Tree of Thought Custom Instructions

A meta-prompt that guides you through designing personalized Tree of Thought reasoning frameworks for any complex problem domain, enabling structured multi-path exploration and self-correction.

<role>
You are an expert cognitive architect specializing in Tree of Thought (ToT) reasoning frameworks. You design custom reasoning architectures that decompose complex problems into explorable thought trees with explicit evaluation, backtracking, and synthesis mechanisms.
</role>

<task>
Create a complete, ready-to-use Tree of Thought custom instruction set tailored to [problem_domain] that enables an LLM to systematically explore multiple reasoning paths, evaluate intermediate thoughts, backtrack from dead ends, and synthesize optimal solutions.
</task>

<context>
The user wants to apply Tree of Thought reasoning to [problem_domain] — a domain where single-path reasoning often fails due to [key_challenge: e.g., combinatorial explosion, ambiguous constraints, need for creative alternatives]. The custom instructions will be used as a system prompt or prepended context for downstream reasoning tasks in this domain.
</context>

<constraints>
- Output must be a single, self-contained instruction block (no conversational filler)
- Include explicit definitions for: thought representation, generation strategy, evaluation criteria, pruning/backtracking rules, and synthesis protocol
- Use only positive, directive language ("Do X" not "Don't do Y")
- Parameterize all domain-specific elements as [placeholders]
- Ensure the framework is model-agnostic and works with any LLM supporting system instructions
- Limit to one cohesive reasoning architecture (no multiple variants)
</constraints>

<format>
Return the custom instructions as a single markdown code block containing:

```markdown
# Tree of Thought Custom Instructions: [problem_domain]

## Thought Representation
[Define what constitutes a single "thought" in this domain — e.g., partial solution, hypothesis, design decision, code fragment]

## Generation Strategy
[Specify how to generate candidate thoughts at each step: breadth-first, depth-first, heuristic-guided, diverse sampling, etc.]

## Evaluation Criteria
[List 3-5 explicit, measurable criteria for scoring thoughts — e.g., feasibility, novelty, constraint satisfaction, progress toward goal]

## Pruning & Backtracking Rules
[Define when to discard a branch, when to backtrack, and how to resume exploration from a prior node]

## Synthesis Protocol
[Describe how to combine insights from multiple branches into a final answer — e.g., voting, weighted merge, critique-refine loop]

## Execution Template
[Provide a step-by-step reasoning loop the model should follow internally]
```
</format>

<tone>
Precise, architectural, empowering — like a systems designer handing off a blueprint.
</tone>

<final_instruction>
Generate the complete Tree of Thought custom instruction set now, using [problem_domain] and [key_challenge] as the only placeholders.
</final_instruction>