graph-of-thoughts
INVOKE after research or brainstorming to synthesize findings. Produces visible aggregation with conflict resolution. Use when multiple inputs need combining into coherent output. Triggers: synthesizing research, combining ideas, merging approaches, integrating findings.
When & Why to Use This Skill
The Graph of Thoughts (GoT) skill is an advanced reasoning framework designed to synthesize complex information from multiple sources into a single, coherent output. By modeling thoughts as a graph rather than a linear chain, it enables sophisticated operations like aggregation, iterative refinement, and conflict resolution. This skill is essential for transforming divergent research or brainstorming sessions into structured, actionable insights with clear provenance.
Use Cases
- Research Synthesis: Aggregating findings from multiple academic papers or market reports to identify consensus and resolve contradictory data points.
- Strategic Planning: Consolidating diverse brainstormed ideas and partial solutions into a unified, logical project roadmap.
- Technical Architecture: Merging different engineering approaches or code snippets into a single, optimized system design while documenting the reasoning for each choice.
- Iterative Document Refinement: Improving complex drafts through structured feedback loops, self-critique, and recursive aggregation of improvements.
| name | graph-of-thoughts |
|---|---|
| description | "INVOKE after research or brainstorming to synthesize findings. Produces visible aggregation with conflict resolution. Use when multiple inputs need combining into coherent output. Triggers: synthesizing research, combining ideas, merging approaches, integrating findings." |
Graph of Thoughts (GoT)
Models reasoning as an arbitrary graph with aggregation, refinement, and feedback operations.
MUST Invoke When
- Synthesizing research from multiple sources
- Combining brainstormed ideas into a coherent plan
- User asks to "combine", "synthesize", or "integrate" findings
- Merging partial solutions from different approaches
- After divergent exploration (ToT) when convergence is needed
- Multiple inputs need to become one coherent output
Output Commitment
This skill produces visible structured output:
- Extracted insights from each source
- Identified agreements and conflicts
- Resolution of conflicts with reasoning
- Unified synthesis with provenance
Do NOT just pick one input—invoke this skill to show synthesis process.
Core Mechanism
Unlike trees (divergent) or chains (linear), graphs support:
- Aggregation: Combining multiple thoughts into one
- Refinement: Iteratively improving a thought
- Splitting: Breaking a thought into components for parallel processing
- Loops: Feedback cycles for iterative improvement
Process
1. Identify input thoughts/sources to synthesize
2. For each input, extract key insights and constraints
3. Identify conflicts or tensions between inputs
4. Aggregate compatible insights into unified positions
5. Resolve conflicts through:
- Prioritization (which source is more authoritative?)
- Synthesis (can both be true in different contexts?)
- Refinement (iterate until coherent)
6. Output synthesized result with clear provenance
Operations
Aggregate
Combine multiple reasoning chains into a single coherent output:
Chain A conclusion: X
Chain B conclusion: Y
Chain C conclusion: Z
→ Aggregated insight: [unified position incorporating X, Y, Z]
Refine
Iteratively improve a thought through feedback:
Draft 1 → Critique → Draft 2 → Critique → Final
Split-then-Aggregate
Parallelize then recombine:
Complex problem → Split into A, B, C → Solve each → Aggregate solutions
When to Apply
- Synthesizing research from multiple sources
- Combining brainstormed ideas into a coherent plan
- Merging partial solutions from different approaches
- Iterative refinement through self-critique
- Any task where insights must converge rather than diverge
Synthesis Pattern
I have these inputs to synthesize:
- Source 1: [insight]
- Source 2: [insight]
- Source 3: [insight]
Step 1 - Extract: What is the core claim/finding from each?
Step 2 - Align: Where do they agree?
Step 3 - Conflict: Where do they disagree? Why?
Step 4 - Resolve: For each conflict, determine resolution
Step 5 - Integrate: Produce unified output with:
- Synthesized position
- Confidence level
- Remaining uncertainties
Anti-Patterns
- Using when inputs don't need integration (just present separately)
- Forcing false synthesis when sources genuinely conflict
- Losing provenance (which insight came from where)
- Aggregating without resolving contradictions