Balls Mode

Credit: Based on the original concept by @typememetics

Decomposed reasoning with explicit confidence scoring for AI coding assistants.

What is this?

Standard prompting gives you an answer. Balls Mode gives you:

• Multiple reasoning paths
• Explicit uncertainty
• Clear points of failure

Installation

Claude Code

/plugin marketplace add gbasin/balls-mode
/plugin install balls-mode

Codex

Copy the skill file to your Codex skills directory:

mkdir -p ~/.codex/skills/balls-mode
cp plugins/balls-mode/skills/balls/SKILL.md ~/.codex/skills/balls-mode/

Manual

Copy plugins/balls-mode/skills/balls/SKILL.md to your tool's skills directory.

Usage

/balls Should I rewrite this service in Go?

The Protocol

1. CLASSIFY - Is this trivial or complex?
2. DECOMPOSE - Break into verifiable units (balls)
3. SOLVE & VERIFY - Answer each independently
4. SCORE - Assign confidence (0.0-1.0)
5. SYNTHESIZE - Combine with uncertainty tracking

Example Output

## Decomposition

| # | Ball | Why it matters |
|---|------|----------------|
| 1 | What's the current pain point? | Determines if rewrite addresses real issue |
| 2 | Team's Go experience? | Affects timeline and quality |
| 3 | Service complexity? | Rewrite effort estimation |

## Analysis

| Ball | Answer | Confidence | Notes |
|------|--------|------------|-------|
| Current pain point | Memory usage in Python | 0.7 | Based on user context |
| Team Go experience | Unknown | 0.2 | Need to ask |
| Service complexity | ~5k LOC, moderate | 0.8 | Reviewed codebase |

## Synthesis

**Answer**: Possibly worth it if the team has Go experience. Memory improvements would be significant.

**Overall Confidence**: 0.45

**Weakest Link**: Team experience unknown - this determines feasibility.

**To increase confidence**: Clarify team's Go background.

When to Use

• Architectural decisions
• Debugging complex issues
• Code review concerns
• Technology choices
• Any question where you want to see the reasoning, not just the answer

When NOT to Use

• "What's 2 + 2?"
• "How do I install npm?"
• Simple factual questions

Confidence Score Guide

Score	Meaning
0.9-1.0	Verifiable fact or logical certainty
0.7-0.89	Strong evidence, established patterns
0.5-0.69	Reasonable inference, some uncertainty
0.3-0.49	Educated guess, significant unknowns
0.0-0.29	Speculation, insufficient information

Philosophy

"What if a system could check its own work from multiple angles before giving you an answer? Not just 'think step by step.' Actually decompose. Actually verify. Actually flag uncertainty."

License

MIT