review-codex

name	review-codex
description	Auto-detect what was built, write a Codex adversarial prompt, dispatch, then verify results
argument-hint	[optional: feature-name or path-to-review.md]

Codex Adversarial Review Pipeline

Full Codex review lifecycle.

Argument handling:

• No argument: auto-detect from git, write prompt, dispatch
• Feature name: write prompt for that feature, dispatch
• Path to .md file: verify existing review, implement fixes

Phase 1: Detect What to Review

1. git diff --name-only HEAD~3..HEAD for recently changed .cs files
2. Group by feature directory
3. Check docs/reviews/ for recent codex-adversarial-*.md
   • Found: Phase 3. Not found: Phase 2.

Phase 2: Write Codex Prompt

2a: Gather files

• Main/Features/{feature}/ — all .cs files
• LOTRAOM.Tests/Features/{feature}/ — tests
• Main/Adapters/ — related adapters
• ModuleData/ — config files

2b: Identify Known Suspects

• Missing adapter wrapping (sealed types in services)
• Entry points over 150 lines
• Culture lookups using .Name.ToString() instead of ICultureMapper
• GameModel overrides not calling base
• Harmony patches on non-existent methods
• Config values hardcoded instead of IConfiguration

2c: Write and dispatch

Include: feature description, LOTRAOM architecture cheatsheet, Known Suspects, file lists.

/codex:rescue [full prompt]

Phase 3: Verify Codex Review

3a: Read the review

3b: Verify each finding against source

3c: Produce assessment table

| # | Codex Severity | Your Severity | Agree? | Reason |

3d: Implement confirmed fixes

1. Make code changes
2. ./build.ps1 -RunTests -MinCoverage 80

Rules

• NEVER implement a fix without reading the source first
• Decompile vanilla targets for Harmony/GameModel findings
• Build and test after every batch of fixes