> ralphy.sh

AI Agent Automation & Auto-Accept for VS Code

Stop clicking. Start building. Ralphy.sh automatically executes AI agent decisions in Cursor, Antigravity, and other AI-powered coding environments.

What It Does

> Auto-Accept on Steroids

Automatically accept file edits, terminal commands, and recovery prompts. Never click "Accept" again.

> Ralph Loop (Autonomous Agents)

Feed it a PRD. Watch it execute tasks end-to-end. Context-aware, stuck-recovery enabled, fully logged.

> Multi-Agent Parallel Processing

Run unlimited AI agents simultaneously across all tabs. Task isolation. Zero manual intervention.

> Security Sandbox

Block dangerous commands automatically. Customize blocklists. Stay safe while automating.

Features at a Glance

Feature	Description
Auto-Accept	Intelligent button detection & automatic approval
Ralph Loop	Autonomous agent execution from PRD documents
Background Mode	Process tabs in parallel, even hidden ones
Stuck Recovery	Auto-escape when agents get stuck
Context Guard	Smart context management & resets
Task Tracking	Full transparency in `.agent/` markdown files
Dashboard	Real-time analytics & action history

Quick Start

# Install from VSIX
code --install-extension ralphy-sh-1.3.9.vsix

Or search "ralphy.sh" / "auto accept" / "ralph loop" in VS Code Extensions.

Launch Auto-Accept

Press Ctrl+Alt+A  (or Cmd+Alt+A on Mac)

Launch Ralph Loop (Autonomous Mode)

Press Ctrl+Alt+R  (or Cmd+Alt+R on Mac)

Or use command palette: Agent Ralph: Start Ralph Loop

How Ralph Loop Works

Load a PRD → Click "Select PRD File" in Ralph panel
Parse Requirements → Agent extracts MUST/SHOULD/COULD tasks
Execute Autonomously → Tasks run one-by-one with context resets
Track Progress → All actions logged in .agent/TASKS.md
Auto-Stop → Halts when all tasks complete or max resets hit

PRD Document Format

For best results, structure your PRD like this:

# Feature: My Feature

## Requirements
- As a user, I MUST be able to login
- The system SHOULD remember preferences
- COULD: Add dark mode support

## Technical Notes
- Use TypeScript for all new files
- Files to modify: `src/auth.ts`, `src/config.ts`

Smart Keywords:

MUST / SHALL → Priority: CRITICAL
SHOULD → Priority: HIGH
COULD / MAY → Priority: MEDIUM

Keyboard Shortcuts

Shortcut	Action
`Ctrl+Alt+A`	Start/Stop Auto-Accept
`Ctrl+Alt+S`	Stop Agent Automation
`Ctrl+Alt+R`	Start Ralph Loop
`Ctrl+Alt+M`	Select AI Model

Core Commands

Agent: Start Agent Automation
Agent: Stop Agent Automation
Agent: Show Dashboard
Agent: Toggle Background Mode (Pro)
Agent: Configure Banned Commands (Pro)
Agent Ralph: Load PRD Document
Agent Ralph: Start Ralph Loop
Agent Ralph: Pause Ralph Loop
Agent Ralph: Resume Ralph Loop
Agent Ralph: Stop Ralph Loop
Agent Ralph: Show Ralph Status
Agent Ralph: Reset Context Window
Agent Automation: Select AI Model

Essential Settings

{
  "agentAutomation.autoAccept": true,              // Enable auto-accept
  "agentAutomation.backgroundMode": true,         // Process tabs in parallel
  "agentAutomation.pollingSpeed": 300,            // Detection interval (ms)
  "agentAutomation.stuckRecovery": true,          // Auto-escape stuck loops
  "agentAutomation.createAgentFolder": false,     // Enable .agent/ tracking
  "agentAutomation.ralph.enabled": false,         // Enable Ralph mode
  "agentAutomation.ralph.contextResetThreshold": 0.8,  // Reset at 80% context
  "agentAutomation.autoInjectRules": true,        // Auto-create .agent-rules.md
  "agentAutomation.enforceJSDoc": true,           // Require JSDoc comments
  "agentAutomation.bannedCommands": [
    "rm -rf /",
    "rm -rf ~",
    "format c:",
    "del /f /s /q"
  ]
}

Agent State Folder (.agent/)

When Ralph mode is enabled, a transparent state folder is created:

.agent/
├── STATUS.md       # Current mode & active task
├── TASKS.md        # Generated task list with checkboxes
├── DECISIONS.md    # Architectural decisions log
├── GUARDRAILS.md   # Patterns & anti-patterns
├── HISTORY.md      # Append-only action log
└── cache/          # Temporary data (git-ignored)

Commit these files to git for full team visibility into agent execution.

Search Keywords

This extension is optimized for searches:

auto accept - Automatically approve AI suggestions
ai agent automation - Autonomous code execution
ralph loop - Continuous autonomous agent loops
ralph tasks - Autonomous task execution from PRDs
autonomous agents - Self-executing AI workflows
agent automation vscode - Orchestrated AI in VS Code

Compatible With

Cursor (Native Integration)
Antigravity (VS Code Extension)
VS Code Built-in LM API (Claude, GPT-4, etc.)
OpenAI (Fallback via API key)
Anthropic (Fallback via API key)

Why Ralphy.sh?

Zero Manual Intervention → Set and forget agent automation
Context Aware → Intelligent resets prevent token overflow
Fully Logged → Every action tracked in markdown
Developer Friendly → Built by developers for developers
Free Forever → MIT License. No paywalls.

Advanced: Custom AI Models

Command Palette → Agent Automation: Select AI Model

Supports:

Claude 3.5 Sonnet (Default)
GPT-4o / GPT-4o-mini
GPT-4 Turbo
Custom models via API keys

Troubleshooting

Agent stuck in loop?

Enable agentAutomation.stuckRecovery: true
Lower agentAutomation.pollingSpeed (faster detection)

Ralph Loop not starting?

Set agentAutomation.createAgentFolder: true
Ensure .agent/ folder exists

Commands not executing?

Check agentAutomation.bannedCommands blocklist
Verify agent permissions in security settings

Development

Built with ❤️ for developers who value their time.

Made for Antigravity & Cursor communities.

License

MIT - Free and Open Source Forever.

Ready to automate? Press Ctrl+Alt+A and watch the magic happen.

Ralphy.sh | AI Agent Automation and Auto-Accept

Ralphy.sh