judge_skill

This skill reviews code changes using codex review to detect bugs, security gaps, and intent mismatches, delivering actionable, evidence-based findings.

Shell

8

GitHub Stars

1

Bundled Files

2 months ago

Catalog Refreshed

4 months ago

First Indexed

Readme & install

Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.

Installation

Preview and clipboard use veilstrat where the catalogue uses aiagentskills.

npx veilstrat add skill simota/agent-skills --skill judge

SKILL.md7.9 KB

Overview

This skill is an automated code-review agent that uses codex review to find correctness, security, and intent-alignment problems in PRs, commits, and pre-commit changes. It classifies findings by severity, produces evidence-backed reports, and routes issues to the right fixer agent. The focus is on actionable, non-stylistic verdicts so teams can triage and fix real risks quickly.

How this skill works

I run codex review with the appropriate mode (PR --base, commit SHA, or --uncommitted) and parse results into severity buckets (CRITICAL/HIGH/MEDIUM/LOW/INFO). I filter contextual false positives, verify intent against the PR or commit message, and attach line-specific evidence and suggested remediation owners. Finally, I generate a structured report and recommend which agent should fix each class of issues.

When to use it

Before merging a pull request to catch regressions or intent mismatch
As a pre-commit gate to catch high-severity problems earlier
To review a specific commit SHA after CI or investigation
When you need a quick surface-level security screen before escalation
When verifying that fixes actually resolve the root cause without regressions

Best practices

Always clarify review scope first (base branch, files, or commit) and ask for missing context
Run with --uncommitted when local changes exist to avoid scope drift
Prioritize CRITICAL/HIGH findings and attach minimal repro evidence for each
Route security findings to the deep-analysis agent (Sentinel) rather than attempting fixes here
Use the report’s intent-alignment section to resolve mismatches before coding changes

Example use cases

Automated PR gate: run on every PR and block merges with critical issues
Pre-commit check for CI-enforced safety before pushing code
Post-fix verification: confirm a Builder’s patch actually resolves the reported root cause
Security triage: surface potential vulnerabilities and hand off to Sentinel for deep dive
Quality handoff: detect cross-file inconsistency and route to refactorer (Zen)

FAQ

No. I only report issues and suggest remediation agents; fixes are performed by Builder or Zen.

How do you handle false positives?

I apply contextual filtering rules from codex review guidance and include evidence so reviewers can quickly validate or dismiss findings.