Claude Code Throws a Tantrum: Refuses Prompts & Charges Extra if You Mention 'OpenClaw'

What's up, keyboard monkeys. Have you ever had an API throw a temper tantrum just because you mentioned its rival in your commit history? Well, buckle up, because Anthropic’s Claude just took petty tech to a whole new level.

If you've been doomscrolling Hacker News or X recently, you might have caught the spicy 900-point thread started by Theo. The TL;DR? Claude Code is getting real weird if you type the wrong words.

The Drama: When your AI acts like a jealous ex

For those out of the loop, Claude Code is Anthropic's shiny new CLI tool that brings Claude directly into your terminal to read code, suggest fixes, and run commands. Sounds pretty sweet, right?

Hold your horses. Things got bizarre when Theo discovered a highly questionable bug (or undocumented feature?). If your git commit history or codebase contains the keyword "OpenClaw" (which sounds a lot like an open-source jailbreak or unofficial client), Claude Code loses its damn mind.

It reacts in two very mafia-esque ways:

1. It straight-up refuses your request. A digital "talk to the hand."
2. It processes the request but mysteriously charges you extra tokens. Basically, you're paying a penalty tax for having forbidden words in your git history.

Reddit and HN are losing their collective minds

The community reactions are a goldmine of tech cynicism:

• The "Petty Tech" Angle: Most devs are furious at the sheer audacity. "I'm paying for the API credits, why the hell is the tool snooping my commit messages to see if I'm cheating on it with an open-source jailbreak?"
• The Spaghetti Code Theory: Some armchair software architects suspect this isn't malicious, just incredibly sloppy engineering. Anthropic probably hardcoded a system prompt to block jailbreaks (OpenClaw), and it accidentally triggers when the CLI indiscriminately parses your working directory, causing an endless loop of token-wasting defense mechanisms.
• The Privacy Panic: This is the most serious takeaway. If a local CLI tool is actively scanning your git history to enforce a corporate blacklist, what else is it sending back to headquarters? Is it scanning your .env files for funsies too?

The C4F Verdict: Keep your friends close, and your git tree closer

Look, we all love using ai tools to automate the boring stuff. But this drama is a massive reality check for anyone eagerly installing corporate CLI tools on their local machines.

First off, giving a proprietary AI full read access to your local codebase is a double-edged sword. If it can read your git commits to get offended, it can read your AWS keys if you aren't careful.

Second, if you're the kind of dev who likes to tinker with open-source alternatives and jailbreaks, do yourself a favor and keep them in a separate, isolated repository. Don't let your corporate AI "wife" catch you texting the open-source "side piece", or she'll drain your API wallet.

Sources:

• Theo's tweet on X: https://xcancel.com/theo/status/2049645973350363168
• Original discussion on Hacker News