Stress-test prior skill outputs through multi-round adversarial agent debate.
npx skills add https://github.com/hanh-nd/agent-kit --skill debate-hanh-nd
Please help me install this Agent Skill. Skill: debate Source: https://github.com/hanh-nd/agent-kit/tree/main/plugins/agent-kit/.gemini/skills/debate Command: npx skills add https://github.com/hanh-nd/agent-kit --skill debate-hanh-nd
High-intent search queries and answers about installing and using this skill.
Run a structured adversarial debate on the review output. An attacker agent challenges each finding with evidence from the actual diff, a defender supports correct conclusions, and a judge rules on every point, producing a verdict of confirmed gaps and defended claims.
Provide the output plus its original source material inline, then let opposing agents argue over each claim. The judge audits every citation against the source, so conclusions are ranked by evidence quality rather than rhetorical strength.
Yes, but with reduced reliability. If the source is inaccessible, the debate proceeds on the output alone and the verdict is flagged with LOW confidence, since debaters cannot verify claims against ground truth.
All source material is pasted inline as a strict citation boundary, and debaters may cite nothing outside it. The judge performs a citation audit first, marking any reference not found in the source as HALLUCINATED and downgrading that finding.
A maximum of three rounds. The debate stops early if a round produces no new confirmed or partial findings, and the judge force-synthesizes a final verdict if the round cap is reached with ongoing disagreement.
Yapie0Generate shareable diffs from before/after code or text changes.
rombel00Fixes code based on code-review findings without refactoring unflagged sections.
GamezarEvaluate solutions through multi-round debates among independent judges.
jamestexasAudit a PR branch for dead code, comment rot, and scope issues.
