CLI Builder for Any Web App

What it does

Clicks through a web app with developer tools open to discover the real requests behind its buttons, and wraps them in a safe command-line tool (or an MCP server) your AI agent can run afterward. All writes run as dry runs until you say go.

When to use it

• When an app you use daily has no public API but does have a web dashboard.
• When you want to automate repetitive click-based tasks in an internal tool.
• When you need your AI agent to operate an app without exposing credentials in chat.

Steps

1. Tell it which web app and which actions you want automated.
2. The agent clicks through the app with developer tools open.
3. The real requests behind each relevant button are mapped.
4. The command-line tool (or MCP) wrapping them is built.
5. Every write runs first as a dry run for your review.
6. You only trigger real execution once you explicitly confirm.

The skill

---
name: app-api-cli-builder
description: Maps the hidden requests behind a web app's buttons and wraps them in a safe command-line tool to automate tasks.
---

# CLI Builder for Any Web App

## 1. What it does

- Clicks through a web app with developer tools open to discover the real requests behind its buttons, and wraps them in a safe command-line tool (or an MCP server) your AI agent can run afterward. All writes run as dry runs until you say go.

## 2. When to use it

- When an app you use daily has no public API but does have a web dashboard.
- When you want to automate repetitive click-based tasks in an internal tool.
- When you need your AI agent to operate an app without exposing credentials in chat.

## 3. Step by step

- Tell it which web app and which actions you want automated.
- The agent clicks through the app with developer tools open.
- The real requests behind each relevant button are mapped.
- The command-line tool (or MCP) wrapping them is built.
- Every write runs first as a dry run for your review.
- You only trigger real execution once you explicitly confirm.

## 4. Safety rules

- Safety rules: everything ships as a draft first; nothing is sent, published, or paid for without your explicit approval; no data, numbers, or results are ever invented; no credentials or secrets are typed into chat.

## 5. Best practices

- Always review the full draft before approving any send, publish, or payment.
- If real information (data, numbers, results) is missing, ask for it to be flagged as pending instead of invented.
- Keep credentials and keys out of chat; set them in environment variables or your secrets manager.

## 6. GoHighLevel integration

- Where relevant, results can be saved as contacts, tags, tasks, or notes inside your GoHighLevel sub-account.
- No automation inside GHL (workflow, campaign, trigger) activates without your review first.

For Claude Code, save it as ~/.claude/skills/app-api-cli-builder/SKILL.md. In Cursor, ChatGPT or another assistant, paste it as project instructions.