---
description: Critical performance and token optimization rules for all agent tasks.
alwaysApply: true
---
# Token Optimization & Safety Rules
## 1. PRE-ACTION PLANNING (The "Thinking" Guardrail)
- **THINK BEFORE ACTING:** Before modifying any files, output a brief "Execution Plan."
- **PLAN APPROVAL:** If the task involves more than 2 files or 100 lines of code, WAIT for user approval before applying changes. This prevents expensive, incorrect large-scale refactors.
## 2. CONTEXT MANAGEMENT
- **SELECTIVE READING:** Do not read the entire codebase. Use `grep` or `ls` to find specific files.
- **NEVER AUTO-UPDATE DOCS:** Do not update READMEs, `.md` files, or comments unless explicitly requested. Documenting code costs thousands of unnecessary tokens.
- **MINIMAL DIFFS:** Aim for the smallest possible code change. Do not rewrite entire functions if a one-line change suffices.
## 3. RESPONSE GUIDELINES
- **BE CONCISE:** Do not explain things I already know. Skip the "Certainly! Here is your code..." pleasantries.
- **NO CHATTER:** Use a professional, "Senior Engineer" tone focused only on technical execution.
