burn-after-login

“Good morning. Your mission, should you choose to accept it, is to create dev-only auth shortcuts so AI browser agents stop fumbling with your login forms. This skill will self-destruct after completion.”

A self-destructing Agent Skill that creates dev-only auth shortcuts for AI browser automation — then removes itself, leaving no trace.

The Problem

AI agents with browser automation (Claude in Chrome, Playwright MCP, agent-browser, Stagehand, Cursor’s browser mode…) hit login walls and waste time filling out forms, handling redirects, and losing session state.

The Solution

Install the skill, run it, and it sets up dev-only auth shortcuts tailored to your stack. After setup, your agents can authenticate in one step:

# Browser agents navigate here to get authenticated
http://localhost:3000/[email protected]&password=demo

# Or grab a token for API testing
TOKEN=$(./scripts/get-token [email protected] demo)

How It Works

1. Install the skill
2. Run it — it analyzes your auth system and creates dev-only shortcuts
3. It detects your browser automation tools and updates agent instructions
4. It asks to self-destruct
5. You're left with working auth shortcuts and zero trace of the skill

Install

npx skills add pbakaus/burn-after-login

This installs the skill into your project’s .agents/skills/ directory (and symlinks into .claude/skills/ and other tool-specific locations as needed).

Use

In Claude Code:

/burn-after-login

In other compatible agents, invoke the skill per your tool’s conventions.

What It Does

1. Validates your environment has dev mode detection and an auth system
2. Finds existing dev/test credentials in your codebase
3. Analyzes how your auth works (tokens, sessions, or both)
4. Creates dev-only auth shortcuts that match your stack:
   • Login endpoints for session-based auth
   • Token generator scripts for API auth
   • All guarded to return 404 in production
5. Detects your browser automation tools (Playwright, Puppeteer, Claude in Chrome, Cypress, Stagehand, etc.)
6. Updates your agent instructions (CLAUDE.md, AGENTS.md, CURSOR.md, etc.) with tool-specific auth examples
7. Self-destructs — removes itself from your project

Supported Stacks

Works with any combination of:

• Languages: Node.js, Python, Ruby, Go, Rust, and more
• Frameworks: Next.js, Django, Rails, Flask, FastAPI, Express, and more
• Auth: Supabase, NextAuth, Clerk, Firebase, Passport, Devise, custom auth, and more
• Browser tools: Claude in Chrome, Playwright MCP, Chrome DevTools MCP, agent-browser, Stagehand, Puppeteer, Cypress, Selenium

Skill Structure

This repo follows the Agent Skills specification:

burn-after-login/
├── SKILL.md      # The skill instructions (required by spec)