breakdown

name	breakdown
description	Break down complex problems before implementation using multi-model reasoning and prompt techniques
user-invocable	true

Breakdown - Problem Decomposition

Decompose complex tasks using first principles, pattern analysis, and multi-model reasoning before implementation. Adapts to your available API keys.

Usage

/breakdown [problem or task description]

When to Use

• Before implementing non-trivial features
• When requirements are unclear or ambiguous
• When facing multiple possible approaches
• Before major refactoring or architectural changes

Instructions

When user invokes /breakdown [problem]:

Step 0: Check Available Tools

The pipeline adapts to available tools. Each step lists a priority chain - use the FIRST available tool.

Step 1: First Principles (Atomic Truths)

Use the FIRST available:

1. mcp__tachibot-mcp__execute_prompt_technique({ technique: "first_principles", tool: "grok_reason", query: "[problem]" })
2. mcp__tachibot-mcp__execute_prompt_technique({ technique: "first_principles", tool: "openai_reason", query: "[problem]" })
3. mcp__tachibot-mcp__execute_prompt_technique({ technique: "first_principles", tool: "gemini_analyze_text", query: "[problem]" })

If execute_prompt_technique is unavailable, call the reasoning tool directly with the prompt: "Apply first principles thinking to: [problem]. Strip to fundamental truths, challenge assumptions, identify atomic units."

Extract: Core truths, assumptions to challenge, atomic units.

Step 2: Decompose (Sub-problems & Dependencies) — Least-to-Most

Use the FIRST available:

1. mcp__tachibot-mcp__execute_prompt_technique({ technique: "least_to_most", tool: "kimi_thinking", query: "..." })
2. mcp__tachibot-mcp__execute_prompt_technique({ technique: "least_to_most", tool: "kimi_decompose", query: "..." })
3. mcp__tachibot-mcp__execute_prompt_technique({ technique: "decompose", tool: "gemini_analyze_text", query: "..." })
4. mcp__tachibot-mcp__execute_prompt_technique({ technique: "decompose", tool: "openai_reason", query: "..." })

Feed step 1 output into the query. Use least_to_most to order sub-tasks from simplest → hardest (atomic parts first, complex parts last — each builds on previous solutions).

Extract: Sub-tasks ordered simplest→hardest, dependency graph, execution order.

Step 3: Patterns (Causality & Cycles)

Use the FIRST available:

1. mcp__tachibot-mcp__execute_prompt_technique({ technique: "patterns", tool: "gemini_analyze_text", query: "..." })
2. mcp__tachibot-mcp__execute_prompt_technique({ technique: "patterns", tool: "grok_reason", query: "..." })
3. mcp__tachibot-mcp__execute_prompt_technique({ technique: "patterns", tool: "openai_reason", query: "..." })

Feed step 2 output into the query.

Extract: Hidden connections, recurring themes, potential cycles/loops.

Step 4: Pre-Mortem + Feasibility (Reality Check)

Use the FIRST available:

1. mcp__tachibot-mcp__execute_prompt_technique({ technique: "pre_mortem", tool: "grok_reason", query: "..." })
2. mcp__tachibot-mcp__execute_prompt_technique({ technique: "pre_mortem", tool: "openai_reason", query: "..." })
3. mcp__tachibot-mcp__execute_prompt_technique({ technique: "feasibility", tool: "gemini_analyze_text", query: "..." })

Use pre_mortem when available: "Assume this failed 6 months from now — why?" This catches failure modes that forward analysis misses (Klein 2007). Fall back to feasibility if pre_mortem isn't available.

Extract: Failure causes ranked by likelihood, early warning signs, mitigations, go/no-go.

Step 5: Synthesize Output

Present structured breakdown:

## Problem
[One sentence]

## First Principles
- [Truth 1]
- [Truth 2]
- Challenged assumption: [X]

## Sub-Tasks
1. **[Task A]** - [purpose]
   - Depends on: none
   - Risk: low

2. **[Task B]** - [purpose]
   - Depends on: Task A
   - Risk: medium

## Patterns Found
- Causality: [X causes Y]
- Cycle: [A -> B -> C -> A]
- Anomaly: [unexpected finding]

## Execution Order (simplest → hardest)
1. [Simplest] - no deps, builds foundation
2. [Next] - builds on 1
...

## Feasibility
| Aspect | Status | Notes |
|--------|--------|-------|
| Technical | OK | [details] |
| Time | Warning | [constraint] |
| Resources | OK | [details] |

## Risks & Mitigations
| Risk | Likelihood | Mitigation |
|------|------------|------------|
| [Risk 1] | Medium | [approach] |

## Verdict
[Go / No-go / Needs more info]

Examples

• /breakdown implement OAuth authentication with refresh tokens
• /breakdown refactor monolith payment module to microservice
• /breakdown add real-time collaboration to document editor