Feat Containerized Sandbox Code Execution

[hassan11196]

Sandbox Code Execution Approval System

Overview

Implements a comprehensive approval mechanism for sandbox code execution, allowing administrators to configure approval requirements at deployment level and users to toggle approval mode per session.

Features

1. Deployment-Level Configuration

• Configure default approval mode (auto or manual) in deployment YAML
• Sandbox timeout, allowed images, and enabled status configuration
• Settings stored in src/utils/sandbox/config.py

sandbox:
  enabled: true
  approval_mode: manual  # or 'auto'
  timeout: 30
  default_image: "python:3.11-slim"

2. Session-Level Approval Toggle

• Users can override deployment defaults per chat session
• Settings icon in chat UI provides approval mode toggle
• Preference stored in Flask session and persists across queries

3. Real-Time Approval Flow

• Threading Architecture: Agent stream runs in background thread while main thread handles SSE event delivery
• Event Queue: Approval requests pass through queue.Queue for thread-safe communication
• Non-Blocking: Chat UI remains responsive while waiting for approval
• Debug Logging: Comprehensive logging throughout approval flow for troubleshooting

4. Enhanced UI

• Approval Modal: Displays code preview with syntax highlighting
• Full Code View: Executed code shown in chat trace with:
  • Language and image badges
  • Line count indicator
  • Syntax-highlighted code block
  • Copy-to-clipboard button
• Visual Feedback: Pending status indicator with pulsing animation
• Session Toggle: Settings panel for enabling/disabling manual approval

5. API Endpoints

• GET/POST /api/sandbox/approval-mode - Get/set session approval preference
• GET /api/sandbox/approval/<id> - Get approval request details
• POST /api/sandbox/approval/<id>/approve - Approve execution
• POST /api/sandbox/approval/<id>/reject - Reject execution
• GET /api/sandbox/approvals/pending - List pending approvals
• GET /api/sandbox/config - Get deployment sandbox configuration

All endpoints protected with require_auth decorator for SSO/RBAC integration.

Implementation Details

Backend Changes

• src/archi/pipelines/agents/tools/sandbox.py: Approval callback integration, mode resolution (deployment → session → auto)
• src/utils/sandbox/approval.py: Approval request storage, resolution tracking, pending request queries
• src/utils/sandbox/config.py: Configuration management with ApprovalMode enum
• src/interfaces/chat_app/app.py:
  • Threading architecture for non-blocking stream + approval events
  • Session management for approval mode preference
  • API endpoints for approval workflow
  • Fixed session structure usage (session['user'] instead of session['uid'])

Frontend Changes

• src/interfaces/chat_app/static/chat.js:
  • handleSandboxApproval() for approval/rejection with credentials: 'include'
  • renderApprovalRequest() enhanced with full code display
  • Session approval mode toggle in settings
  • Event handling for approval_request SSE events
• src/interfaces/chat_app/static/chat.css:
  • Approval modal styling with code block formatting
  • Badge styling for language/image/lines indicators
  • Pulsing animation for pending status

Security Considerations

• All approval endpoints require authentication via require_auth decorator
• Session-based approval tracking prevents cross-user approval hijacking
• Username/email captured for audit trail on all approvals/rejections
• Cookies sent with credentials: 'include' for proper session handling

Testing

Enable manual approval mode and verify:

1. Approval modal appears when sandbox tool executes
2. Approving allows code execution to proceed
3. Rejecting cancels execution
4. Code displays correctly in chat trace
5. Session toggle persists across queries
6. Debug logs show callback invocation and event flow

Technical Architecture

User Query → Agent Pipeline → Sandbox Tool
                                    ↓
                            Check Approval Mode
                                    ↓
                         ┌─────────┴──────────┐
                         ↓                     ↓
                    Auto Mode            Manual Mode
                         ↓                     ↓
                    Execute              Callback
                                              ↓
                                         Event Queue
                                              ↓
                                         SSE Stream
                                              ↓
                                         UI Modal
                                              ↓
                                    User Approves/Rejects
                                              ↓
                                      API Endpoint
                                              ↓
                                    Resolve Approval
                                              ↓
                                    Continue/Cancel

Future Enhancements

• Per-user approval mode defaults stored in database
• Approval request expiration/timeout
• Approval history viewer in UI
• Webhook notifications for approval requests
• Multi-level approval workflows (require N approvers)