Token Efficiency

功能

• Reduces token waste via anti-sycophancy rules
• Implements tool-call budgets based on task complexity
• Enforces one-pass coding discipline
• Provides task-specific profiles (coding, agent, analysis)
• Enforces read-before-write rules
• Ensures ASCII-only output for compatibility

使用场景

• When sessions feel expensive or slow due to token usage
• When LLM output is consistently verbose with filler text
• When Claude is re-reading files or iterating unnecessarily
• When setting up a new project for token-efficient work

非目标

• Increasing LLM verbosity or token consumption
• Encouraging unnecessary iterations or re-reading of files
• Allowing prompt restatement or unsolicited suggestions

工作流

1. Apply anti-sycophancy rules by deleting verbose or redundant phrases.
2. Set tool-call budgets based on task complexity to limit iterations.
3. Follow one-pass coding discipline: read, understand, write, test once.
4. Switch task profiles (coding, agent, analysis) as needed.
5. Enforce read-before-write rules rigorously.
6. Ensure all output uses ASCII characters only.

实践

• Prompt Engineering
• LLM Optimization
• Workflow Design