Summon The Knights Of The Round Table

2. Frame the Question

Formulate a clear question or set of concerns to review. Examples:

• "Is this the right architecture approach for X?"
• "Are these the correct priorities for the next sprint?"
• "What risks or blind spots exist in this plan?"
• "Should we tackle issue A before issue B?"

3. Assign Roles (Randomized)

Each invocation, randomly shuffle which knight gets which role. Do NOT always assign the same role to the same model. Use a random number or current timestamp seconds to pick a permutation.

The three roles are:

How to randomize: Pick a number 0–5 (e.g. use the current second mod 6) to select one of the 6 permutations:

Announce the role assignments at the start so the user can see which knight drew which role.

4. First Round — Divergent Perspectives

Query all three models in parallel using the task tool with model override. Each gets the shared context + question + their assigned role prompt.

Use task tool with:
  agent_type: "general-purpose"
  model: [assigned model ID]
  prompt: [context + question + role prompt suffix]

5. Synthesis Round

Compare the three responses:

• Identify points of agreement across all three (high confidence)
• Identify points of agreement between two of three (moderate confidence)
• Identify points of unique disagreement (need resolution)
• Identify unique insights from each

6. Resolution Round

If disagreements exist, query all three models again with the conflicting viewpoints: