the-judge

Reviews pull requests and posts one consolidated evidence-backed GitHub review via the gh CLI.

Updated Sep 15, 2026
One-click install
npx skills add https://github.com/Peterson-Benhame/agent-skills --skill the-judge-peterson-benhame
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: the-judge
Source: https://github.com/Peterson-Benhame/agent-skills/tree/main/packages/skills-catalog/skills/%28quality%29/the-judge
Command: npx skills add https://github.com/Peterson-Benhame/agent-skills --skill the-judge-peterson-benhame

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Code reviews often drown authors in low-value comments, unverified claims about libraries, and endless re-review rounds. This Skill reviews a pull request like a senior engineer: every finding carries verified evidence, external claims are checked against current official documentation, and everything posts as a single consolidated GitHub review. ## Core Features & Use Cases - Evidence-first review: Internal claims require verified file:line citations; external claims require URLs from official docs fetched during the review, never memory. - Deterministic checks first: Runs the repo's own linters, type checkers, and tests, plus a regex-based bypass scan, so no tokens are wasted on what tooling already catches. - Noise gate and verdict mapping: A deterministic gate script enforces comment quality, caps nits at 5, and maps findings to APPROVE, COMMENT, or REQUEST_CHANGES. - Convergence contract: Round 1 is exhaustive; later rounds only verify resolution and new blockers, ending review ping-pong. - Use Case: Ask "judge this PR before I merge" and receive one GitHub review with inline comments anchored to diff lines, a research log, and a severity-weighed verdict. ## Quick Start Ask the agent to review the pull request for the current branch, for example by saying "judge this PR before I merge".

Frequently Asked Questions about the-judge

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

FAQPage Schema
How do I review a GitHub pull request with an AI agent?

Invoke the skill with a phrase like "judge this PR" or "revise esse PR" on a branch with an open PR. It fetches the diff via the gh CLI, runs the repo's checks, researches external dependencies, and posts one consolidated review with inline comments.

How does the review verdict get decided?

Findings are classified as blocker, should-fix, nit, or pre-existing. Any blocker produces REQUEST_CHANGES, zero blockers with should-fix items produces COMMENT, and otherwise APPROVE. A gate script verifies the verdict matches the findings before posting.

Does it work with repositories in any programming language?

Yes, the review passes are language-agnostic and the bypass scanner covers markers from JavaScript, TypeScript, Python, Go, Ruby, and Java. It detects the repo's own lint, typecheck, and test tooling from package.json, Makefile, pyproject.toml, or CI config.

Why does posting a review fail with HTTP 422?

GitHub returns 422 when you APPROVE or REQUEST_CHANGES your own PR, or when an inline comment anchors to a line outside the diff. The post script auto-falls back to COMMENT for own PRs and reports which anchors failed.

What happens on a second review of the same PR?

Round 2 and later only verify whether previous findings were resolved and scan new commits for blockers introduced by fixes. No new non-blocker findings are raised, and the gate enforces this convergence contract mechanically.

When should I not use this PR review skill?

Do not use it for reviewing prose or documents, fixing CI failures, resolving merge conflicts, writing fixes itself, or responding to review comments. It only judges pull request diffs and posts reviews.