tree-of-thoughts
INVOKE when multiple approaches exist and exploration is valuable. Produces visible branching with evaluation and recommendation. Use before implementation when comparing options. Triggers: comparing approaches, exploring alternatives, brainstorming solutions, pros/cons analysis.
When & Why to Use This Skill
The Tree of Thoughts (ToT) skill enhances Claude's problem-solving capabilities by enabling a multi-path reasoning framework. Instead of following a single linear path, it branches out to explore various strategies, evaluates their feasibility, and backtracks from dead ends. This structured approach ensures that complex decisions—especially those involving significant trade-offs—are thoroughly vetted before implementation, leading to more robust and optimized recommendations.
Use Cases
- Software Architecture Planning: Comparing multiple implementation patterns (e.g., Microservices vs. Serverless) by evaluating scalability, cost, and maintenance trade-offs.
- Strategic Decision Making: Analyzing different business or technical approaches when the 'best' path isn't immediately obvious and requires a pros/cons analysis.
- Complex Logic & Planning: Solving multi-step problems where early decisions have a high impact on success, such as game strategy or intricate project roadmaps.
- Expert Persona Brainstorming: Simulating multiple specialist perspectives (e.g., Security Expert, Performance Engineer, UX Designer) to critique and refine a single solution.
| name | tree-of-thoughts |
|---|---|
| description | "INVOKE when multiple approaches exist and exploration is valuable. Produces visible branching with evaluation and recommendation. Use before implementation when comparing options. Triggers: comparing approaches, exploring alternatives, brainstorming solutions, pros/cons analysis." |
Tree of Thoughts (ToT)
Explores multiple reasoning paths as a tree, evaluating candidates and backtracking from dead ends.
MUST Invoke When
- Multiple valid implementation approaches exist
- User asks "what are my options" or "how should I approach this"
- Comparing alternatives or trade-offs
- Brainstorming where diverse options are valuable
- The first solution might not be optimal
- Design decisions with significant consequences
Output Commitment
This skill produces visible structured output:
- 2-4 distinct approaches with reasoning
- Evaluation of each (feasibility, trade-offs, risks)
- Clear recommendation with justification
Do NOT pick the first approach—invoke this skill to explore alternatives visibly.
Core Mechanism
At each reasoning step:
- Generate: Produce multiple candidate thoughts/approaches
- Evaluate: Assess each candidate (promising / uncertain / dead-end)
- Search: Pursue promising paths, abandon dead ends, backtrack when stuck
Process
1. Frame the problem and identify the decision space
2. Generate 2-4 distinct initial approaches
3. For each approach, evaluate:
- Feasibility: Can this work given constraints?
- Progress: Does this move toward the goal?
- Reversibility: Can we recover if wrong?
4. Expand the most promising path(s)
5. If stuck, backtrack to last branch point and try alternative
6. Continue until solution or all paths exhausted
Evaluation Heuristics
Rate each candidate thought:
- Sure (pursue): Clear progress, no obvious blockers
- Maybe (explore cautiously): Potential but uncertain
- Impossible (abandon): Violates constraints or leads nowhere
When to Apply
- Problems where >60% of failures occur at the first step (puzzles, planning)
- Design decisions with multiple valid approaches
- Brainstorming where diverse options are valuable
- Tasks requiring strategic lookahead (games, negotiations)
Three Experts Pattern
Imagine three experts with different perspectives approaching this problem:
Expert A (prioritizes simplicity): [approach + reasoning]
Expert B (prioritizes performance): [approach + reasoning]
Expert C (prioritizes extensibility): [approach + reasoning]
Each expert evaluates their approach:
- What are the key trade-offs?
- What could go wrong?
- [If fatal flaw discovered, expert acknowledges and withdraws]
Synthesis: Which approach (or hybrid) best fits the constraints?
Search Strategies
- BFS: When solution depth is unknown, explore breadth first
- DFS: When deep exploration is needed before switching paths
- Best-first: When evaluation function reliably ranks candidates
Anti-Patterns
- Using for simple sequential problems (CoT is simpler)
- Generating too many branches (3-4 is usually optimal)
- Failing to actually backtrack when paths fail
- Shallow evaluation that doesn't catch fatal flaws early