context_mastery（语境掌握）

Context Mastery & Token Optimization Protocol

1. The "Grep-First" Rule (Token Saver)

• Problem: read_file on a 2000-line file consumes ~500-1000 tokens instantly.
• Solution: Always use grep or read_file_range first to inspect specific functions or imports.
• Banned: cat or read_file on entire directories or massive files (e.g., package-lock.json, giant logs) unless absolutely necessary.

2. Incremental Summarization

• Trigger: When the conversation exceeds ~20 turns or you feel the context window filling.
• Action:

1. Summarize what has been achieved in the last 10 turns. 2. Write it to scratchpad.md or active_task.md. 3. Explicitly state "I am clearing internal context of validated steps" (if tool allows) or just rely on the scratchpad for future lookup.

3. Focused Context Loading

• Don't: "Read all files in src/ to understand the project." (Too expensive).
• Do:

1. Read CLAUDE.md / README.md. 2. Read directory listing ls -R. 3. Selectively read *only* the interfaces (types.ts) or entry points (index.ts) related to the current task.

4. Context Efficiency Checklist

• Did I use grep instead of reading the whole file?
• Did I check if I already have this info in MEMORY.md?
• Is my next prompt efficient? (Avoid repeating massive code blocks back to the user).