. Each lesson identifies what went wrong (Issue) and what should have been done instead (Fix). Many early notes mention the legacy backend/npm-app/WebSocket stack; those references are now historical with the SDK-first architecture.
 
 ## 2025-10-21T02:19:38.224Z — add-sidebar-fades (257cb37)
 
@@ -1624,4 +1624,3 @@ Do not include implementation details in your response; focus on ensuring all lo
 
 - **Issue:** Docs/examples not fixed to valid JSON after removing parentInstructions.
   **Fix:** Remove dangling keys/braces and ensure examples compile; replace spawnableAgents with subagents.
-

.agents/__tests__/context-pruner.test.ts

@@ -217,10 +217,15 @@ describe('context-pruner handleSteps', () => {
     expect(resultMessages.length).toBeLessThan(messages.length)
 
     // Should contain replacement messages
-    const hasReplacementMessage = resultMessages.some(
-      (m: any) =>
-        typeof m.content === 'string' &&
-        m.content.includes('Previous message(s) omitted due to length'),
+    const hasReplacementMessage = resultMessages.some((m: any) =>
+      Array.isArray(m.content)
+        ? m.content.some(
+            (part: any) =>
+              part?.type === 'text' &&
+              part.text.includes('Previous message(s) omitted due to length'),
+          )
+        : typeof m.content === 'string' &&
+          m.content.includes('Previous message(s) omitted due to length'),
     )
     expect(hasReplacementMessage).toBe(true)
   })
@@ -243,10 +248,14 @@ describe('context-pruner handleSteps', () => {
     const resultMessages = results[0].input.messages
 
     // Important message should be preserved
-    const importantMessage = resultMessages.find(
-      (m: any) =>
-        typeof m.content === 'string' &&
-        m.content.includes('Important message'),
+    const importantMessage = resultMessages.find((m: any) =>
+      Array.isArray(m.content)
+        ? m.content.some(
+            (part: any) =>
+              part?.type === 'text' && part.text.includes('Important message'),
+          )
+        : typeof m.content === 'string' &&
+          m.content.includes('Important message'),
     )
     expect(importantMessage).toBeDefined()
   })

CONTRIBUTING.md

@@ -79,22 +79,18 @@ Before you begin, you'll need to install a few tools:
 8. **Start development services** (requires 3 terminals):
 
    ```bash
-   # Terminal 1 - Backend server (start first)
-   bun run start-server
-   # Expected: 🚀 Server is running on port 4242
-
-   # Terminal 2 - Web server (start second)
+   # Terminal 1 - Web server
    bun run start-web
    # Expected: Ready on http://localhost:3000
 
-   # Terminal 3 - CLI client (start last)
+   # Terminal 2 - CLI client
    bun run start-bin
    # Expected: Welcome to Codebuff! + agent list
    ```
 
    Now, you should be able to run the CLI and send commands, but it will error out because you don't have any credits.
 
-   **Note**: CLI requires both backend and web server running for authentication.
+   **Note**: The CLI uses the SDK to talk directly to the web API; no separate backend process is required.
 
 9. **Giving yourself credits**:
 
@@ -148,9 +144,8 @@ In order to run the CLI from other directories, you need to first publish the ag
 
 Codebuff is organized as a monorepo with these main packages:
 
-- **backend/**: WebSocket server, LLM integration, agent orchestration
-- **npm-app/**: CLI application that users interact with
 - **web/**: Next.js web application and dashboard
+- **cli/**: CLI application that users interact with
 - **python-app/**: Python version of the CLI (experimental)
 - **common/**: Shared code, database schemas, utilities
 - **sdk/**: TypeScript SDK for programmatic usage
@@ -250,15 +245,15 @@ Build specialized agents in `.agents/` for different languages, frameworks, or w
 
 ### 🔧 **Tool System**
 
-Add new capabilities in `backend/src/tools.ts` - file operations, API integrations, development environment helpers. The sky's the limit!
+Add new capabilities in `sdk/src/tools` or the agent runtime (packages/agent-runtime) - file operations, API integrations, development environment helpers. The sky's the limit!
 
 ### 📦 **SDK Improvements**
 
 Make the SDK in `sdk/` even more powerful with new methods, better TypeScript support, or killer integration examples.
 
 ### 💻 **CLI Magic**
 
-Enhance the user experience in `npm-app/` with smoother commands, better error messages, or interactive features that make developers smile.
+Enhance the user experience in `cli/` with smoother commands, better error messages, or interactive features that make developers smile.
 
 ### 🌐 **Web Dashboard**
 

authentication.knowledge.md

@@ -2,13 +2,13 @@
 
 ## Overview
 
-Codebuff implements secure authentication between CLI (npm-app), backend, and web application using fingerprint-based device identification.
+Codebuff implements secure authentication between the CLI, the API, and the web application using fingerprint-based device identification.
 
 ## Core Authentication Flow
 
 ```mermaid
 sequenceDiagram
-    participant CLI as npm-app
+    participant CLI as CLI
     participant Web as web app
     participant DB as Database
 
@@ -30,7 +30,6 @@ sequenceDiagram
 ### 1. First Time Login / Missing Credentials
 
 - CLI generates fingerprint from hardware info + 8 random bytes
-- Uses `calculateFingerprint()` in `npm-app/src/fingerprint.ts`
 - Continues to core flow with new fingerprintId
 
 ### 2. Logout Flow