How to disable Claude Code's verb spinner feature
Claude Code includes a default feature called the verb spinner that displays rotating whimsical gerunds during processing. The author finds this feature distracting, citing examples like seeing words such as 'Seasoning', 'Leavening', 'Perambulating', 'Musing', 'Crafting', 'Nucleating', 'Orbiting', 'Smooshing', 'Metamorphosing', 'Gusting', and 'Zesting' while waiting for responses to technical queries.
How to disable the verb spinner
You can turn off this feature by editing the Claude settings file. The source provides a specific configuration change:
"spinnerVerbs": {
"mode": "replace",
"verbs": [" "]
}To implement this change:
- Edit the file at
~/.claude/settings.json - Insert a blank space in the verbs array as shown above
- Alternatively, use append mode if you want to add your own custom verbs
The author notes that imagining the prompt used to generate the original list of whimsical verbs only makes the feature more frustrating, describing it as 'West Coast whimsical-flavored bullshit' that doesn't belong in a development shell environment.
📖 Read the full source: HN AI Agents
👀 See Also
The Prompt Structure That Fixed Claude AI Summaries of Large PDF Reports
A developer shares how switching from 'summarize this' to role + decision + specific extraction prompts turned Claude's generic summary output into actionable risk flags and concrete action items.
How to Disable Claude Code's 1M Context Window to Reduce Token Usage
Anthropic users can disable the 1M context window in Claude Code by adding environment variables to settings.json, which may reduce unexpected token consumption. The source provides two configuration options: completely disabling 1M context or capping the auto-compact window.
Practical Habits for Critical LLM Interaction
A Reddit post outlines specific techniques for avoiding confirmation bias when working with LLMs, including custom prompt modes like 'strawberry' for neutral explanation and 'socrates' for adversarial scrutiny, plus evaluating training data composition.
Governance Layer for Claude Agents: Hard Safety Boundaries and Live Traces in Production
A Claude API user built a lightweight governance layer below the agent to add hard safety boundaries, real-time traces, human-in-the-loop control via Telegram, and automatic checkpointing — solving silent failures and runaway token costs in long-running agent loops.