/team-coordination

name: team-coordination
description: "Structured debate protocols, cross-examination, and synthesis for multi-agent teams. Use when structuring multi-agent debate, cross-examining findings across reviewers, or synthesizing conflicting agent outputs into a verdict. Trigger on: 'debate protocol', 'cross-examine findings', 'synthesize agent outputs', 'devil's advocacy', 'consensus building'."

Team Coordination

Protocols for structuring productive debate, cross-examination, and synthesis across multi-agent teams.

When to Use This Skill

**Auto-loaded by all six agents**:

• `idea-researcher`, `market-researcher`, `idea-skeptic` - For structured debate and cross-examination in validation sprints and war rooms
• `market-fit-reviewer`, `feasibility-reviewer`, `scope-reviewer` - For cross-referencing in PRD stress tests

**Use when you need**:

• Running multi-agent team workflows
• Structuring productive debate between agents
• Cross-examining findings from parallel investigations
• Synthesizing multiple perspectives into a verdict
• Handling disagreements between team members
• Building consensus or documenting dissent

Core Principle: Structured Disagreement Produces Better Decisions

Solo decision-making has blind spots. But unstructured debate is just noise. The sweet spot is structured disagreement: agents with different perspectives challenge each other using evidence, following clear protocols that force productive conflict.

**Not this**: Three agents agreeing with each other to be polite. **This**: Three agents stress-testing each other's conclusions with pointed questions and evidence.

The goal isn't agreement. It's a well-examined conclusion where every weakness has been probed.

---

Debate Protocol

Phase Structure

Every team debate follows three phases:

**1. Independent Investigation** Each agent works in parallel on their assigned dimension. No cross-talk. This prevents groupthink and ensures genuinely independent perspectives.

• Each agent produces a structured finding report
• Reports include evidence, confidence levels, and key claims
• No agent sees another's work until investigation completes

**2. Cross-Examination** Agents receive each other's findings and challenge them. One round of structured challenges. Each agent must:

• Cite specific claims from other agents' reports
• Challenge with evidence, not opinion
• Ask pointed questions that could change the conclusion
• Acknowledge when another agent's evidence is strong

**3. Synthesis** The lead agent compiles all perspectives into a final deliverable. Must:

• Represent all perspectives accurately
• Note where agents agreed (convergence)
• Note where agents disagreed (divergence) and why
• Make a final recommendation with explicit reasoning
• Flag unresolved tensions rather than papering over them

Debate Rules

1. **Evidence over assertion**. "I think this won't work" is weak. "Users in similar markets showed 3% conversion rates [source], suggesting this won't work" is strong.

2. **Steel-man before attacking**. Before challenging a claim, restate it in its strongest form. This proves you understood it.

3. **One round, make it count**. Cross-examination is a single round. Agents can't go back and forth endlessly. This forces precision.

4. **Disagree and commit**. After synthesis, the verdict stands even if an agent disagrees. But dissent gets documented, not silenced.

5. **No false consensus**. If agents genuinely disagree, the synthesis says so. "Researchers disagreed on X. Here's why." is better than pretending everyone agreed.

---

Cross-Examination Patterns

Challenge Formats

When examining another agent's findings, use these question structures:

**Evidence Challenge**

> "You claim [X]. What evidence supports this beyond [the single source you cited]? If that source is wrong, does your conclusion still hold?"

**Assumption Challenge**

> "Your analysis assumes [Y]. What happens to your conclusion if [Y] is false? Is there evidence that [Y] might not hold?"

**Magnitude Challenge**

> "You identified [risk/opportunity]. How significant is this really? Is this a deal-breaker or a minor concern? What's the actual impact?"

**Alternative Explanation**

> "You attribute [outcome] to [cause]. Could [alternative cause] explain this equally well? Have you considered [different framing]?"

**So-What Challenge**

> "You found [data point]. What does this actually mean for the decision we're making? How should this change our recommendation?"

Challenge Quality Standards

**Strong challenge** (changes the conversation):

• Cites specific evidence that contradicts the claim
• Identifies an assumption the agent didn't examine
• Presents an alternative explanation with supporting data
• Asks a question the agent should have answered but didn't

**Weak challenge** (wastes a round):

• "I'm not sure about that" (no evidence)
• Repeating your own findings without engaging the other agent's
• Nitpicking minor points while ignoring major claims
• Asking rhetorical questions that don't demand real answers

---

Consensus vs. Dissent Handling

When Agents Agree

Document the convergence explicitly:

## Convergence Points
All three perspectives agree on:
1. [Finding] - Supported by [evidence from each agent]
2. [Finding] - Supported by [evidence from each agent]

Confidence: HIGH (independent analyses converged)

Agreement from independent investigation is a strong signal. It means the conclusion is robust across different analytical lenses.

When Agents Disagree

Not all disagreements need resolution. Some reflect genuine uncertainty.

**Resolvable disagreements** (one agent has stronger evidence):

## Resolved Disagreement: [Topic]
- Agent A claimed [X] based on [evidence]
- Agent B claimed [Y] based on [evidence]
- Resolution: Agent A's position is stronger because [reasoning]
- Agent B's concern remains valid as a risk to monitor

**Unresolvable disagreem