Verdict Synthesis Process and Output Templates in Council of High Intelligence

The verdict synthesis process is a dedicated Chairman step that aggregates multi-round deliberation transcripts into a structured final verdict using mode-specific output templates defined in SKILL.md.

The Council of High Intelligence repository implements a structured deliberation framework where multiple AI agents analyze problems through distinct rounds before a designated Chairman synthesizes the final output. This verdict synthesis process ensures that diverse perspectives are consolidated into actionable, auditable conclusions with consistent formatting across different operational modes.

How the Verdict Synthesis Process Works

The Chairman Step (STEP 7)

The synthesis occurs exclusively during STEP 7 (in Full mode) and is performed only by the Chairman selected during STEP 1.7. According to the execution sequence in SKILL.md, the coordinator does not participate in the synthesis logic—it merely forwards the fully constructed prompt to the Chairman’s model and returns the response unchanged to the user【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/SKILL.md#L70-L78】.

The Chairman is determined through one of three methods:

• Auto-routing based on provider availability
• Configuration override specified in configs/auto-route-defaults.yaml
• Explicit CLI flag using --chairman

Prompt Construction and Execution

The Chairman receives a single, fully-formed prompt containing:

• The original problem statement
• Complete transcripts from Round 1 (independent analyses), Round 2 (cross-examination), and Round 3 (final crystallisation) with real member names restored
• Explicit instructions to weigh arguments by validity, surface genuine disagreement, and catalog knowledge gaps

The prompt terminates with a placeholder that the system replaces with the concrete verdict template defined in the "Output Templates" section of SKILL.md【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/SKILL.md#L70-L78】. If the Chairman call fails, the coordinator executes a fallback synthesis annotated with Chairman: <name> (FAILED — synthesized by coordinator fallback)【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/SKILL.md#L107-L112】.

Output Templates for Each Operating Mode

The repository defines three distinct output templates corresponding to the three operating modes. Each template structures the final verdict into standardized sections while accommodating mode-specific constraints.

Full Mode: Council Verdict Template

Located in SKILL.md under "### Council Verdict (Full Mode)"【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/SKILL.md#L75-L108】, this template produces the most comprehensive output. The sections appear in the following order:

1. Problem – Restatement of the original query
2. Council Composition – List of participating members and their roles
3. Chairman – Identity of the synthesizing agent
4. Provider Routing – Model routing details for each participant
5. Acceptable Compromises – Middle-ground positions the council finds valid
6. Kill Criteria – Conditions that would invalidate the recommendation
7. Concrete Next Step – Specific, actionable recommendation
8. Unresolved Questions – Explicit acknowledgment of unknowns
9. Recommended Next Steps – Follow-up actions for the user
10. Consensus & Agreement – Summary of alignment and dissent

Quick Mode: Quick Verdict Template

Defined under "### QUICK STEP 3: Synthesize Quick Verdict (CHAIRMAN)" within the Quick Mode Sequence【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/SKILL.md#L31-L39】, this template accelerates the process by skipping Round 2 cross-examination. Key differences include:

• Quick-mode-specific composition listing only the participating members
• Single-provider default routing notation
• Condensed deliberation history reflecting only two rounds

Duo Mode: Duo Verdict Template

Found under "### DUO STEP 4: Synthesize Duo Verdict (CHAIRMAN)"【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/SKILL.md#L68-L71】, this template facilitates dialectic between two members. The structure mirrors the Full mode template but includes:

• Only two members in the composition section
• An explicit constraint noting that the Chairman must not be one of the duo participants

A canonical enumeration of all three templates appears in demos/verdict-template.md【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/demos/verdict-template.md】 for quick reference.

Execution Flow and Checkpoints

The verdict synthesis process represents the culmination of an eight-step pipeline:

1. Panel selection (STEP 0)
2. Provider routing (STEP 1)
3. Problem restatement (STEP 1.5)
4. Deliberation rounds (Rounds 1-3)
5. Chairman selection (STEP 1.7)
6. Synthesize verdict (STEP 7 for Full mode, or analogous steps for Quick/Duo)
7. Append session metadata (STEP 8)

Each checkpoint is announced to the user with [CHECKPOINT] markers before proceeding, ensuring full transparency and auditability of the synthesis chain.

Practical Examples

Full Mode Council

# Convene the full council on a pricing strategy question

/council --full "What is the right pricing strategy for our SaaS product?"

The coordinator executes three deliberation rounds, selects a Chairman (e.g., gemini via auto-routing), and invokes the synthesis prompt. The final output conforms to the Council Verdict template structure defined in SKILL.md.

Quick Mode Analysis

# Fast 2-round analysis with trimmed verdict

/council --quick "Should we add caching here?"

This command skips cross-examination, uses the Quick Verdict template, and still runs the Chairman synthesis (STEP 3 of Quick mode) to ensure quality control.

Duo Mode Dialectic

# Two-person dialectic on architectural decisions

/council --duo "Should we use microservices or a monolith?"

The two selected members exchange positions through Steps 1-3, then a non-participating Chairman synthesizes the Duo Verdict using the dedicated template.

Key Source Files and Implementation Details

File	Purpose
SKILL.md	Master specification containing execution sequences, step definitions, synthesis prompts, and all output templates (Full, Quick, Duo)【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/SKILL.md】
demos/verdict-template.md	Reference document enumerating the three verdict templates and pointing back to SKILL.md【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/demos/verdict-template.md】
scripts/detect-providers.sh	Generates provider detection JSON used for Chairman routing and selection
configs/auto-route-defaults.yaml	Default provider-model mappings and Chairman defaults
agents/	Directory containing individual member definitions (e.g., council-aristotle.md) used to reconstruct the deliberation transcript

Summary

• The verdict synthesis process is exclusively performed by the Chairman agent selected in STEP 1.7, not the coordinator.
• The Chairman receives a composite prompt containing the original problem and complete deliberation transcripts from all three rounds.
• Three distinct output templates exist: Council Verdict (Full), Quick Verdict (Quick), and Duo Verdict (Duo), each defined in SKILL.md.
• If the Chairman fails, the coordinator falls back to direct synthesis with explicit failure annotation.
• All checkpoints are announced via [CHECKPOINT] markers to maintain auditability.

Frequently Asked Questions

Who performs the verdict synthesis in the Council of High Intelligence?

The Chairman selected during STEP 1.7 performs the synthesis exclusively. The coordinator merely forwards the prompt and returns the response; it does not modify or participate in the synthesis unless the Chairman call fails, in which case a fallback synthesis occurs with explicit failure notation【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/SKILL.md#L107-L112】.

What happens if the Chairman fails to generate a verdict?

If the Chairman's model call fails, the coordinator executes a fallback synthesis directly and annotates the output with Chairman: <name> (FAILED — synthesized by coordinator fallback). This ensures the user always receives a verdict even when the primary synthesis mechanism encounters errors【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/SKILL.md#L107-L112】.

How do the output templates differ between Full, Quick, and Duo modes?

All templates share core sections (Problem, Chairman, Acceptable Compromises, Kill Criteria, Concrete Next Step), but differ in scope. Full mode includes complete Round 1-3 transcripts and full council composition. Quick mode omits Round 2 cross-examination and uses single-provider routing. Duo mode restricts composition to two members and requires the Chairman to be external to the dialectic pair【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/SKILL.md#L31-L39】【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/SKILL.md#L68-L71】【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/SKILL.md#L75-L108】.

Where are the verdict templates defined in the codebase?

The canonical definitions reside in SKILL.md under their respective mode sections: "### Council Verdict (Full Mode)" for Full mode, "### QUICK STEP 3: Synthesize Quick Verdict (CHAIRMAN)" for Quick mode, and "### DUO STEP 4: Synthesize Duo Verdict (CHAIRMAN)" for Duo mode【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/SKILL.md#L75-L108】. A summary reference exists in demos/verdict-template.md【/cache/repos/github.com/0xNyk/council-of-high-intelligence/main/demos/verdict-template.md】.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}