The 'Yes Man' Pattern: Automating AI Coding Assistants

Let the AI do its thing. Stop babysitting prompts.

---

AI coding assistants are powerful. They can write code, run tests, fix bugs, even commit changes. But they keep asking for permission.

"Should I proceed?" Yes. "Do you want me to run this?" Yes. "Shall I continue?" Yes.

After the hundredth confirmation, I built agent-yes: a wrapper that automatically handles these prompts so AI assistants can work uninterrupted.

The Problem

Every AI CLI tool asks for confirmation before taking actions:

Claude: I'll update the test file. Proceed?
> [waiting for input]

Claude: I'll run the tests now. Continue?
> [waiting for input]

Claude: Tests passed. Should I commit?
> [waiting for input]

For simple tasks, this is fine. For complex multi-step tasks, you're constantly babysitting. You can't walk away and let it work.

I wanted to say "run all tests and commit changes" and come back to a completed task.

The Solution

agent-yes wraps AI CLI tools and automatically responds to prompts:

# Instead of
claude
> run all tests and commit changes
> [manual confirmations...]

# Just run
claude-yes -- run all tests and commit changes
# Walk away. Come back to committed code.

It detects when the AI is asking for confirmation and automatically sends "Yes".

Supported AI Tools

agent-yes works with 11 AI coding assistants:

Command	AI Tool
claude-yes	Claude (Anthropic)
copilot-yes	GitHub Copilot
gemini-yes	Google Gemini
codex-yes	OpenAI Codex
cursor-yes	Cursor
grok-yes	Grok (xAI)
qwen-yes	Qwen (Alibaba)
auggie-yes	Augment Code

Same interface, any AI backend.

How It Works

Under the hood, agent-yes uses node-pty to spawn a pseudo-terminal and monitor output:

// Simplified architecture
const pty = spawn(cliCommand, args);

pty.onData((data) => {
  // Check if AI is asking for confirmation
  if (matchesPromptPattern(data)) {
    pty.write("Yes\n");
  }

  // Check if AI is ready for new input
  if (matchesReadyPattern(data)) {
    // Can queue next task
  }

  // Check for fatal errors
  if (matchesFatalPattern(data)) {
    process.exit(1);
  }
});

Each supported CLI has its own configuration:

const claudeConfig = {
  binary: "claude",
  readyPatterns: [/^>\s*$/m, /What would you like/],
  enterPatterns: [/Yes, proceed/i, /Continue\?/i],
  fatalPatterns: [/Error:/i, /failed/i],
};

Usage Patterns

Basic: Run and Confirm

claude-yes -- refactor the auth module

The AI works through the task, automatically confirming each step.

Auto-Exit When Done

claude-yes --exit-on-idle=60s "run all tests and commit changes"

Exits automatically when the AI has been idle for 60 seconds. Perfect for CI/CD.

Multiple Tasks

claude-yes << EOF
1. Run the test suite
2. Fix any failing tests
3. Update the changelog
4. Commit with a descriptive message
EOF

Library Usage

import claudeYes from "agent-yes";

await claudeYes({
  prompt: "solve all TODOs in the codebase",
  cli: "claude",
  exitOnIdle: 30000,
  continueOnCrash: true,
  logFile: "claude.log",
});

Docker

docker run --rm -v $(pwd):/workspace -w /workspace \
  ghcr.io/snomiao/agent-yes:latest \
  -- run tests and fix failures

Real-World Workflows

Morning Automation

# Start your day with automated cleanup
claude-yes --exit-on-idle=120s << EOF
1. Update dependencies
2. Run tests
3. Fix any deprecation warnings
4. Commit changes as "chore: morning cleanup"
EOF

PR Review Assistant

# Auto-review and fix a PR
claude-yes -- review the changes in this PR, fix any issues, update tests

Documentation Generator

# Generate docs for new code
claude-yes -- generate JSDoc comments for all exported functions in src/

Safety Considerations

Warning: agent-yes automatically executes commands. Only use on:

• Trusted codebases
• Sandboxed environments
• Code you've reviewed

Prompt injection risk: Malicious code could embed instructions that the AI follows. In automated workflows, this is a real concern.

Mitigations:

• Review code before running agent-yes
• Use in sandboxed containers
• Set up git hooks to review changes before commit
• Use --exit-on-idle to prevent infinite loops

Installation

# Global install
npm install -g agent-yes

# Or use directly with npx
npx claude-yes -- your task here

# Or with Bun
bunx claude-yes -- your task here

Why This Pattern Matters

AI assistants are becoming agentic - they can plan, execute, and iterate. But the current UX treats them like tools that need constant supervision.

The "Yes Man" pattern acknowledges that for many tasks, we trust the AI to make reasonable decisions. We don't need to approve every file write or every test run.

This doesn't mean blind trust. It means:

• Define the task clearly upfront
• Let the AI execute autonomously
• Review the final result

It's the difference between micromanaging and delegating.

What's Next

The pattern extends to:

• Batch operations: Process multiple repos overnight
• CI/CD integration: AI-powered code review and fixes
• Scheduled maintenance: Weekly dependency updates, security patches

As AI assistants get more capable, the "Yes Man" wrapper becomes essential infrastructure.

---

Try It

Install: npm install -g agent-yes

GitHub: github.com/snomiao/agent-yes

Quick start:

claude-yes -- help me solve all todos in this repo

---

Snowstar Miao builds automation tools for AI-assisted development.