feat(v3r2): Wave 3 — Permission Stack, MX TAG v2, Hook Handler, Constitution Pipeline (#741)

* feat(v3r2): Wave 1+2 Bundle — Constitution + Runtime Foundation

Wave 1 — Constitution Foundation (CON-002, CON-003, SPC-001):
- CON-002: 5-layer amendment safety gate (frozen guard, canary check,
  contradiction detector, rate limiter, human oversight)
- CON-003: Rule tree consolidation (merge 4 deleted files, relocate
  content, frontmatter migration)
- SPC-001: EARS hierarchical acceptance criteria parser with spec
  linter subcommand

Wave 2 — Runtime Core (RT-005, RT-001, RT-004):
- RT-005: Multi-layer settings resolution with provenance tags
  (8-source priority stack: policy > user > project > local >
  plugin > skill > session > builtin)
- RT-001: Hook JSON-OR-ExitCode dual protocol (HookResponse schema,
  structured output for all 27 hook events)
- RT-004: Typed session state with phase checkpoints and
  resumable permission context

Supporting changes:
- Template mirrors for all rules/skills changes
- LSP client decision document
- Master wave plan document
- CLI subcommands: constitution, spec lint/view, state, doctor config

🤖 MoAI <email@mo.ai.kr>

* feat(runtime): SPEC-V3R2-RT-002 Permission Stack + Bubble Mode 구현

8-소스 Permission Stack (SrcPolicy > SrcUser > SrcProject > SrcLocal >
SrcPlugin > SrcSkill > SrcSession > SrcBuiltin) + 5 PermissionMode
(Default, AcceptEdits, BypassPermissions, Plan, Bubble) + BubbleDispatcher
for fork agent AskUserQuestion routing + moai doctor permission 서브커맨드

🧪 AC 15개 전체 통과 · 📦 7 files · 📊 ~750 LOC

🗿 MoAI <email@mo.ai.kr>

* feat(spc): SPEC-V3R2-SPC-002 @mx TAG v2 — Sidecar Index + TagScanner 구현

16-언어 TagScanner (Go/Python/TypeScript/Rust/Java/Kotlin/C#/Ruby/PHP/
Elixir/C++/Scala/R/Dart/Swift) + JSON sidecar 인덱스 (.moai/state/
mx-index.json) + 원자적 쓰기 (temp+rename) + 7일 TTL 만료 아카이빙 +
AnchorID resolver API (SPC-004 placeholder)

🧪 AC 15개 전체 통과 · 📦 9 files · 📊 ~800 LOC

🗿 MoAI <email@mo.ai.kr>

* feat(runtime): SPEC-V3R2-RT-006 Hook Handler 27-Event Coverage + P-H02 CRITICAL 수정

5 UPGRADE: subagentStop (P-H02 tmux pane 누수 수정), postToolUseFailure
(에러 분류 7종), instructionsLoaded (40K char budget), fileChanged
(MX re-scan 16언어), configChange (YAML validation reload)

1 REMOVE: setup.go orphan 삭제
4 RETIRE-OBS-ONLY: notification/elicitation/elicitationResult/taskCreated
신규 audit_test.go: 핸들러 수 패리티 + retired event 검증

🧪 AC 17개 전체 통과 · 📦 18 files · 📊 ~1,000 LOC

🗿 MoAI <email@mo.ai.kr>

* fix(lint): golangci-lint 20개 이슈 수정 (errcheck/staticcheck/unused)

- errcheck: defer f.Close(), os.Remove(), json.Unmarshal 등 반환값 검사 추가
- ST1005: 에러 문자열 첫 글자 소문자화 (Layer → layer)
- SA9003: 빈 분기 제거 (isFork bypass 로직 코멘트로 대체)
- QF1003: if/else → tagged switch 전환
- unused: parseScoreFromProgress에 nolint 지시어 추가 (TODO 예약)
- unused: evolutionLogPath 인라인 처리

🗿 MoAI <email@mo.ai.kr>

Author: GoosLab

.claude/rules/moai/core/moai-constitution.md

@@ -1,6 +1,5 @@
 ---
-description: Core constitutional principles for MoAI orchestrator - HARD rules that must always be followed
-globs:
+paths: "**/.claude/**"
 ---
 
 # MoAI Constitution

.claude/rules/moai/development/coding-standards.md

@@ -1,6 +1,5 @@
 ---
-description: MoAI-specific coding standards for instruction documents and configuration files
-globs: .claude/**/*.md, .claude/**/*.yaml, .moai/**/*.yaml, CLAUDE.md
+paths: ".claude/**/*.md,.claude/**/*.yaml,.moai/**/*.yaml,CLAUDE.md"
 ---
 
 # Coding Standards

.claude/rules/moai/workflow/context-window-management.md

@@ -73,7 +73,7 @@ This rule applies to all MoAI workflows:
 ## Cross-references
 
 - `feedback_large_spec_wave_split.md` (auto-memory) — wave-split mitigation for SPECs with 30+ tasks
-- `.claude/rules/moai/workflow/file-reading-optimization.md` — token budget per file read
+- `.claude/skills/moai/references/file-reading-optimization.md` — token budget per file read
 - `output-styles/moai/moai.md` §6 (Persistence & Context Awareness) — orchestrator persistence pattern
 - CLAUDE.md §11 (Error Handling) — token-limit recovery flow
 

.claude/rules/moai/workflow/spec-workflow.md

@@ -45,9 +45,55 @@ Token Strategy:
 - Selective file loading
 - Enables 70% larger implementations
 
-Development Methodology:
-- Configured in quality.yaml (development_mode: ddd or tdd)
-- See @workflow-modes.md for detailed methodology cycles
+Development Methodology (configured in quality.yaml development_mode):
+
+### DDD Mode — ANALYZE-PRESERVE-IMPROVE
+
+Best for existing projects with < 10% test coverage. Uses manager-ddd agent.
+
+**ANALYZE**: Read existing code, map domain boundaries, identify side effects and implicit contracts.
+**PRESERVE**: Write characterization tests capturing current behavior. Create behavior snapshots for regression detection.
+**IMPROVE**: Make small incremental changes. Run characterization tests after each change. Refactor with test validation.
+
+### TDD Mode — RED-GREEN-REFACTOR (default)
+
+Best for all development work, new projects, and brownfield with 10%+ coverage. Uses manager-tdd agent.
+
+**RED**: Write a failing test describing desired behavior. Verify it fails. One test at a time.
+**GREEN**: Write simplest implementation that passes. No premature optimization.
+**REFACTOR**: Clean up while keeping tests green. Extract patterns, remove duplication.
+
+Brownfield enhancement: Pre-RED step reads existing code to understand current behavior before writing the failing test.
+
+### Methodology Auto-Detection
+
+| Project State | Test Coverage | Recommendation |
+|--------------|---------------|----------------|
+| Greenfield (new) | N/A | TDD |
+| Brownfield | >= 10% | TDD |
+| Brownfield | < 10% | DDD |
+
+Manual override: `quality.development_mode` in quality.yaml, `MOAI_DEVELOPMENT_MODE` env var, or `moai init --mode <ddd|tdd>`.
+
+### Pre-submission Self-Review
+
+Before marking implementation complete: review full diff against SPEC acceptance criteria. Ask "Is there a simpler approach?" and "Would removing any changes still satisfy the SPEC?" Skip for single-file changes under 50 lines, bug fixes with reproduction test, or user-approved annotation cycle changes.
+
+### Drift Guard
+
+After each methodology cycle, compare planned files against actual modifications. Warns at <= 30% drift. Triggers re-planning (Phase 2.7) above 30%.
+
+### Team Mode Methodology
+
+Each teammate applies the methodology within their file ownership scope. team-validator validates compliance. team-tester exclusively owns test files.
+
+### MX Tag Integration
+
+| Phase | TDD Action | DDD Action |
+|-------|-----------|-----------|
+| Test/Analyze | RED: add `@MX:TODO` | ANALYZE: 3-Pass scan, identify targets |
+| Implement/Preserve | GREEN: remove `@MX:TODO` | PRESERVE: validate tags, add `@MX:LEGACY` |
+| Refactor/Improve | REFACTOR: add `@MX:NOTE` | IMPROVE: update tags, add `@MX:NOTE` |
 
 Success Criteria:
 - All SPEC requirements implemented

.claude/rules/moai/workflow/team-pattern-cookbook.md

@@ -1,6 +1,6 @@
 ---
 description: "Five team execution patterns with role profiles and communication protocols"
-paths: ".moai/config/sections/workflow.yaml,.claude/rules/moai/workflow/team-protocol.md"
+paths: ".moai/config/sections/workflow.yaml,.claude/rules/moai/workflow/worktree-integration.md"
 ---
 
 <!-- Source: revfactory/harness — Apache License 2.0 — see .claude/rules/moai/NOTICE.md -->