judgment-day

Run blind dual adversarial code reviews with re-judging until approval or escalation.

2|Updated Jul 29, 2025

Search Tags:#code quality #dual review #adversarial judging #issue confirmation #re-judgment loop #confirmed-only fixes #PR triage

Ready to run

npx skills add https://github.com/kurojs/EnderDots --skill judgment-day-kurojs

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: judgment-day
Source: https://github.com/kurojs/EnderDots/tree/main/.config/opencode/skills/judgment-day
Command: npx skills add https://github.com/kurojs/EnderDots --skill judgment-day-kurojs

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

It reduces false positives and missed bugs by forcing two independent, blind adversarial judges to review the same target using the same injected standards, then only fixing confirmed findings.

Core Features & Use Cases

Blind dual/adversarial review: Launches two parallel judges that never self-review the code and return severity-tagged findings.
Deterministic confirmation workflow: Separates confirmed, suspect, contradiction, and informational items, with explicit escalation rules when judges disagree.
Confirmed-only repair loop: Optionally asks before fixing Round 1 confirmed issues, then re-runs both judges after each fix iteration to re-validate.

Quick Start

Use the judgment-day skill to dual-review and re-judge a specific repository feature or pull request when you ask for “Judgment Day” or “juzgar” with the target clearly specified.

Frequently Asked Questions about judgment-day

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

FAQPage Schema

How do I reduce false positives in adversarial code review?▼

Adversarial code review reduces false positives by running blind dual judges that independently evaluate the same target. Only confirmed findings from both judges proceed to fixes, ensuring theoretical warnings are separated from real issues.

How does blind dual review work for pull request triage?▼

Blind dual review works by launching two parallel judges that never self-review the code and return severity-tagged findings. The confirmation workflow separates confirmed, suspect, contradiction, and informational items with explicit escalation rules.

What is a confirmed-only fix loop in code quality assessment?▼

A confirmed-only fix loop gates repairs so that only Round 1 confirmed issues are fixed. After each fix iteration, both judges re-run to re-validate the changes, ensuring strict terminal outcomes of APPROVED or ESCALATED.

Can I use dual adversarial judging for architecture slices and error handling?▼

Yes, dual adversarial judging applies to reviewing files, features, PRs, or architecture slices. It assesses correctness, edge cases, error handling, performance, and security against injected project standards using parallel judge execution.

What happens when two independent code review judges disagree?▼

When two independent judges disagree, the workflow applies explicit escalation rules. Contradictions between judges trigger strict terminal outcomes, resulting in an ESCALATED status rather than proceeding with unconfirmed fixes.

How to normalize severity tags in PR triage with dual reviewers?▼

PR triage with dual reviewers normalizes severity by distinguishing real issues from theoretical warnings. The workflow categorizes findings into confirmed, suspect, contradiction, and informational items before applying confirmed-only fix gating.