How Word Count Limits Differ Across Deliberation Rounds in Council of High Intelligence

The Council of High Intelligence framework enforces strict word-count caps that vary by deliberation mode and round, ranging from 400 words for initial analysis in Full mode down to 50 words for final statements in Duo mode.

The Council of High Intelligence is an open-source deliberation framework that structures multi-agent reasoning through controlled conversation flows. According to the source code in SKILL.md, word count limits differ across deliberation rounds to optimize token budgets, maintain focus, and produce succinct, auditable verdicts throughout the decision-making process.

Word Count Limits by Deliberation Mode

The framework defines three distinct deliberation modes—Full, Quick, and Duo—each with specific round-based constraints documented in SKILL.md.

Full Mode: Deep Multi-Round Deliberation

Full mode implements a three-round progressive constraint system designed for thorough analysis.

• Round 1 (Independent Analysis): Responses are capped at 400 words (SKILL.md, providing ample space for comprehensive first-pass essays and blind-first analysis.
• Round 2 (Cross-Examination): The limit tightens to 300 words (SKILL.md, with the additional requirement that participants engage with at least two peers, forcing focused critique.
• Round 3 (Final Crystallization): A stringent 100-word cap (SKILL.md compels each member to distill their position to the essential conclusion.

Quick Mode: Rapid Binary Assessment

Quick mode compresses deliberation into two rounds with uniform limits to prioritize speed over depth.

• Round 1 (Rapid Analysis): Embedded single-round analysis is limited to 200 words (SKILL.md.
• Round 2 (Final Positions): Anonymous decisive statements are capped at 200 words (SKILL.md, maintaining brevity for rapid verdicts.

This mode explicitly skips the cross-examination step found in Full mode.

Duo Mode: Two-Member Dialectic

Duo mode orchestrates a structured debate between two council members with escalating brevity requirements.

• Round 1 (Opening Positions): Each member receives 300 words (SKILL.md for full opening statements.
• Round 2 (Direct Response): The exchange of arguments permits 200 words (SKILL.md with no anonymization.
• Round 3 (Final Statements): Closing statements are strictly limited to 50 words (SKILL.md, ensuring a crisp final stance.

Source Code Implementation

These constraints are hardcoded in the execution sequence definitions within SKILL.md. The word count limits are enforced through prompt engineering in the agent definitions located in agents/*.md, which work in conjunction with the execution logic to ensure compliance. User-facing documentation in README.md provides invocation examples that trigger these specific limit profiles.

Practical Usage Examples

The following commands demonstrate how each mode applies its specific word count constraints:

# Full mode – triad example (architectural triad)

/council --triad architecture "Should we adopt a monorepo?"

# → Round 1 responses are capped at 400 words each.

# → Round 2 cross-examination is capped at 300 words each.

# → Round 3 final crystallization is capped at 100 words each.

# Quick mode – rapid check

/council --quick "Do we need caching in the auth flow?"

# → The single analysis round is limited to 200 words.

# → The final position round is also limited to 200 words.

# Duo mode – dialectic between two members

/council --duo "Should we rewrite the monolith as micro-services?"

# → Opening positions: 300 words each.

# → Direct response: 200 words each.

# → Final statements: 50 words each.

Summary

• Full mode uses a descending progression of 400 → 300 → 100 words across three rounds to force increasingly focused analysis.
• Quick mode maintains a consistent 200-word limit across both rounds to enable rapid binary assessments.
• Duo mode scales from 300 words down to 50 words to balance depth with brevity in two-member debates.
• All limits are explicitly defined in SKILL.md at specific line numbers and enforced through the agent prompt architecture.

Frequently Asked Questions

What is the highest word count limit in the Council of High Intelligence framework?

The maximum word count allowance is 400 words, which occurs in Round 1 of Full mode (Independent Analysis) as defined at line 347 of SKILL.md. This generous initial allocation accommodates comprehensive independent analysis before subsequent rounds force increasing concision.

Why do word count limits decrease in later rounds of Full mode?

The descending progression from 400 to 300 to 100 words serves a specific pedagogical purpose: Round 1 enables thorough blind-first essays, Round 2 forces focused peer critique by tightening constraints, and Round 3 requires final crystallization that distills complex arguments into essential conclusions. This structure ensures each deliberation phase remains focused and produces actionable verdicts.

How do Quick mode and Duo mode word limits compare?

Quick mode maintains uniform 200-word limits across both rounds (lines 564 and 659), prioritizing speed and binary decision-making. Duo mode employs a variable descending scale from 300 words down to 50 words (lines 636, 659, and 665), allowing more initial depth for opening positions while demanding extreme brevity for final statements in the two-member dialectic.

Where are the word count limits defined in the source code?

All word count constraints are centrally defined in SKILL.md at specific line numbers corresponding to each mode and round. These limits are implemented through the agent prompt templates located in agents/*.md and triggered by command-line invocations documented in README.md.

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