agon

Instantiate Advocate and Skeptic positions to test claims and generate Convergence Reports.

4|2|Updated Dec 31, 2025

Search Tags:#reasoning #argumentation #debate #epistemology #adversarial AI #claim evaluation

Ready to run

npx skills add https://github.com/TylerGarlick/abraxas --skill agon

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: agon
Source: https://github.com/TylerGarlick/abraxas/tree/main/skills/agon
Command: npx skills add https://github.com/TylerGarlick/abraxas --skill agon

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill addresses the inherent bias in AI models towards single-voice reasoning by instantiating opposing viewpoints to rigorously test claims and reveal genuine points of agreement and divergence.

Core Features & Use Cases

Adversarial Debate: Pit an Advocate and a Skeptic against a claim to uncover its strengths and weaknesses.
Steelman & Falsify: Generate the strongest possible version of a weak claim or identify precise conditions under which a claim would be false.
Convergence Reporting: Produce detailed reports highlighting areas of consensus and contention between opposing arguments.
Use Case: When evaluating a complex hypothesis, use /agon debate to see if an Advocate can defend it and a Skeptic can find flaws, then review the Convergence Report to understand where the truth likely lies.

Quick Start

Use the agon skill to debate the claim that AI will achieve general intelligence by 2030.

Frequently Asked Questions about agon

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

FAQPage Schema

How do I test a factual claim using adversarial reasoning?▼

Steelman a weak claim by generating the strongest possible version of the hypothesis, while falsification identifies the precise conditions under which a strong claim would definitively fail.

What is structured adversarial argumentation and when do I need it?▼

Structured adversarial argumentation is a method of rigorously testing hypotheses by enforcing position asymmetry between opposing sides, needed when evaluating complex claims to overcome inherent AI single-voice reasoning bias.

How do I identify areas of agreement and divergence in a debate?▼

Generate a Convergence Report after an adversarial debate to highlight specific areas of consensus, contention, and unresolved questions between opposing arguments, revealing where the truth likely lies.

What is the best way to steelman a weak hypothesis or falsify a strong one?▼

Steelman a weak claim by generating the strongest possible version of the hypothesis, while falsification identifies the precise conditions under which a strong claim would definitively fail.

Can I use adversarial AI to evaluate epistemic rigor without external dependencies?▼

You can evaluate epistemic rigor without external dependencies by using enforced position asymmetry and Janus labeling to independently structure opposing viewpoints and test claims.