---

name: tot description: Tree-of-thoughts reasoning - explore multiple branches of thought and evaluate different perspectives. version: "1.0.0" status: stable category: reasoning triggers:

• "/tot"
• "tree of thoughts"
• "explore multiple approaches"
• "consider different perspectives" aliases:
• /tot workflow_steps:
• parse_input
• spawn_branches
• evaluate_branches
• return_recommendation

---

Tree-of-Thoughts Reasoning

When to Use

• Complex problems requiring multiple approaches
• High-stakes decisions where single-point reasoning is risky
• Creative tasks where lateral thinking helps
• Architectural decisions with competing constraints
• Debugging complex issues where multiple hypotheses exist

Instructions

Just tell me what you want to reason about. I'll spawn multiple subagents to explore different approaches in parallel, then evaluate which branch is most reliable.

Workflow

Step 1: Parse Input

Extract the core problem or question from your request.

Step 2: Spawn Branches

Create 3-5 parallel subagents, each exploring a different reasoning approach:

• Analytical Branch: Step-by-step logical decomposition
• Creative Branch: Lateral thinking, novel solutions
• Skeptical Branch: Critique-first, find flaws in assumptions
• Pragmatic Branch: Focus on practical implementation
• Synthesis Branch: Integrate multiple perspectives

Step 3: Evaluate Branches

Each branch returns:

• Approach description
• Key insights
• Confidence score (0-1)
• Recommendations

I evaluate using:

• Self-consistency (do branches converge?)
• Evidence quality (backed by verification?)
• Risk assessment (what could go wrong?)

Step 4: Return Recommendation

Synthesize the highest-confidence approach and present:

• Recommended path with rationale
• Alternative considerations
• Confidence score
• Next steps

Research Basis

Based on "Can AIs Like ChatGPT Think?" (ai-consciousness.org):

• 18× improvement on Game of 24 (4% → 74% success)
• Parallel branch exploration + self-consistency evaluation
• Outperforms single-threaded chain-of-thought

Configuration

Environment Variables:

• TOT_ENABLED=true - Enable/disable Tree-of-Thoughts (default: true)
• TOT_BRANCHES=3 - Number of parallel branches (default: 3)
• TOT_TIMEOUT=300 - Timeout per branch in seconds (default: 300)

Examples

User: /tot Should I use Redis or Memcached for caching?

Response:

Exploring 3 reasoning branches in parallel...

Branch 1 (Analytical): Performance comparison...
Branch 2 (Pragmatic): Implementation complexity...
Branch 3 (Skeptical): Operational overhead...

RECOMMENDATION: Redis (confidence: 0.8)
- Reasons: Data structures, persistence, ecosystem
- Alternative: Memcached for pure read-heavy workloads
- Next step: Prototype Redis connection pooling

Implementation Notes

This skill uses the tot_core.py module for the Tree-of-Thoughts reasoning engine. The core module provides:

• ThoughtBranch dataclass for branch representation
• TreeOfThoughts class with explore_branches() and evaluate_branches() methods
• Agent tool integration for parallel subagent spawning
• Async/await support for parallel execution

See Also

• Self-reflection gate (Phase 1): Detects low-confidence claims
• Chain-of-Draft (Phase 3): Optimizes verbose reasoning