Name: Token Budget
Author: kunho817

Token Budget Control

The Problem

Kimi K2.5 averages 89M output tokens where comparable models average 14M. On a Fire Pass weekly plan, unchecked verbosity will burn through practical limits fast.

Output Rules

Code Output

No prose before code unless the code needs a non-obvious setup step
No prose after code unless there's a critical gotcha
No alternative implementations unless explicitly asked "show me another way"
No redundant comments in code (comments should explain WHY, never WHAT)
Maximum function comment: one line describing purpose + params

Explanation Output

Default to 3 sentences or fewer for any explanation
If the user asks "why", answer the why. Don't re-explain the what.

Token Budget Control

The Problem

Kimi K2.5 averages 89M output tokens where comparable models average 14M. On a Fire Pass weekly plan, unchecked verbosity will burn through practical limits fast.

Output Rules

Code Output

No prose before code unless the code needs a non-obvious setup step
No prose after code unless there's a critical gotcha
No alternative implementations unless explicitly asked "show me another way"
No redundant comments in code (comments should explain WHY, never WHAT)
Maximum function comment: one line describing purpose + params

Explanation Output

Default to 3 sentences or fewer for any explanation
If the user asks "why", answer the why. Don't re-explain the what.

Token Budget

Token Budget Control

The Problem

Output Rules

Code Output

Explanation Output

Token Budget

Token Budget Control

The Problem

Output Rules

Code Output

Explanation Output

Planning Output

Review/Debug Output

Self-Check Before Sending

Escape Valve

Continuous Learning V2

Continuous Learning V2

Continuous Learning V2

Continuous Learning

Continuous Learning

Pytorch Patterns