@ai-dev-methodologies/rlp-desk 0.5.1 → 0.5.2

package/README.md

@@ -49,8 +49,12 @@ You'll be asked to confirm each item:
 
 ### 3. Run
 
-```
-/rlp-desk run calculator
+```bash
+# Recommended (cross-engine + final consensus):
+/rlp-desk run <slug> --mode tmux --worker-model spark:high --consensus final-only --debug
+
+# Claude-only:
+/rlp-desk run <slug> --debug
 ```
 
 The leader loop runs autonomously — spawning workers, verifying results, and tracking progress until completion or a circuit breaker triggers.
@@ -232,28 +236,39 @@ When all US pass individually, the final ALL verify runs **sequentially per-US**
 
 ### Run Options
 
-| Flag | Default | Description |
-|------|---------|-------------|
-| `--max-iter N` | 100 | Maximum iterations before timeout |
-| `--mode agent\|tmux` | agent | Execution mode (see below) |
-| `--worker-model MODEL` | sonnet | Claude worker model (haiku/sonnet/opus) |
-| `--worker-engine claude\|codex` | claude | Worker engine |
-| `--verifier-model MODEL` | auto | Auto-selected: +1 tier (same-engine) or cross-engine |
-| `--verifier-engine claude\|codex` | auto | Opposite of worker engine if codex available |
-| `--codex-model MODEL` | gpt-5.4 | Codex model (spark requires GPT Pro) |
-| `--codex-reasoning LEVEL` | medium | low/medium/high/xhigh |
-| `--verify-mode per-us\|batch` | per-us | Verification strategy (see below) |
-| `--lock-worker-model` | off | Disable progressive model upgrade on failure |
-| `--debug` | off | Debug logging to `logs/<slug>/debug.log` |
-| `--with-self-verification` | off | Campaign-level post-loop analysis report |
+| Option | Default | Description |
+|--------|---------|-------------|
+| `--mode agent\|tmux` | agent | agent=LLM Leader, tmux=shell Leader |
+| `--worker-model MODEL` | haiku | Worker model. `name`=claude, `name:reasoning`=codex |
+| `--lock-worker-model` | off | Disable auto model upgrade on failure |
+| `--verifier-model MODEL` | sonnet | per-US verification model (lighter) |
+| `--final-verifier-model MODEL` | opus | final ALL verification model (stricter) |
+| `--consensus off\|all\|final-only` | off | Cross-engine consensus scope |
+| `--consensus-model MODEL` | gpt-5.4:medium | per-US cross-verifier (lighter) |
+| `--final-consensus-model MODEL` | gpt-5.4:high | final cross-verifier (stricter) |
+| `--verify-mode per-us\|batch` | per-us | per-us: verify each US → final ALL |
+| `--cb-threshold N` | 6 | Consecutive failures → BLOCKED |
+| `--max-iter N` | 100 | Max iterations → TIMEOUT |
+| `--iter-timeout N` | 600 | Per-iteration timeout seconds (tmux only) |
+| `--debug` | off | Debug logging |
+| `--with-self-verification` | off | Post-campaign SV report |
+
+#### Per-US vs Final Verification
+
+RLP Desk runs two distinct verification passes:
+
+- **Per-US** (`--verifier-model`, default: sonnet) — runs after each user story completes. Lightweight and fast, catches issues early before later stories build on broken foundations.
+- **Final ALL** (`--final-verifier-model`, default: opus) — runs once after all user stories pass individually. Stricter and more thorough, catches cross-US integration issues and anything per-US missed.
+
+When `--consensus` is enabled, a second cross-engine verifier runs alongside each pass: `--consensus-model` for per-US and `--final-consensus-model` for the final ALL gate. Both engines must pass.
 
 ### Init Presets
 
 After `brainstorm`, `init` detects your environment and presents run command presets:
 
-- **Codex detected** → recommends cross-engine mode (`--worker-model gpt-5.4:high --verify-consensus`)
-- **GPT Pro (spark)** → offers spark preset (`--worker-model gpt-5.3-codex-spark:high`)
-- **Claude-only** → defaults to `--worker-model sonnet` with opus verifier
+- **Codex detected (GPT Pro / spark)** → recommends cross-engine mode (`--worker-model spark:high --consensus final-only`)
+- **Codex detected (large PRD, AC > 15)** → offers gpt-5.4 preset (`--worker-model gpt-5.4:high --consensus final-only`)
+- **Claude-only** → defaults to `--debug` with haiku worker and opus final verifier
 - **Basic** → minimal flags for quick iteration
 
 The brainstorm phase evaluates complexity (US count, file scope, logic, dependencies, code impact) and recommends a starting model. You can override any recommendation.
@@ -332,9 +347,8 @@ RLP Desk supports two execution engines for Worker and Verifier. **Claude is the
 
 ### Claude (default)
 
-```
+```bash
 /rlp-desk run calculator
-/rlp-desk run calculator --worker-engine claude --verifier-engine claude
 ```
 
 Uses Claude Code's `Agent()` tool (agent mode) or `claude -p` CLI (tmux mode). Supports dynamic model routing (haiku/sonnet/opus).
@@ -345,17 +359,17 @@ Uses Claude Code's `Agent()` tool (agent mode) or `claude -p` CLI (tmux mode). S
 # Install codex CLI first
 npm install -g @openai/codex
 
-# Run with codex worker
-/rlp-desk run calculator --worker-engine codex
+# Run with codex worker (spark requires GPT Pro)
+/rlp-desk run calculator --worker-model spark:high
 
 # Customize model and reasoning effort
-/rlp-desk run calculator --worker-engine codex --codex-model gpt-5.4 --codex-reasoning high
+/rlp-desk run calculator --worker-model gpt-5.4:high
 
-# Mix engines: codex worker, claude verifier
-/rlp-desk run calculator --worker-engine codex --verifier-engine claude
+# Cross-engine: codex worker, claude verifier (recommended)
+/rlp-desk run calculator --worker-model spark:high --consensus final-only --debug
 ```
 
-Uses the `codex` CLI via `Bash()` (agent mode) or as an interactive TUI (tmux mode). The `codex` binary is only required when an engine is set to `codex`.
+The engine is inferred automatically from the `--worker-model` value: a plain model name (e.g. `haiku`) routes to Claude, while `name:reasoning` format (e.g. `spark:high`) routes to Codex. The `codex` binary is only required when a codex model is specified.
 
 | Engine | Agent Mode | Tmux Mode | Dynamic Routing |
 |--------|-----------|-----------|-----------------|

package/docs/plans/frolicking-churning-honey.md

@@ -0,0 +1,253 @@
+# Plan: rlp-desk 옵션 인터페이스 정리 + brainstorm/init 추천 보강
+
+## Context
+
+rlp-desk 옵션이 유기적으로 성장하면서 20개+로 늘어남. 중복, 충돌, 미구분 문제 발생:
+- `--worker-engine`, `--worker-codex-model`, `--worker-codex-reasoning`은 `--worker-model`이 이미 통합 처리 (parse_model_flag)
+- consensus 옵션 3중 충돌 (`--verify-consensus`, `--final-consensus`, `--consensus-scope`)
+- per-US Verifier vs Final Verifier 구분 없음
+- brainstorm 모델 추천이 2열(Worker/Verifier)만 있어 per-US/Final 구분 안 됨
+- gpt-5.3-codex 경로가 유저 결정 없이 코드에 존재
+- brainstorm/init에서 정책 기반 추천이 부족 (batch capacity, spark vs gpt-5.4 기준 등)
+
+## 근거 문서
+
+| 근거 | 위치 | 핵심 내용 |
+|------|------|----------|
+| 검증 철학 | `memory/feedback_verification_philosophy.md` | 기본값: haiku/sonnet/opus, per-US 가볍게, final 엄격, cross-engine 추천 |
+| CB threshold | `memory/feedback_cb_threshold_six.md` | CB=6 확정, claude 3단계 커버 |
+| 모델 구분 | `memory/feedback_no_assumption.md` | spark=Pro 적극 추천, gpt-5.4=fallback, gpt-5.3-codex 미확정 |
+| Task sizing | `memory/feedback_us_size_limit.md` | Worker comfortable zone 안에 분할 |
+| 현행 옵션 | `src/commands/rlp-desk.md:186-207` | 현재 20개+ 옵션 목록 |
+| parse_model_flag | `src/scripts/lib_ralph_desk.zsh:38-57` | `모델명`=claude, `모델명:추론`=codex 통합 파싱 |
+| get_next_model | `src/scripts/lib_ralph_desk.zsh:81-110` | 엔진별 업그레이드 경로 |
+| check_model_upgrade | `src/scripts/lib_ralph_desk.zsh:116-168` | Worker만 상향, Verifier 고정 |
+
+---
+
+## 최종 옵션 (14개)
+
+### 1. 실행
+
+| # | 옵션 | 고정 기본값 | 설명 |
+|---|------|------------|------|
+| 1 | `--mode agent\|tmux` | `agent` | agent=LLM Leader, tmux=shell Leader |
+
+### 2. Worker
+
+| # | 옵션 | 고정 기본값 | 포맷 | 설명 |
+|---|------|------------|------|------|
+| 2 | `--worker-model` | `haiku` | `haiku` 또는 `spark:high` | `모델명`=claude, `모델명:추론`=codex. parse_model_flag()가 engine/model/reasoning 자동 분리 |
+| 3 | `--lock-worker-model` | off | flag | 실패 시 자동 모델 업그레이드(check_model_upgrade) 비활성화 |
+
+### 3. Verifier (per-US)
+
+| # | 옵션 | 고정 기본값 | 포맷 | 설명 |
+|---|------|------------|------|------|
+| 4 | `--verifier-model` | `sonnet` | `sonnet` 또는 `gpt-5.4:high` | per-US 검증 모델. 캠페인 내내 고정 (progressive upgrade 없음) |
+
+### 4. Verifier (final ALL)
+
+| # | 옵션 | 고정 기본값 | 포맷 | 설명 |
+|---|------|------------|------|------|
+| 5 | `--final-verifier-model` | `opus` | `opus` 또는 `gpt-5.4:high` | final ALL 검증 모델. 미지정 시 opus 고정 (per-US와 독립) |
+
+### 5. Consensus
+
+| # | 옵션 | 고정 기본값 | 설명 |
+|---|------|------------|------|
+| 6 | `--consensus` | `off` | `off`: 단일 엔진 / `all`: 매 verify 교차 / `final-only`: final ALL에서만 교차 |
+| 7 | `--consensus-model` | `gpt-5.4:medium` | per-US 교차 verifier. 가볍게 |
+| 8 | `--final-consensus-model` | `gpt-5.4:high` | final 교차 verifier. 엄격하게. spark 불가(100k limit) |
+
+consensus 동작:
+
+| 상황 | 주 verifier | 교차 verifier |
+|------|------------|--------------|
+| per-US, 주=claude | `--verifier-model` (sonnet) | `--consensus-model` (gpt-5.4:medium) |
+| per-US, 주=codex | `--verifier-model` (gpt-5.4:high 등) | claude `opus` 고정 |
+| final, 주=claude | `--final-verifier-model` (opus) | `--final-consensus-model` (gpt-5.4:high) |
+| final, 주=codex | `--final-verifier-model` (gpt-5.4:high 등) | claude `opus` 고정 |
+
+- 양쪽 모두 pass해야 통과. 엔진 우선권 없음.
+- spark는 consensus 교차에 사용 불가 (100k output limit).
+
+### 6. 검증 전략
+
+| # | 옵션 | 고정 기본값 | 설명 |
+|---|------|------------|------|
+| 9 | `--verify-mode` | `per-us` | `per-us`: US마다 검증→final ALL / `batch`: 전부 후 한번에 |
+
+### 7. 안전장치
+
+| # | 옵션 | 고정 기본값 | 설명 |
+|---|------|------------|------|
+| 10 | `--cb-threshold` | `6` | 연속 fail N회 → BLOCKED. consensus 시 자동 ×2 (=12) |
+| 11 | `--max-iter` | `100` | 최대 iteration → TIMEOUT |
+| 12 | `--iter-timeout` | `600` | iteration당 timeout 초 (tmux만 적용) |
+
+### 8. 로깅/분석
+
+| # | 옵션 | 고정 기본값 | 설명 |
+|---|------|------------|------|
+| 13 | `--debug` | off | `~/.claude/ralph-desk/analytics/<slug>/debug.log` |
+| 14 | `--with-self-verification` | off | campaign 후 SV 리포트 → 다음 brainstorm 피드백(§8½) |
+
+---
+
+## 제거 옵션 (내부 구현으로 숨김)
+
+| 제거 옵션 | 대체 방법 |
+|-----------|----------|
+| `--worker-engine` | `--worker-model` 포맷에서 자동 추론 |
+| `--worker-codex-model` | `--worker-model spark:high`에 포함 |
+| `--worker-codex-reasoning` | `--worker-model spark:high`에 포함 |
+| `--verifier-engine` | `--verifier-model` 포맷에서 자동 추론 |
+| `--verifier-codex-model` | `--verifier-model`에 포함 |
+| `--verifier-codex-reasoning` | `--verifier-model`에 포함 |
+| `--verify-consensus` | `--consensus all` |
+| `--final-consensus` | `--consensus final-only` |
+| `--consensus-scope` | `--consensus`에 통합 |
+| `--consensus-fail-fast` | 제거 (복잡도 대비 가치 낮음) |
+
+내부적으로 env var (`WORKER_ENGINE`, `WORKER_CODEX_MODEL` 등)는 유지 — `parse_model_flag()`가 `--worker-model`에서 분리하여 설정.
+
+---
+
+## Brainstorm 모델 추천 구조 (신규)
+
+### Claude-only (codex 미설치)
+
+| 복잡도 | Worker | per-US Verifier | Final Verifier | Consensus |
+|--------|--------|----------------|----------------|-----------|
+| LOW | haiku | sonnet | opus | off |
+| MEDIUM | sonnet | opus | opus | off |
+| HIGH | opus | opus | opus | off |
+| CRITICAL | opus | opus | opus + human | off |
+
+### Cross-engine (codex 설치, ★ 추천)
+
+| 복잡도 | Worker | per-US Verifier | Final Verifier | Consensus |
+|--------|--------|----------------|----------------|-----------|
+| LOW | spark:high | sonnet | opus | final-only |
+| MEDIUM | spark:high | opus | opus | final-only |
+| HIGH | gpt-5.4:high | opus | opus | all |
+| CRITICAL | gpt-5.4:high | opus | opus + human | all |
+
+Worker 모델 선택 기준:
+- **spark:high** — 기본 추천. Pro 토큰풀 분리 = 비용 절감. PRD AC ≤ 15개
+- **gpt-5.4:high** — spark 100k output limit 초과 시 fallback. PRD AC > 15개
+
+### Brainstorm 추천 예시 (codex 있을 때)
+
+```
+★ 추천: cross-engine + final-consensus
+/rlp-desk run <slug> --mode tmux --worker-model spark:high --consensus final-only --debug
+
+대규모 PRD (spark 100k 초과):
+/rlp-desk run <slug> --mode tmux --worker-model gpt-5.4:high --consensus final-only --debug
+
+극도로 중요 (전체 consensus):
+/rlp-desk run <slug> --mode tmux --worker-model gpt-5.4:high --consensus all --debug
+
+claude-only:
+/rlp-desk run <slug> --debug
+```
+
+---
+
+## Batch Capacity Check (신규)
+
+brainstorm에서 batch 모드 + 대규모 PRD 감지 시 자동 경고:
+
+| 조건 | 경고 | 제안 |
+|------|------|------|
+| batch + spark + AC > 10 | "spark 100k output limit 주의" | wave split 또는 gpt-5.4 전환 |
+| batch + gpt-5.4 + AC > 15 | "단일 batch에 AC 과다" | wave split (US 3-4개씩 묶음) |
+| per-us + 어떤 모델이든 | 경고 없음 | US별 처리라 limit 문제 없음 |
+
+---
+
+## 확정 사항
+
+1. **교차 verifier claude 기본값**: `opus` 확정 (주 verifier가 codex일 때)
+2. **gpt-5.3-codex 경로**: get_next_model()에서 **제거** 확정 (spark + gpt-5.4만 유지)
+3. **`--verifier-model` 기본값**: `sonnet` (per-US 가볍게, feedback_verification_philosophy.md)
+4. **`--consensus-model` 기본값**: `gpt-5.4:medium` (per-US 교차, 가볍게)
+5. **`--final-consensus-model` 기본값**: `gpt-5.4:high` (final 교차, 엄격하게)
+6. **교차 claude** (주 verifier가 codex일 때): `opus` 고정
+
+---
+
+## Brainstorm/Init 추천 흐름 (보강)
+
+brainstorm 단계 7~8번을 아래로 교체:
+
+### brainstorm step 7: 모델 추천
+
+1. 복잡도 평가 (5-factor table, 현행 유지)
+2. codex 설치 여부 확인 (`command -v codex`)
+3. **4열 추천 테이블 제시** (Worker / per-US Verifier / Final Verifier / Consensus)
+4. spark vs gpt-5.4 판단: AC 총 개수 세어서
+   - AC ≤ 15 → spark:high 추천 ("Pro 토큰풀 분리, 비용 절감")
+   - AC > 15 → gpt-5.4:high 추천 ("spark 100k output limit 초과")
+5. verify-mode가 batch이면 batch capacity check 실행 (아래 참조)
+
+### brainstorm step 8 (init 후 명령어 제시): 추천 순서
+
+codex 있을 때:
+```
+★ 추천: cross-engine + final-consensus (비용 절감 + blind-spot 커버)
+/rlp-desk run <slug> --mode tmux --worker-model spark:high --consensus final-only --debug
+
+대규모 PRD (AC > 15, spark limit 초과):
+/rlp-desk run <slug> --mode tmux --worker-model gpt-5.4:high --consensus final-only --debug
+
+극도로 중요 (매 US 교차검증):
+/rlp-desk run <slug> --mode tmux --worker-model gpt-5.4:high --consensus all --debug
+
+claude-only (교차검증 없음):
+/rlp-desk run <slug> --debug
+```
+
+codex 없을 때:
+```
+★ 추천:
+/rlp-desk run <slug> --mode tmux --debug
+
+⚠️ codex 미설치 — 같은 엔진은 같은 blind spot을 공유합니다.
+cross-engine 교차검증: npm install -g @openai/codex
+```
+
+### init 후 옵션 설명
+
+init 완료 후 명령어 제시 시 아래 설명 포함:
+
+```
+옵션 설명:
+  --worker-model          Worker 모델 (haiku|sonnet|opus 또는 spark:high|gpt-5.4:high)
+  --verifier-model        per-US 검증 모델 (기본: sonnet, 가볍게)
+  --final-verifier-model  final ALL 검증 모델 (기본: opus, 엄격하게)
+  --consensus             교차검증 (off|all|final-only)
+  --consensus-model       per-US 교차 모델 (기본: gpt-5.4:medium)
+  --final-consensus-model final 교차 모델 (기본: gpt-5.4:high)
+```
+
+---
+
+## 변경 대상 파일
+
+| 파일 | 변경 내용 |
+|------|----------|
+| `src/commands/rlp-desk.md` | 옵션 14개로 정리, brainstorm 추천 4열 테이블, batch capacity check, init 추천 예시 보강 |
+| `src/governance.md` | consensus 섹션 정리, `--final-verifier-model` + `--final-consensus-model` 문서화, §7b 교차검증 테이블 |
+| `src/scripts/run_ralph_desk.zsh` | `--consensus` 통합 플래그, `--final-verifier-model`, `--final-consensus-model` 파싱, deprecated 옵션 경고 |
+| `src/scripts/lib_ralph_desk.zsh` | gpt-5.3-codex 경로 제거 (spark + gpt-5.4만 유지) |
+| `README.md` | 옵션 레퍼런스 테이블 14개로 갱신, 추천 예시, per-US/final 구분 설명 |
+
+## 검증
+
+1. 기존 테스트 regression: `bash tests/test_cb_and_analytics.sh && bash tests/test_us004_progressive_upgrade.sh && bash tests/test_v052_improvements.sh`
+2. 신규 테스트: `tests/test_option_cleanup.sh` — 옵션 파싱, 기본값, deprecated 경고
+3. `zsh -n` syntax check
+4. Self-verification 3 scenarios (CLAUDE.md gate)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-dev-methodologies/rlp-desk",
-  "version": "0.5.1",
+  "version": "0.5.2",
   "description": "Fresh-context iterative loops for Claude Code — autonomous task completion with independent verification",
   "scripts": {
     "postinstall": "node scripts/postinstall.js",