deep-review

name	deep-review
description	Launch parallel deep-dive agents to review completed work for quality, standards, compatibility, and completeness
argument-hint	[feature-name]

Deep Review

Launch 4 parallel review agents to audit: $ARGUMENTS

Step 0: Identify Target Files

Find all files related to $ARGUMENTS:

• Main/Features/$ARGUMENTS/ — feature code
• LOTRAOM.Tests/Features/$ARGUMENTS/ — tests
• Main/Adapters/ — related adapters
• ModuleData/ — related XML config
• docs/ — feature documentation

Step 1: Launch 4 Parallel Agents (Sonnet)

Agent 1 — Standards Compliance

Review all target files for:

• ADR-002: Entry points <150 lines, delegate to services
• ADR-003: No #region blocks
• ADR-004: No [Obsolete] without full migration in same PR
• ADR-005: No #if DEBUG except IoC.cs
• ADR-007: Adapter pattern — no sealed TaleWorlds types in services
• No hardcoded secrets — IConfiguration for all config values
• Culture lookups — uses ICultureMapper.GetLotrCultureName(), never .Name.ToString()
• DryIoC registration — correct lifetime (Reuse.Singleton vs Transient)

Agent 2 — Bannerlord Compatibility

• Verify Harmony patch targets exist in the game DLLs
• Check GameModel overrides call base.Method() for unhandled cases
• Verify method signatures against decompiled source in External/Decompiled/
• Check for v1.2 → current API breaking changes
• Verify CampaignBehavior lifecycle hooks are correct

Agent 3 — Efficiency & Performance

• No unnecessary allocations in hot paths
• String concatenation in loops (use StringBuilder)
• LINQ in performance-critical code
• Correct use of ?. and ?? operators
• Thread safety in shared state
• Dictionary/HashSet vs List for lookups

Agent 4 — Completeness

• Test coverage — do tests exist? (80% minimum)
• CHANGELOG updated — [Unreleased] section reflects changes
• Documentation — public APIs have XML doc comments
• ADR created — if architectural decision was made
• Build passes — ./build.ps1
• No TODO/FIXME without linked issue
• GitHub issue — exists for this feature/fix

Step 2: Compile Report

DEEP REVIEW: [target]
=====================

CRITICAL: N findings
HIGH: N findings
MEDIUM: N findings
LOW: N findings

[Grouped findings by agent, with file:line references]

VERDICT: CLEAN / NEEDS FIXES / BLOCKED