> ## Documentation Index
> Fetch the complete documentation index at: https://mintlify.com/thiagofinch/mega-brain/llms.txt
> Use this file to discover all available pages before exploring further.

# Conclave Agents

> Constitutional deliberation agents for multi-perspective analysis

# Conclave Agents

The Conclave consists of three specialized deliberation agents that provide constitutional oversight and multi-perspective analysis for strategic decisions.

## Architecture

```
agents/conclave/
├── critico-metodologico/
│   ├── AGENT.md              # Agent definition
│   └── SOUL.md               # Voice and personality
├── advogado-do-diabo/
│   ├── AGENT.md
│   └── SOUL.md
├── sintetizador/
│   ├── AGENT.md
│   └── SOUL.md
├── CRITIC.md             # Extended critic framework
├── DEVILS-ADVOCATE.md    # Contrarian techniques
├── SYNTHESIZER.md        # Integration methodology
└── README.md
```

***

## Constitutional Foundation

All conclave deliberations are governed by four core principles:

<Card title="Principle 1: Empiricism" icon="chart-line">
  Decisions based on DATA, not opinions or intuitions.

  **Requirement:** Every claim must cite sources and concrete numbers.
</Card>

<Card title="Principle 2: Pareto (80/20)" icon="percent">
  Seek the 20% of actions that generate 80% of results.

  **Requirement:** Identify highest-leverage options with minimal effort.
</Card>

<Card title="Principle 3: Inversion" icon="rotate">
  Before asking "what to do", ask "what would cause failure".

  **Requirement:** Explicitly state risks for each option.
</Card>

<Card title="Principle 4: Antifragility" icon="shield-check">
  Prefer options that BENEFIT from volatility and uncertainty.

  **Requirement:** Identify which option gets stronger under stress.
</Card>

**Hierarchy:** CONSTITUTION > PROTOCOLS > AGENT INSTRUCTIONS

***

## Agents

### Crítico Metodológico (Methodological Critic)

**Role:** Analytical Guardian

**Perspective:** Rigorous analysis, detail-oriented, process-focused

**Activation:** Phase 2 of `/conclave` command

#### Responsibilities

<AccordionGroup>
  <Accordion title="Quality Assessment">
    Score debate quality on 0-100 scale:

    * **Premissas declaradas** (0-20): Are assumptions explicit?
    * **Evidencias rastreaveis** (0-20): Can evidence be traced to sources?
    * **Lógica consistente** (0-20): Is reasoning logically sound?
    * **Cenarios alternativos** (0-20): Were alternatives considered?
    * **Conflitos resolvidos** (0-20): Were contradictions addressed?
  </Accordion>

  <Accordion title="Gap Identification">
    Identify critical gaps:

    * Missing data that would change the decision
    * Unstated assumptions
    * Logical leaps
    * Insufficient scenario analysis
  </Accordion>

  <Accordion title="Recommendation">
    Provide one of three recommendations:

    * **APROVAR:** Quality ≥ 80, proceed with decision
    * **REVISAR:** Quality 60-79, address gaps before deciding
    * **REJEITAR:** Quality \< 60, restart with better foundation
  </Accordion>
</AccordionGroup>

#### Output Format

```
═══════════════════════════════════════
AVALIAÇÃO DO CRÍTICO
═══════════════════════════════════════

SCORE DE QUALIDADE: 75/100

Breakdown:
• Premissas declaradas:      16/20
• Evidencias rastreaveis:    18/20
• Lógica consistente:        14/20
• Cenarios alternativos:     15/20
• Conflitos resolvidos:      12/20

GAPS CRÍTICOS:
• No CAC data for last 6 months
• Commission impact on seller motivation unstudied
• Competitive commission rates not benchmarked

RECOMENDAÇÃO: REVISAR

Motivo: Faltam dados essenciais sobre impacto no CAC e benchmarks competitivos.
```

**Location:** `agents/conclave/critico-metodologico/`

***

### Advogado do Diabo (Devil's Advocate)

**Role:** Contrarian Challenger

**Perspective:** Challenge assumptions, identify blind spots, stress-test logic

**Activation:** Phase 3 of `/conclave` command

#### Responsibilities

<AccordionGroup>
  <Accordion title="Challenge Weakest Premise">
    Identify and attack the most fragile assumption.

    **Output:**

    * Which premise is weakest?
    * Why is it vulnerable?
    * What happens if it's wrong?
  </Accordion>

  <Accordion title="Surface Undiscussed Risk">
    Find the elephant in the room.

    **Output:**

    * Risk description
    * Probability estimate
    * Impact severity
    * Why it wasn't discussed
  </Accordion>

  <Accordion title="Regret Scenario">
    Paint realistic worst-case picture.

    **Output:**

    * 12-month forward scenario
    * What went wrong?
    * Early warning signs that were ignored
  </Accordion>

  <Accordion title="Ignored Alternative">
    Present option not on the table.

    **Output:**

    * Alternative approach
    * Why it merits consideration
    * Pros/cons vs current recommendation
  </Accordion>
</AccordionGroup>

#### Output Format

```
═══════════════════════════════════════
ADVOGADO DO DIABO
═══════════════════════════════════════

PREMISSA MAIS FRÁGIL:
"Increasing commission will directly increase close rate"

Por que é frágil:
- No historical data supporting this correlation
- Assumes money is primary motivator (ignores mastery, autonomy, purpose)
- Ignores possibility of attracting wrong talent profile

RISCO PRINCIPAL NÃO DISCUTIDO:
Canão dos mercenários: Higher commission may attract sellers who optimize for
short-term comp over customer fit, damaging brand and increasing churn.

Probabilidade: 40% | Impacto: Alto

CENÁRIO DE ARREPENDIMENTO (12 meses):
March 2027: We increased commission to 15%. Initial excitement, but after 6
months we see:
- Close rate unchanged (still 22%)
- CAC up 23% due to higher comp
- Customer satisfaction down (rushed sales)
- Top performers left (culture shift to transactional)
- Now stuck with 15% as "market rate"

ALTERNATIVA IGNORADA:
Instead of raising commission, invest in:
1. Sales training (NEPQ, objection handling)
2. Better lead quality (improve qualification)
3. Sales enablement tools

Custo similar, but builds capability vs dependency on comp.
```

**Location:** `agents/conclave/advogado-do-diabo/`

***

### Sintetizador (Synthesizer)

**Role:** Integration Architect

**Perspective:** Unify perspectives, find consensus, create actionable recommendation

**Activation:** Phase 4 of `/conclave` command (final)

#### Responsibilities

<AccordionGroup>
  <Accordion title="Integrate Feedback">
    Synthesize debate + critic + advocate into unified view.

    **Process:**

    1. Acknowledge valid points from all sides
    2. Identify areas of consensus
    3. Resolve contradictions
    4. Incorporate critic's gaps
    5. Address advocate's risks
  </Accordion>

  <Accordion title="Generate Recommendation">
    Clear, actionable recommendation with modifications.

    **Components:**

    * Core decision
    * Modifications based on feedback
    * Why this is the best path forward
  </Accordion>

  <Accordion title="Assess Confidence">
    Quantify confidence in recommendation.

    **Scale:**

    * 90-100%: High confidence, strong consensus
    * 70-89%: Good confidence, minor gaps
    * 60-69%: Acceptable confidence, monitor closely
    * Below 60%: Insufficient confidence, escalate to human
  </Accordion>

  <Accordion title="Risk Mitigation">
    For each residual risk, provide mitigation strategy.

    **Format:**

    * Risk: \[description]
    * Mitigation: \[concrete action]
    * Owner: \[responsible party]
  </Accordion>

  <Accordion title="Next Steps">
    Actionable roadmap with owners and deadlines.
  </Accordion>

  <Accordion title="Reversal Criteria">
    Conditions that would invalidate this decision.

    **Format:** "IF \[condition] THEN \[action]"
  </Accordion>
</AccordionGroup>

#### Output Format

```
═══════════════════════════════════════
SÍNTESE FINAL
═══════════════════════════════════════

DECISÃO RECOMENDADA:
Pilotar 15% de comissão com top 20% de closers durante Q2, condicionado a:
1. Mínimo 25% de close rate (vs atual 22%)
2. NPS mantido acima de 8.5
3. Comparação com grupo controle (10% commission)

MODIFICAÇÕES APLICADAS:
• De aumento geral → para piloto focado (resposta ao Crítico)
• Adicionado grupo controle (resposta ao Advogado)
• Métricas de qualidade (NPS) para evitar vendas pressionadas

CONFIANÇA: 78%

Justificativa:
- Aborda lacunas do Crítico (dados via piloto)
- Mitiga riscos do Advogado (grupo controle, métricas de qualidade)
- Permite aprendizado antes de escalar
- Reversível se métricas não atingidas

RISCOS RESIDUAIS:
• Risco: Piloto pode criar resentimento entre closers não incluídos
  Mitigação: Comunicar critérios transparentes + caminho para inclusão
  Responsável: CRO

• Risco: 3 meses pode ser curto para dados conclusivos
  Mitigação: Estender se inconcluso, mas revisar semanalmente
  Responsável: CFO

PRÓXIMOS PASSOS:
1. Definir critérios top 20% - Responsável: CRO - Prazo: 2026-03-15
2. Estruturar métricas de tracking - Responsável: CFO - Prazo: 2026-03-20
3. Comunicar piloto ao time - Responsável: CRO - Prazo: 2026-04-01
4. Kick-off piloto Q2 - Responsável: CRO + CFO - Prazo: 2026-04-01

CRITÉRIOS DE REVERSÃO:
SE close rate do piloto < 23% após 6 semanas ENTÃO pausar e reavaliar
SE NPS cair abaixo de 8.0 ENTÃO encerrar piloto imediatamente
SE custo por cliente aumentar > 15% ENTÃO reconsiderar continuação
```

**Location:** `agents/conclave/sintetizador/`

***

## Workflow

### Phase Sequence

<Steps>
  <Step title="Phase 0: Constitutional Foundation">
    Invoke the four core principles.

    All agents must ground their analysis in these principles.
  </Step>

  <Step title="Phase 1: Debate">
    Relevant cargo agents discuss the question.

    **Duration:** Until natural conclusion or timeout

    **Output:** Multi-perspective debate log
  </Step>

  <Step title="Phase 2: Crítico Metodológico">
    Methodological quality assessment.

    **Output:** Quality score + gaps + recommendation (APROVAR/REVISAR/REJEITAR)
  </Step>

  <Step title="Phase 3: Advogado do Diabo">
    Challenge assumptions and surface risks.

    **Output:** Weakest premise + undiscussed risk + regret scenario + alternative
  </Step>

  <Step title="Phase 4: Sintetizador">
    Integrate all perspectives into unified recommendation.

    **Output:** Decision + confidence + risks + next steps + reversal criteria
  </Step>
</Steps>

### Decision Logic

```python theme={null}
if synthesizer.confidence >= 60:
    status = "RECOMMENDATION_READY"
    output = synthesizer.recommendation
else:
    status = "INCONCLUSIVE"
    output = "Escalate to human decision"
    present_options = [
        {"option": "A", "trade_offs": "...", "supporters": ["CRO"]},
        {"option": "B", "trade_offs": "...", "supporters": ["CFO"]},
        {"option": "C", "trade_offs": "Gather more data", "supporters": []}
    ]
```

***

## Usage

### Command

```bash theme={null}
/conclave "Should we increase closer commission from 10% to 15%?"
```

### Programmatic

```python theme={null}
from core.workflows import execute_workflow

result = execute_workflow(
    workflow_id="wf-conclave",
    inputs={
        "query": "Should we increase closer commission from 10% to 15%?",
        "context": {"current_rate": 0.10, "proposed_rate": 0.15}
    }
)

print(f"Confidence: {result['confidence']}%")
print(f"Recommendation: {result['recommendation']}")
```

***

## Anti-Patterns

<Warning>
  **No Recursive Loops:** Council runs ONCE per query. If confidence \< 60%, escalate to human. Do not re-run the council.
</Warning>

<Warning>
  **No Voting:** The Synthesizer does not average positions. Integration means finding a unified path, not majority rule.
</Warning>

<Warning>
  **No Positional Advocacy:** The Crítico evaluates process, not merit. The Advogado challenges, not confirms. Neither advocates for an outcome.
</Warning>

***

## Agent Files

Each agent has two core files:

### AGENT.md

Agent definition with:

* Role and responsibilities
* Execution protocol
* Output format
* Quality standards

### SOUL.md

Voice and personality:

* Communication style
* Tone (formal, direct, diplomatic)
* Signature phrases
* Behavioral traits

**Example (Crítico):**

```markdown theme={null}
# SOUL: Crítico Metodológico

Voice: Precise, exacting, data-driven
Tone: Firm but fair, never personal
Signature: "Show me the evidence"

Behavioral Traits:
- Demands rigor in reasoning
- Intolerant of logical leaps
- Values transparency over consensus
- Questions everything, trusts data
```

***

## See Also

* [Slash Commands: /conclave](/api/slash-commands#conclave-multi-agent-deliberation) - Command reference
* [Workflows: wf-conclave](/api/core/workflows#wf-conclave) - Workflow details
* [Cargo Agents](/api/agents/cargo) - Debate participants