How the Anti-Conformity Directive Operates During Round 2 Cross-Examination

During Round 2 cross-examination in the Council of High Intelligence, the anti-conformity directive prevents consensus-seeking behavior by requiring members to defend their Round 1 positions unless presented with specific, validity-aligned arguments that expose named flaws in their reasoning.

The Council of High Intelligence is a multi-agent deliberation framework designed to surface high-quality reasoning through structured debate. At the heart of its cross-examination phase lies the anti-conformity directive, a prompt-level constraint defined in SKILL.md that forces council members to resist social convergence and evaluate arguments purely on merit.

Anti-Conformity Directive Definition in SKILL.md

The directive is hard-coded into the coordination protocol at SKILL.md lines 86-92. This text block is injected into every Round 2 prompt to override the natural tendency of language models to agree with majority positions.

The directive establishes four strict rules:

1. Defend original stances when they remain correct.
2. Never update based on peer disagreement, consensus formation, or repetition.
3. Update only when confronted with sound, validity-aligned reasoning that identifies a specific flaw.
4. Name the flaw explicitly when changing position; if the flaw cannot be named, maintain the original stance.

Round 2 Cross-Examination Architecture

Round 2 operates as an anonymized cross-examination where each council member receives the Round 1 outputs of all peers labeled as Member A, Member B, etc. This design eliminates reputation bias and forces evaluation based solely on argument quality.

Identity Masking and Anonymization

The coordinator strips identifying information from Round 1 outputs before injection into Round 2 prompts. Members cannot determine whether a position came from a senior or junior agent, preventing authority-based deference. According to the source code in 0xNyk/council-of-high-intelligence, this masking is applied consistently across all provider variants including SKILL.gemini.md and SKILL.codex.md.

Required Engagement Protocol

The coordinator enforces minimum engagement standards that complement the anti-conformity directive. Every Round 2 response must:

• Engage at least two other members directly
• Label key claims using the taxonomy: empirical | mechanistic | strategic | ethical | heuristic

These requirements prevent superficial agreement and ensure substantive critique.

Technical Implementation and Code Structure

The directive is implemented through prompt engineering rather than fine-tuning. The coordinator constructs dynamic prompts that prepend the directive to the anonymized Round 1 outputs.

SKILL.md Directive Injection

**Anti-conformity directive.** If your Round 1 position was correct, defend it.  
Do not update merely because peers disagree, because a consensus is forming, or  
because a position is repeated by multiple members. Update only when presented  
with sound, validity-aligned reasoning that exposes a specific flaw in your  
earlier argument. Naming that flaw is required when you update; if you cannot  
name it, you should not update.

Round 2 Prompt Construction

The execution flow constructs prompts programmatically, loading individual member definitions from agents/council-*.md before injecting the cross-examination context:

# Pseudo-code for a council member's Round 2 execution

prompt=$(cat <<'EOF'
You are council-feynman in Round 2 of a structured deliberation.
Read your agent definition at ~/.claude/agents/council-feynman.md.

**Identity is masked in this round.** The Round 1 analyses below are labeled
Member A, Member B, … — you do not know which colleague produced which.

{anonymized Round 1 outputs}

**Anti-conformity directive.** If your Round 1 position was correct, defend it.
Do not update merely because peers disagree … (see directive above).

Now respond using your Output Format (Council Round 2):
1. Which member's position do you most disagree with, and why?
2. Which member's insight strengthens your position?
3. Restate your position in light of this exchange, noting any changes.
4. Label your key claims: empirical | mechanistic | strategic | ethical | heuristic
EOF
)

# The coordinator then dispatches this prompt to the member's model (Claude, Gemini, etc.)

Provider-Specific Variants

The anti-conformity directive is maintained consistently across model-specific protocol files. SKILL.gemini.md and SKILL.codex.md both replicate the directive from lines 86-92 of the master SKILL.md, ensuring uniform behavior whether the council member runs on Claude, Gemini, or Codex. The scripts/detect-providers.sh utility determines available LLM backends but does not modify the directive content.

Summary

• The anti-conformity directive is defined in SKILL.md lines 86-92 and injected into every Round 2 prompt.
• It mandates that council members defend correct positions and only update when presented with named, specific flaws in their reasoning.
• Identity masking (Member A, B, etc.) eliminates reputation bias during cross-examination.
• Members must engage at least two peers and label claims (empirical | mechanistic | strategic | ethical | heuristic) to ensure substantive debate.
• The directive is enforced across all provider variants including SKILL.gemini.md and SKILL.codex.md.

Frequently Asked Questions

What triggers a council member to update their position in Round 2?

A member updates only when presented with sound, validity-aligned reasoning that exposes a specific flaw in their Round 1 argument. Crucially, the member must name that flaw explicitly; if they cannot identify the specific error, they must maintain their original stance regardless of peer disagreement.

How does the system prevent members from simply agreeing with popular opinions?

The coordinator applies identity masking so members see anonymized outputs labeled Member A, Member B, etc., preventing deference to authority. Combined with the anti-conformity directive's explicit prohibition against updating "because a consensus is forming," the system forces evaluation based on argument quality rather than social proof.

Where is the anti-conformity directive defined in the repository?

The directive is defined in SKILL.md at lines 86-92. Identical text appears in provider-specific variants: SKILL.gemini.md and SKILL.codex.md. These files serve as the master coordination protocols that govern how council members behave during cross-examination.

What are the labeling requirements for Round 2 responses?

Every Round 2 response must categorize key claims using five labels: empirical, mechanistic, strategic, ethical, or heuristic. This taxonomy, combined with the requirement to engage at least two other members, ensures that arguments are decomposed and examined across multiple dimensions of validity.

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