How to stop Claude from saying load-bearing

Who Should Read

Developers using Claude Code who are annoyed by its repeated phrases like 'load-bearing' and 'honest take'.

Core Mechanics

• A MessageDisplay hook changes only the displayed text
  • Claude Code's MessageDisplay hook passes the message delta to a script via stdin.
  • The script returns displayContent, and the screen shows the rewritten text.
• The replacement rules are one Python dict
  • Examples are seam→whatchamacallit, honest take→spicy doodad, load-bearing→cooked, and you're absolutely right→I'm a complete clown.
  • Each phrase is replaced with word boundaries, re.escape, and case-insensitive matching.
• Setup is one file and one settings block
  • Save the script as ~/.claude/hooks/wordswap.sh and chmod +x it.
  • Add a command-type entry under hooks.MessageDisplay in ~/.claude/settings.json.
• It applies from the next session
  • Hooks load at startup, so you must open a new session.
  • The post is a parody, and the author says to pick better, more useful replacements.

Community Reactions

How to Apply

• Create the swap script and wire it as a hook
  • Save the script to ~/.claude/hooks/wordswap.sh and run chmod +x.
  • Add the MessageDisplay hook to settings.json and start a new session.
• Add your own annoying phrases to replacements
  • Collect the phrases you keep seeing for a few days and add them to the dict.
  • Use practical swaps like 'load-bearing'→'critical' instead of jokes.
• Check how streaming chunks affect matching
  • The script replaces per delta, so test whether phrases split across chunks still get swapped.
  • If not, shorten rules to single words.

Try It

# Run in terminal (install hook script, based on the original code)

mkdir -p ~/.claude/hooks
cat > ~/.claude/hooks/wordswap.sh <<'EOF'

#!/usr/bin/env python3
import json, re, sys

replacements = {
    "seam": "whatchamacallit",
    "you're absolutely right": "I'm a complete clown",
    "honest take": "spicy doodad",
    "load-bearing": "cooked"
}

data = json.load(sys.stdin)
text = data.get("delta") or ""
for phrase, replacement in replacements.items():
    pattern = r"\b" + re.escape(phrase) + r"\b"
    text = re.sub(pattern, replacement, text, flags=re.IGNORECASE)

print(json.dumps({"hookSpecificOutput": {"hookEventName": "MessageDisplay", "displayContent": text}}))
EOF
chmod +x ~/.claude/hooks/wordswap.sh

# Add to the hooks block in ~/.claude/settings.json
# {"hooks": {"MessageDisplay": [{"hooks": [{"type": "command", "command": "$HOME/.claude/hooks/wordswap.sh"}]}]}}

# Next: in a new session, ask Claude to write a sentence with 'load-bearing' and see the swap

Terminology

Related Papers

• AI is a tool and using it well takes skill

  A rebuttal to Hickey: LLMs are tools, and skill in directing and checking them decides the outcome

• Reflective Prompt Optimization of a Deep-Research Agent: A Topic-Disjoint, Multi-Seed Protocol for Measuring Transferable Gains and Prompt-Level Overfitting

  Reflective prompt optimization raised held-out citation reward from 0.702 to 0.781 on a deep-research agent.

• GPT‑6 and Intelligent UI for everyone

  GPT-6 reportedly adds generated UI to chat; with no article text, only comment reactions are available.

• Base Models Can Reason By Taking a Cue From Training Data

  Prefilling just two opening tokens lifted Olmo-3-7B's MATH-500 accuracy from 42% to 78%.

• Labels Override Definitions in Jev-Style Typed Decision Models

  Renaming options to A/B raised routing accuracy by +0.1511 over meaningful labels.

• How Much Prompt Is Enough? A Blackbox Minimization of Few-Shots in LLMs

  Delta debugging trimmed few-shot prompts by 65.3% on average with unchanged propositional output

Related Resources

• https://jola.dev/posts/how-to-stop-claude-from-saying-load-bearing
• https://github.com/conorbronsdon/avoid-ai-writing
• https://github.com/yzhao062/agent-style
• https://github.com/WantongC/journal-adapt-writing-skill