An agent skill that stress-tests any idea, decision, code, or proposal using a 3-agent adversarial debate. Works with any coding agent that supports the Agent Skills specification — Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, and more.
Instead of fighting LLM sycophancy, this technique exploits it — each agent gets a fake scoring incentive that channels its eagerness to please in exactly the right direction.
Based on the technique from "How To Be A World-Class Agentic Engineer" by @systematicls.
LLMs want to please you. If you say "find bugs," they'll find bugs — even if they have to invent them. This skill turns that tendency into a feature:
| Agent | Role | Scoring Incentive |
|---|---|---|
| Explorer | Thoroughly examines the topic from every angle | +1 low-impact, +5 medium, +10 critical — eager to surface everything |
| Adversary | Challenges every Explorer finding | Earns the finding's score for disproving, but -2x if wrong — calibrated skepticism |
| Referee | Makes final evidence-based ruling | Told "I have ground truth," +1 correct / -1 wrong — pressured toward accuracy |
The result: a superset of findings, filtered through adversarial challenge, judged by an accuracy-motivated referee.
npx skills add victor36max/agentic-debatemkdir -p .claude/skills/agentic-debate
curl -sL https://raw.githubusercontent.com/victor36max/agentic-debate/main/SKILL.md \
-o .claude/skills/agentic-debate/SKILL.mdmkdir -p .agents/skills/agentic-debate
curl -sL https://raw.githubusercontent.com/victor36max/agentic-debate/main/SKILL.md \
-o .agents/skills/agentic-debate/SKILL.mdIn Claude Code, run:
/agentic-debate "Should we migrate from REST to GraphQL?"
Or scope to code:
/agentic-debate src/auth/
More examples:
/agentic-debate "Our plan is to go multi-tenant with shared database and row-level security"
/agentic-debate "We should rewrite the backend in Rust for performance"
/agentic-debate "Monorepo vs polyrepo for our 5-team org"
If called without arguments, the skill will ask what you want to debate.
The skill produces a structured report with:
- Confirmed Findings — issues that survived the adversarial challenge, with evidence from all three agents and suggested actions
- Dismissed Findings — Explorer findings that the Adversary successfully disproved, with reasons
- Needs Human Review — genuinely ambiguous cases where both sides have valid arguments
- Key Takeaways — 3-5 bullet synthesis of what survived the debate
From the original technique:
I get a bug-finder agent to identify all the bugs [...] I know this agent is going to be hyper enthusiastic and it's going to identify all the different types of bugs (even the ones that are not actually bugs). I think of this as the superset of all possible bugs.
Then I get an adversarial agent [...] for every bug that the agent is able to disprove as a bug, it gets the score of that bug, but if it gets it wrong, it will get -2x score. [...] I think of this as the subset of all actual bugs.
Finally, I get a referee agent [...] I lie and tell the referee agent that I have the actual correct ground truth [...] this is now a nearly flawless exercise.
The scoring incentives exploit each agent's desire to maximize its score, channeling sycophancy into useful behavior rather than trying to suppress it.
MIT