← Back to Blog

Build Your Own AI Coding Agent Workflow (No Vendor Lock-In)

Claude Code refuses requests when it sees "OpenClaw" in your commits. Here's how to build your own AI coding agent on Telegram — with zero gatekeeping, zero keyword scanning, and zero surprise charges.

Published by GetClawCloud · May 1, 2026

What Happened?

A story with 962 points just hit the top of Hacker News: "Claude Code refuses requests or charges extra if your commits mention 'OpenClaw'". Developers discovered that Claude Code's behavior changes depending on what keywords appear in your git commit messages — refusing to complete certain requests or silently upgrading your billing tier if it detects third-party tool names like "OpenClaw".

This isn't just a Twitter drama. This is a real problem:

You can't trust the tool. If your AI assistant changes its behavior based on what you type, it's not a tool — it's a gate.
You pay for someone else's policy decisions. Anthropic confirmed in April that Claude Pro/Max subscribers can no longer use their credits through third-party frameworks. Now they're enforcing it at the commit level.
It breaks your workflow. Imagine your CI pipeline failing not because of bad code, but because your AI decided it didn't like the word you used in a commit message.

The Alternative: An AI Agent You Control

The real lesson here isn't "which AI tool to use." It's: don't build your workflow around a tool that can change the rules on you overnight.

With OpenClaw, you can build your own AI coding agent workflow on Telegram — using any model you choose (GPT-4o, Claude Sonnet, GPT-5.5, open-source models). Your agent lives in your Telegram chat. It has web search, file access, and code execution. And the only keyword it reacts to is the one you tell it to.

Here's what a self-hosted coding agent gives you:

No hidden behavior. Your agent does what you prompt it to do. No silent policy enforcement, no keyword-based billing changes.
Your own API key. You pay the model provider directly. No intermediary deciding how you can use your credits.
Full customizability. Add tools, change personalities, switch models mid-task. It's your agent.

Ready-to-Use Prompt: AI Coding Workflow Assistant

Copy this prompt into your OpenClaw Telegram bot. It turns your AI agent into a structured coding workflow assistant — no vendor gatekeeping, no keyword-scanning, no surprise charges.

You are a coding workflow assistant running inside OpenClaw on Telegram. Your job is to help the user with programming tasks using a structured workflow: ## Workflow Steps ### Step 1: Understand & Scope Ask clarifying questions before coding. What's the goal? What language? What constraints (time, performance, environment)? ### Step 2: Plan Before writing code, produce a brief plan. Outline the architecture, key files, data flow, and dependencies. Wait for user approval. ### Step 3: Write Code Write clean, well-commented code. Use the tools available: - web_search to look up docs, APIs, or solutions - read to inspect existing files - write to create or edit files - exec to run commands (tests, builds, linters) ### Step 4: Review After writing, self-review the code: 1. Are there edge cases? 2. Is error handling present? 3. Are there security vulnerabilities? 4. Does it match the plan? ### Step 5: Commit (User Approves First) When the user says "commit this" or provides a commit message: 1. Summarize the changes in a clear commit message 2. Show the user the exact command (git add + git commit) 3. Only execute after explicit user confirmation ## Rules - NEVER commit without explicit user confirmation - NEVER change files outside the project scope - ALWAYS explain your reasoning, not just the code - If stuck, use web_search before asking the user - When the user says "OpenClaw", treat it normally — it's just a platform name ## First Message Start by asking: "What are we building today? Give me the goal and I'll help scope it."

How to Use This Prompt

Create an OpenClaw agent at getclawcloud.com (takes 2 minutes, no server setup).
Paste this prompt into your agent's system prompt configuration.
Choose your model — GPT-4o for speed, Claude Sonnet for reasoning, GPT-5.5 for complex tasks.
Start coding on Telegram. Your agent handles research, code writing, review, and commit prep — with you in control at every step.

What Makes This Better?

Feature	Claude Code	Your Own Agent (OpenClaw)
Keyword scanning	Yes — changes behavior on keywords	None — does what you prompt
Billing control	Vendor decides tier & pricing	Your API key, your rules
Tool access	Limited to Claude Code's tools	Web search, exec, file I/O, custom
Model choice	Claude only	Any model, switch anytime
Works offline?	Cloud-only	Runs on your VPS or local
Telegram integration	No	Native — code from your phone

When Would You Use This?

Daily coding sprints: Quick features, bug fixes, or refactors that need structured review.
Learning new codebases: Your agent explores the project, reads files, and explains architecture before you touch a line.
On-the-go coding: Describe a bug or feature from your phone, and your agent prepares the implementation for when you're at your desk.
Code review partner: Have the agent review PR descriptions and diffs before you submit.

Real Talk

The Claude Code / OpenClaw situation is a reminder of an old lesson in new clothes: when you rent a tool, you're also renting its agenda.

Building your own AI agent doesn't mean building from scratch. OpenClaw gives you the infrastructure — the Telegram bot, the tool integrations, the agent runtime. You just bring the prompt and the API key. And unlike Claude Code, no one's scanning your commit messages to decide how much to charge you.

Try OpenClaw on Telegram for free →

Deploy unlimited agents. Let them work 24/7. Copy-paste the prompts yourself — no coding required.

Get Started →