fux-debate

Orchestrates a two-agent adversarial debate to author and ratify constitutional rules.

1|Updated Jun 1, 2026
One-click install
npx skills add https://github.com/arpitarya/fux --skill fux-debate-arpitarya
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: fux-debate
Source: https://github.com/arpitarya/fux/tree/main/archive/v0.1/fux/data/skills/debate
Command: npx skills add https://github.com/arpitarya/fux --skill fux-debate-arpitarya

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Rules added to a project's constitution on a single agent's say-so are a single point of failure; this Skill forces a genuine two-agent debate with blind first passes and anti-sycophancy gates before any rule is ratified by a human. ## Core Features & Use Cases - Blind two-agent debate: Spawns two sub-agents with identical briefs and no assigned sides, each required to raise at least one concrete objection to the proposed rule. - Anti-sycophancy gates: Convergence only counts after both agents genuinely attempt to falsify the rule; instant agreement on a constitutional-tier rule forces an extra adversarial round. - Human ratification with audit trail: Writes the full transcript to .fux/debates/<rule-id>.md and hands the user the fux ratify command, which hashes the transcript into ratification.debate_hash and opens a PR on protected branches. - Decision capture: Routes the concluded debate as a tamper-evident ADR via fux capture-decision with a money/PII firewall for sensitive routes. - Use Case: A team wants to add the rule "never commit money docs or PII" as a constitutional invariant; the host agent runs the debate, captures dissent, and the human ratifies it with an auditable hashed transcript. ## Quick Start Run /fux debate "never commit money docs or PII; plans live in elgar" --tier constitutional and follow the debate procedure to ratify the rule.

Frequently Asked Questions about fux-debate

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I author a constitutional rule with a two-agent debate?▼

Run /fux debate with your proposed rule and --tier constitutional. The host agent spawns two sub-agents with identical briefs and no assigned sides, runs blind first passes plus adversarial rounds, then hands you the fux ratify command to complete ratification yourself.

What is the difference between standard and constitutional tier rules?▼

A rule is constitutional only if a wrong answer costs money, PII, audit, or trust and the rule never legitimately changes. Constitutional rules get an extra adversarial round on instant agreement and enter the apex via fux ratify; everything else is standard tier.

Does the debate skill make API calls or cost extra tokens?▼

The only intelligence spent is the host session's own tokens and the two sub-agents it spawns. Fux itself runs no model and makes no API call; it only provides the deterministic harness that captures, hashes, and ratifies the transcript.

What happens if the two debate agents do not converge?▼

The host escalates to the human with both final positions side by side and the unresolved crux. No rule is authored or ratified until the human acts as tie-breaker, after which the transcript including their ruling is written and ratified.

Why does the debate require each agent to raise an objection?▼

The anti-sycophancy gates prevent rubber-stamping: an agent that raises no concrete objection is sent back, and convergence only counts after both agents genuinely tried to falsify the rule. This surfaces failure modes a lone author would rationalize away.

Can I ratify a rule directly on a protected branch?▼

No. On a protected branch with a remote, fux ratify automatically writes on a new constitution/<rule-id> branch and opens a PR, which merges only after required fux gate and ai-review checks pass. The --no-pr flag does a local offline ratify instead.