deep-review

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

Deep Review

Launch 4 parallel review agents to audit the specified system or class: $ARGUMENTS

Step 0: Identify Target Files

Find all files related to $ARGUMENTS:

• src_new/scripts/ — core logic and class offenses
• src_new/triggers/ — game text pattern matching
• src_new/aliases/ — user command shortcuts
• src_new/tests/ — unit tests
• .claude/classes/ — class documentation

List all relevant files and their line counts.

Step 1: Launch 4 Parallel Review Agents (Sonnet)

Agent 1 — Lua Standards & Quality

Review all target files for:

• Lua 5.1 compatibility — no Lua 5.2+ features (goto, bitwise ops, string.format %a)
• Variable scoping — local declarations, no accidental globals (check luacheck would pass)
• Naming conventions — camelCase for locals, UPPER_CASE for constants, dot.notation for modules
• Dead code — unreachable branches, unused locals, commented-out blocks
• Error handling — pcall/xpcall around GMCP parsing, nil checks on table lookups
• Performance — avoid string concatenation in hot loops, prefer table.concat
• YAML header correctness — valid frontmatter, matching group hierarchy

Rate each finding: CRITICAL / HIGH / MEDIUM / LOW

Agent 2 — Combat System Correctness

Review target files for combat logic errors:

• Affliction IDs — verify all affliction names match Achaea's canonical set
• Balance/equilibrium gating — commands sent only when on balance
• Lock progression — affliction combinations follow valid lock routes (reference .claude/classes/lock_types.md)
• Illusion detection — proper GMCP cross-referencing to reject fake afflictions
• Curing priority — affliction priorities don't conflict with server-side curing
• State machine correctness — state transitions are complete (no missing cases)
• GMCP data trust — using gmcp.Char.Vitals, gmcp.Char.Afflictions correctly
• Class-specific mechanics — verify ability names, skill costs, cooldowns match game data

Agent 3 — Trigger & Pattern Correctness

Review trigger files for:

• Regex correctness — patterns match intended game text, no catastrophic backtracking
• Trigger ordering — numbered prefixes maintain correct fire order
• False positive risk — patterns too broad that match unintended text
• Capture group usage — captured values used correctly in trigger scripts
• Multi-line trigger coverage — multi-line game output handled properly
• Duplicate patterns — two triggers matching the same text
• Missing patterns — known game output not covered by any trigger

Agent 4 — Completeness & Documentation

Check that the work is complete:

• Test coverage — do unit tests exist for new/modified logic?
• CHANGELOG updated — does CHANGELOG.md reflect the changes?
• Class doc updated — if class offense changed, is .claude/classes/<class>.md current?
• AGENTS.md updated — if new patterns/conventions introduced, is AGENTS.md current?
• Build passes — ./build.sh succeeds
• Version consistency — 3-point version sync intact
• Missing files — any triggers referenced in code but not created?

Step 2: Compile Report

Merge findings from all 4 agents:

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

Step 3: Fix Issues (if requested)

For each CRITICAL or HIGH finding:

1. Make the fix
2. Run lua5.1 src_new/tests/test_runner.lua
3. Run ./build.sh