triflux 10.0.0 → 10.0.2

package/skills/tfx-codex-swarm-workspace/iteration-1/full-swarm-all-prds/without_skill/outputs/analysis.md

@@ -0,0 +1,92 @@
+# Codex Swarm Analysis — Full 7-PRD Execution Plan
+
+## Overview
+
+- **Source**: 7 PRD files in `.omx/plans/`
+- **Issues**: #24, #25, #26, #27, #28, #29, #30
+- **Classification**: 5 implement, 2 investigate
+- **User directive**: implement -> `$plan` -> `$autopilot`, investigate -> `$plan` only
+- **Execution**: Codex CLI in `--full-auto` mode, one psmux session per PRD, all in one WT window as tabs
+
+## What Was Generated
+
+| File | Purpose |
+|------|---------|
+| `classification.md` | PRD-by-PRD type classification with rationale |
+| `routing.md` | Codex profile routing (model, effort level) per PRD |
+| `commands.sh` | Full executable script: worktrees, prompts, psmux, codex, WT tabs |
+| `analysis.md` | This summary |
+
+## Execution Pipeline
+
+```
+Phase 1: git worktree add (7 isolated worktrees)
+Phase 2: Copy PRD files into worktrees
+Phase 3: Generate per-task prompt files with OMX skill instructions
+Phase 4: psmux new-session + codex send-keys (7 sessions)
+Phase 5: psmux attach --wt-tab (1 WT window, 7 tabs)
+```
+
+## Session Map
+
+| Tab | Session | Issue | Type | Model | Skill Flow |
+|-----|---------|-------|------|-------|------------|
+| 1 | codex-swarm-24 | file-transfer | implement | gpt-5.3-codex (high) | $plan -> $autopilot |
+| 2 | codex-swarm-25 | resize-blank | investigate | gpt-5.4 (high) | $plan only |
+| 3 | codex-swarm-26 | session-cleanup | implement | gpt-5.3-codex (high) | $plan -> $autopilot |
+| 4 | codex-swarm-27 | dashboard-anchor | implement | gpt-5.3-codex (high) | $plan -> $autopilot |
+| 5 | codex-swarm-28 | guard-deadlock | implement | gpt-5.3-codex (high) | $plan -> $autopilot |
+| 6 | codex-swarm-29 | cwd-propagation | implement | gpt-5.3-codex (high) | $plan -> $autopilot |
+| 7 | codex-swarm-30 | terminal-minimize | investigate | gpt-5.4 (high) | $plan only |
+
+## Isolation Strategy
+
+Each session runs in its own git worktree under `.codex-swarm/wt-issue-{N}` on branch `codex/issue-{N}`. This prevents file conflicts between parallel Codex sessions. PRD files are copied into each worktree so the Codex agent can read them locally.
+
+## Key Design Decisions
+
+### 1. Model Selection
+- **Implement tasks** (5): `gpt-5.3-codex` with high reasoning effort. Codex-optimized model for code generation.
+- **Investigate tasks** (2): `gpt-5.4` with high reasoning effort. General model better suited for analysis and investigation planning.
+- All 7 PRDs sized as L (standard) based on line count (69-76 lines) and affected file counts (2-5 files).
+
+### 2. OMX Skill Routing
+- User explicitly requested: implement -> autopilot, investigate -> plan only.
+- Implement prompts instruct: `$plan` first, then `$autopilot` for autonomous execution.
+- Investigate prompts instruct: `$plan` only, explicitly stating "do not implement".
+
+### 3. WT Layout
+- User requested "one WT with tabs" -- all 7 sessions are tabs in a single WT window.
+- First session uses `--wt-new-window`, remaining 6 use `--wt-tab`.
+- wt.exe fallback is provided as a commented-out alternative.
+
+### 4. No Execution
+- Per instructions, commands are generated but NOT executed.
+- The `commands.sh` script is ready to run when the user decides to launch.
+
+## Monitoring Commands
+
+```bash
+# Check status of all sessions
+for N in 24 25 26 27 28 29 30; do
+  echo "=== Issue #$N ==="
+  psmux capture-pane --session "codex-swarm-$N" --lines 5
+done
+
+# List active sessions
+psmux list-sessions | grep codex-swarm
+
+# Kill all swarm sessions (cleanup)
+for N in 24 25 26 27 28 29 30; do
+  psmux kill-session --name "codex-swarm-$N"
+done
+git worktree prune
+rm -rf .codex-swarm/
+```
+
+## Risk Notes
+
+- **Issue #25 and #30** (investigate): These have "evidence insufficient" validity verdicts. The investigation plans may conclude that no code fix is needed. Using `$plan` only is appropriate.
+- **Issue #28** (partially valid): The deadlock claim is overstated per validity review, but the UX improvement is still warranted. Autopilot should handle the bounded scope.
+- **Issue #29**: Touches the most files (5 across CLI parse, headless, and backend layers). Most likely to need careful cross-file coordination.
+- **Parallel execution**: 7 simultaneous Codex sessions will consume significant API quota. All are routed to high-effort profiles.

package/skills/tfx-codex-swarm-workspace/iteration-1/full-swarm-all-prds/without_skill/outputs/classification.md

@@ -0,0 +1,71 @@
+# PRD Classification — 7 PRDs in .omx/plans/
+
+## Classification Criteria
+
+Based on PRD content analysis against SKILL.md Step 3 rules:
+- **implement**: keywords "구현", "implement", "추가", "변경", "fix", actionable code changes
+- **investigate**: keywords "조사", "investigation", "재현", "reproduce", runtime-only evidence needed
+
+User override: implement -> `$plan` -> `$autopilot`, investigate -> `$plan` only.
+
+---
+
+## PRD-by-PRD Classification
+
+### 1. prd-issue-24-remote-spawn-file-transfer.md
+- **Type**: implement
+- **Rationale**: "Add an explicit file-transfer phase before remote Claude launch", "Copy transferred files", "rewrite the prompt". Clear implementation deliverable with bounded code changes in `scripts/remote-spawn.mjs`.
+- **Validity verdict**: mostly valid
+- **OMX skill**: `$plan` -> `$autopilot`
+
+### 2. prd-issue-25-remote-spawn-resize-blank-screen.md
+- **Type**: investigate
+- **Rationale**: "Produce a deterministic reproduction procedure", "investigation checklist", "instrumentation-first workflow". Explicitly framed as investigation, not implementation. Verdict was "evidence insufficient".
+- **Validity verdict**: evidence insufficient
+- **OMX skill**: `$plan` (investigation mode)
+
+### 3. prd-issue-26-remote-spawn-session-cleanup.md
+- **Type**: implement
+- **Rationale**: "Add a normal-exit watcher", "trigger killPsmuxSession(sessionName)". Clear implementation target: add cleanup logic to spawn lifecycle.
+- **Validity verdict**: valid
+- **OMX skill**: `$plan` -> `$autopilot`
+
+### 4. prd-issue-27-hud-dashboard-anchor.md
+- **Type**: implement
+- **Rationale**: "Replace focus-relative split behavior with a dedicated dashboard target strategy", "Remove or narrow mf up focus manipulation". Concrete implementation changes to `hub/team/headless.mjs`.
+- **Validity verdict**: valid
+- **OMX skill**: `$plan` -> `$autopilot`
+
+### 5. prd-issue-28-headless-guard-spawn-deadlock.md
+- **Type**: implement
+- **Rationale**: "Make the denial message action-oriented and specific", "Add regression tests around the guard decision matrix". Implementation + test additions to `scripts/headless-guard.mjs`.
+- **Validity verdict**: partially valid
+- **OMX skill**: `$plan` -> `$autopilot`
+
+### 6. prd-issue-29-headless-cwd-propagation.md
+- **Type**: implement
+- **Rationale**: "Add a cwd or equivalent field to the headless start argument parsing", "Propagate that field through start-headless.mjs". Multi-file implementation across CLI parse, headless, backend layers.
+- **Validity verdict**: valid
+- **OMX skill**: `$plan` -> `$autopilot`
+
+### 7. prd-issue-30-remote-spawn-terminal-minimize.md
+- **Type**: investigate
+- **Rationale**: "Reproduce the symptom reliably", "Build a minimal repro script", "Collect terminal state evidence". Explicitly an investigation PRD. Verdict was "evidence insufficient".
+- **Validity verdict**: evidence insufficient
+- **OMX skill**: `$plan` (investigation mode)
+
+---
+
+## Summary Table
+
+| # | PRD file | Type | OMX Skill | Validity |
+|---|----------|------|-----------|----------|
+| 24 | prd-issue-24-remote-spawn-file-transfer | implement | `$plan` -> `$autopilot` | mostly valid |
+| 25 | prd-issue-25-remote-spawn-resize-blank-screen | investigate | `$plan` only | evidence insufficient |
+| 26 | prd-issue-26-remote-spawn-session-cleanup | implement | `$plan` -> `$autopilot` | valid |
+| 27 | prd-issue-27-hud-dashboard-anchor | implement | `$plan` -> `$autopilot` | valid |
+| 28 | prd-issue-28-headless-guard-spawn-deadlock | implement | `$plan` -> `$autopilot` | partially valid |
+| 29 | prd-issue-29-headless-cwd-propagation | implement | `$plan` -> `$autopilot` | valid |
+| 30 | prd-issue-30-remote-spawn-terminal-minimize | investigate | `$plan` only | evidence insufficient |
+
+**Totals**: 5 implement, 2 investigate

package/skills/tfx-codex-swarm-workspace/iteration-1/full-swarm-all-prds/without_skill/outputs/commands.sh

@@ -0,0 +1,264 @@
+#!/usr/bin/env bash
+# =============================================================================
+# Codex Swarm — 7 PRDs from .omx/plans/
+# Generated for: triflux @ C:/Users/SSAFY/Desktop/Projects/cli/triflux
+# 5 implement ($plan -> $autopilot), 2 investigate ($plan only)
+# All sessions in one WT window with tabs
+# =============================================================================
+
+set -euo pipefail
+
+PROJECT_ROOT="C:/Users/SSAFY/Desktop/Projects/cli/triflux"
+SWARM_DIR="$PROJECT_ROOT/.codex-swarm"
+PROMPTS_DIR="$SWARM_DIR/prompts"
+BASH_EXE='C:\Program Files\Git\bin\bash.exe'
+
+# ─── Phase 1: Worktree Creation ─────────────────────────────────────────────
+
+echo "=== Phase 1: Creating worktrees ==="
+
+git -C "$PROJECT_ROOT" worktree add "$SWARM_DIR/wt-issue-24" -b codex/issue-24 2>/dev/null || \
+git -C "$PROJECT_ROOT" worktree add "$SWARM_DIR/wt-issue-24" codex/issue-24
+
+git -C "$PROJECT_ROOT" worktree add "$SWARM_DIR/wt-issue-25" -b codex/issue-25 2>/dev/null || \
+git -C "$PROJECT_ROOT" worktree add "$SWARM_DIR/wt-issue-25" codex/issue-25
+
+git -C "$PROJECT_ROOT" worktree add "$SWARM_DIR/wt-issue-26" -b codex/issue-26 2>/dev/null || \
+git -C "$PROJECT_ROOT" worktree add "$SWARM_DIR/wt-issue-26" codex/issue-26
+
+git -C "$PROJECT_ROOT" worktree add "$SWARM_DIR/wt-issue-27" -b codex/issue-27 2>/dev/null || \
+git -C "$PROJECT_ROOT" worktree add "$SWARM_DIR/wt-issue-27" codex/issue-27
+
+git -C "$PROJECT_ROOT" worktree add "$SWARM_DIR/wt-issue-28" -b codex/issue-28 2>/dev/null || \
+git -C "$PROJECT_ROOT" worktree add "$SWARM_DIR/wt-issue-28" codex/issue-28
+
+git -C "$PROJECT_ROOT" worktree add "$SWARM_DIR/wt-issue-29" -b codex/issue-29 2>/dev/null || \
+git -C "$PROJECT_ROOT" worktree add "$SWARM_DIR/wt-issue-29" codex/issue-29
+
+git -C "$PROJECT_ROOT" worktree add "$SWARM_DIR/wt-issue-30" -b codex/issue-30 2>/dev/null || \
+git -C "$PROJECT_ROOT" worktree add "$SWARM_DIR/wt-issue-30" codex/issue-30
+
+echo "  7 worktrees created under $SWARM_DIR/"
+
+# ─── Phase 2: Copy PRD files into worktrees ──────────────────────────────────
+
+echo "=== Phase 2: Staging PRD files ==="
+
+for N in 24 25 26 27 28 29 30; do
+  mkdir -p "$SWARM_DIR/wt-issue-$N/.omx/plans"
+  cp "$PROJECT_ROOT/.omx/plans/prd-issue-$N-"*.md "$SWARM_DIR/wt-issue-$N/.omx/plans/" 2>/dev/null || true
+done
+
+echo "  PRD files staged into worktrees"
+
+# ─── Phase 3: Generate prompt files ─────────────────────────────────────────
+
+echo "=== Phase 3: Generating prompts ==="
+
+mkdir -p "$PROMPTS_DIR"
+
+# --- Issue #24: implement -> $plan -> $autopilot ---
+cat > "$PROMPTS_DIR/prompt-24.md" << 'PROMPT_EOF'
+triflux 프로젝트의 태스크를 구현해야 합니다.
+
+태스크 파일을 먼저 읽으세요: .omx/plans/prd-issue-24-remote-spawn-file-transfer.md
+
+작업 순서:
+1. 태스크 파일을 읽고 요구사항을 파악하세요
+2. $plan 스킬을 사용하여 구현 계획을 수립하세요
+3. 계획이 완성되면 $autopilot 스킬로 자율 구현을 진행하세요
+4. Acceptance Criteria를 모두 충족하는지 검증하세요
+
+프로젝트 정보:
+- triflux: Claude Code용 멀티모델 CLI 오케스트레이터
+- 언어: JavaScript/ESM (Node.js), 테스트: npm test
+- 주요 수정 대상: scripts/remote-spawn.mjs
+PROMPT_EOF
+
+# --- Issue #25: investigate -> $plan only ---
+cat > "$PROMPTS_DIR/prompt-25.md" << 'PROMPT_EOF'
+triflux 프로젝트의 이슈를 조사해야 합니다.
+
+태스크 파일을 먼저 읽으세요: .omx/plans/prd-issue-25-remote-spawn-resize-blank-screen.md
+
+작업 순서:
+1. 태스크 파일을 읽고 조사 요구사항을 파악하세요
+2. $plan 스킬을 사용하여 조사 계획을 수립하세요
+3. 재현 절차, 계측 방법, 원인 분류 기준을 포함하세요
+4. 구현은 하지 마세요 — 조사 계획 수립까지만 진행하세요
+
+프로젝트 정보:
+- triflux: Claude Code용 멀티모델 CLI 오케스트레이터
+- 언어: JavaScript/ESM (Node.js), 테스트: npm test
+- 조사 대상: SSH PTY resize, psmux pane capture, WT alternate screen
+PROMPT_EOF
+
+# --- Issue #26: implement -> $plan -> $autopilot ---
+cat > "$PROMPTS_DIR/prompt-26.md" << 'PROMPT_EOF'
+triflux 프로젝트의 태스크를 구현해야 합니다.
+
+태스크 파일을 먼저 읽으세요: .omx/plans/prd-issue-26-remote-spawn-session-cleanup.md
+
+작업 순서:
+1. 태스크 파일을 읽고 요구사항을 파악하세요
+2. $plan 스킬을 사용하여 구현 계획을 수립하세요
+3. 계획이 완성되면 $autopilot 스킬로 자율 구현을 진행하세요
+4. Acceptance Criteria를 모두 충족하는지 검증하세요
+
+프로젝트 정보:
+- triflux: Claude Code용 멀티모델 CLI 오케스트레이터
+- 언어: JavaScript/ESM (Node.js), 테스트: npm test
+- 주요 수정 대상: scripts/remote-spawn.mjs, hub/server.mjs
+PROMPT_EOF
+
+# --- Issue #27: implement -> $plan -> $autopilot ---
+cat > "$PROMPTS_DIR/prompt-27.md" << 'PROMPT_EOF'
+triflux 프로젝트의 태스크를 구현해야 합니다.
+
+태스크 파일을 먼저 읽으세요: .omx/plans/prd-issue-27-hud-dashboard-anchor.md
+
+작업 순서:
+1. 태스크 파일을 읽고 요구사항을 파악하세요
+2. $plan 스킬을 사용하여 구현 계획을 수립하세요
+3. 계획이 완성되면 $autopilot 스킬로 자율 구현을 진행하세요
+4. Acceptance Criteria를 모두 충족하는지 검증하세요
+
+프로젝트 정보:
+- triflux: Claude Code용 멀티모델 CLI 오케스트레이터
+- 언어: JavaScript/ESM (Node.js), 테스트: npm test
+- 주요 수정 대상: hub/team/headless.mjs (dashboard attach path)
+PROMPT_EOF
+
+# --- Issue #28: implement -> $plan -> $autopilot ---
+cat > "$PROMPTS_DIR/prompt-28.md" << 'PROMPT_EOF'
+triflux 프로젝트의 태스크를 구현해야 합니다.
+
+태스크 파일을 먼저 읽으세요: .omx/plans/prd-issue-28-headless-guard-spawn-deadlock.md
+
+작업 순서:
+1. 태스크 파일을 읽고 요구사항을 파악하세요
+2. $plan 스킬을 사용하여 구현 계획을 수립하세요
+3. 계획이 완성되면 $autopilot 스킬로 자율 구현을 진행하세요
+4. Acceptance Criteria를 모두 충족하는지 검증하세요
+
+프로젝트 정보:
+- triflux: Claude Code용 멀티모델 CLI 오케스트레이터
+- 언어: JavaScript/ESM (Node.js), 테스트: npm test
+- 주요 수정 대상: scripts/headless-guard.mjs, tests/unit/headless-guard.test.mjs
+PROMPT_EOF
+
+# --- Issue #29: implement -> $plan -> $autopilot ---
+cat > "$PROMPTS_DIR/prompt-29.md" << 'PROMPT_EOF'
+triflux 프로젝트의 태스크를 구현해야 합니다.
+
+태스크 파일을 먼저 읽으세요: .omx/plans/prd-issue-29-headless-cwd-propagation.md
+
+작업 순서:
+1. 태스크 파일을 읽고 요구사항을 파악하세요
+2. $plan 스킬을 사용하여 구현 계획을 수립하세요
+3. 계획이 완성되면 $autopilot 스킬로 자율 구현을 진행하세요
+4. Acceptance Criteria를 모두 충족하는지 검증하세요
+
+프로젝트 정보:
+- triflux: Claude Code용 멀티모델 CLI 오케스트레이터
+- 언어: JavaScript/ESM (Node.js), 테스트: npm test
+- 주요 수정 대상: hub/team/cli/commands/start/parse-args.mjs, start-headless.mjs, hub/team/headless.mjs, hub/team/backend.mjs
+PROMPT_EOF
+
+# --- Issue #30: investigate -> $plan only ---
+cat > "$PROMPTS_DIR/prompt-30.md" << 'PROMPT_EOF'
+triflux 프로젝트의 이슈를 조사해야 합니다.
+
+태스크 파일을 먼저 읽으세요: .omx/plans/prd-issue-30-remote-spawn-terminal-minimize.md
+
+작업 순서:
+1. 태스크 파일을 읽고 조사 요구사항을 파악하세요
+2. $plan 스킬을 사용하여 조사 계획을 수립하세요
+3. 재현 스크립트, 터미널 상태 캡처 방법, 판별 기준을 포함하세요
+4. 구현은 하지 마세요 — 조사 계획 수립까지만 진행하세요
+
+프로젝트 정보:
+- triflux: Claude Code용 멀티모델 CLI 오케스트레이터
+- 언어: JavaScript/ESM (Node.js), 테스트: npm test
+- 조사 대상: remote-spawn.mjs WT tab/pane creation, headless.mjs focus manipulation
+PROMPT_EOF
+
+echo "  7 prompt files generated in $PROMPTS_DIR/"
+
+# ─── Phase 4: psmux Session Creation + Codex Launch ─────────────────────────
+
+echo "=== Phase 4: Creating psmux sessions and launching Codex ==="
+
+# --- Issue #24: implement, codex53_high ---
+psmux new-session --name "codex-swarm-24" --dir "$SWARM_DIR/wt-issue-24"
+psmux send-keys --target "codex-swarm-24:0" \
+  "codex -c 'model=\"gpt-5.3-codex\"' -c 'model_reasoning_effort=\"high\"' --full-auto \"\$(cat $PROMPTS_DIR/prompt-24.md)\"" Enter
+
+# --- Issue #25: investigate, gpt54_high ---
+psmux new-session --name "codex-swarm-25" --dir "$SWARM_DIR/wt-issue-25"
+psmux send-keys --target "codex-swarm-25:0" \
+  "codex -c 'model=\"gpt-5.4\"' -c 'model_reasoning_effort=\"high\"' --full-auto \"\$(cat $PROMPTS_DIR/prompt-25.md)\"" Enter
+
+# --- Issue #26: implement, codex53_high ---
+psmux new-session --name "codex-swarm-26" --dir "$SWARM_DIR/wt-issue-26"
+psmux send-keys --target "codex-swarm-26:0" \
+  "codex -c 'model=\"gpt-5.3-codex\"' -c 'model_reasoning_effort=\"high\"' --full-auto \"\$(cat $PROMPTS_DIR/prompt-26.md)\"" Enter
+
+# --- Issue #27: implement, codex53_high ---
+psmux new-session --name "codex-swarm-27" --dir "$SWARM_DIR/wt-issue-27"
+psmux send-keys --target "codex-swarm-27:0" \
+  "codex -c 'model=\"gpt-5.3-codex\"' -c 'model_reasoning_effort=\"high\"' --full-auto \"\$(cat $PROMPTS_DIR/prompt-27.md)\"" Enter
+
+# --- Issue #28: implement, codex53_high ---
+psmux new-session --name "codex-swarm-28" --dir "$SWARM_DIR/wt-issue-28"
+psmux send-keys --target "codex-swarm-28:0" \
+  "codex -c 'model=\"gpt-5.3-codex\"' -c 'model_reasoning_effort=\"high\"' --full-auto \"\$(cat $PROMPTS_DIR/prompt-28.md)\"" Enter
+
+# --- Issue #29: implement, codex53_high ---
+psmux new-session --name "codex-swarm-29" --dir "$SWARM_DIR/wt-issue-29"
+psmux send-keys --target "codex-swarm-29:0" \
+  "codex -c 'model=\"gpt-5.3-codex\"' -c 'model_reasoning_effort=\"high\"' --full-auto \"\$(cat $PROMPTS_DIR/prompt-29.md)\"" Enter
+
+# --- Issue #30: investigate, gpt54_high ---
+psmux new-session --name "codex-swarm-30" --dir "$SWARM_DIR/wt-issue-30"
+psmux send-keys --target "codex-swarm-30:0" \
+  "codex -c 'model=\"gpt-5.4\"' -c 'model_reasoning_effort=\"high\"' --full-auto \"\$(cat $PROMPTS_DIR/prompt-30.md)\"" Enter
+
+echo "  7 psmux sessions created, Codex launched in each"
+
+# ─── Phase 5: WT Tab Attach — All sessions in one window ────────────────────
+
+echo "=== Phase 5: Attaching all sessions to one WT window ==="
+
+# First session opens a new WT window
+psmux attach --session "codex-swarm-24" --wt-new-window
+
+# Remaining 6 sessions attach as tabs in the same window
+psmux attach --session "codex-swarm-25" --wt-tab
+psmux attach --session "codex-swarm-26" --wt-tab
+psmux attach --session "codex-swarm-27" --wt-tab
+psmux attach --session "codex-swarm-28" --wt-tab
+psmux attach --session "codex-swarm-29" --wt-tab
+psmux attach --session "codex-swarm-30" --wt-tab
+
+echo "  All 7 sessions attached as tabs in one WT window"
+
+# ─── Fallback: wt.exe direct (if psmux attach --wt-tab unavailable) ─────────
+
+# Uncomment below if psmux attach --wt-tab is not available:
+#
+# PSMUX_WIN=$(command -v psmux.exe || echo "psmux")
+# wt.exe -w new \
+#   --title "swarm-24 file-transfer" -d "$SWARM_DIR/wt-issue-24" "$BASH_EXE" -c "psmux attach codex-swarm-24" \; \
+#   new-tab --title "swarm-25 resize-blank" -d "$SWARM_DIR/wt-issue-25" "$BASH_EXE" -c "psmux attach codex-swarm-25" \; \
+#   new-tab --title "swarm-26 session-cleanup" -d "$SWARM_DIR/wt-issue-26" "$BASH_EXE" -c "psmux attach codex-swarm-26" \; \
+#   new-tab --title "swarm-27 dashboard-anchor" -d "$SWARM_DIR/wt-issue-27" "$BASH_EXE" -c "psmux attach codex-swarm-27" \; \
+#   new-tab --title "swarm-28 guard-deadlock" -d "$SWARM_DIR/wt-issue-28" "$BASH_EXE" -c "psmux attach codex-swarm-28" \; \
+#   new-tab --title "swarm-29 cwd-propagation" -d "$SWARM_DIR/wt-issue-29" "$BASH_EXE" -c "psmux attach codex-swarm-29" \; \
+#   new-tab --title "swarm-30 terminal-minimize" -d "$SWARM_DIR/wt-issue-30" "$BASH_EXE" -c "psmux attach codex-swarm-30"
+
+echo ""
+echo "=== Codex Swarm Complete ==="
+echo "  Sessions: 7 (5 implement + 2 investigate)"
+echo "  Window:   1 WT window, 7 tabs"
+echo "  Monitor:  psmux capture-pane --session codex-swarm-{N} --lines 5"

package/skills/tfx-codex-swarm-workspace/iteration-1/full-swarm-all-prds/without_skill/outputs/routing.md

@@ -0,0 +1,113 @@
+# Codex Profile Routing — 7 PRDs
+
+## Routing Method
+
+Following SKILL.md Step 4 profile routing tables.
+
+### Size Estimation Signals
+
+| Signal | Method |
+|--------|--------|
+| PRD length | line count |
+| Affected files | file paths mentioned in PRD |
+| High-cost keywords | "architecture", "migration", "refactoring" |
+| Dependency depth | references to other issues/PRDs |
+
+### Size Thresholds
+
+| Size | PRD lines | Affected files | High-cost keywords |
+|------|-----------|----------------|--------------------|
+| XL | 80+ | 6+ | 2+ |
+| L | 40-80 | 3-5 | 0-1 |
+| M | 20-40 | 1-2 | 0 |
+| S | <20 | 1 | 0 |
+
+---
+
+## Per-PRD Routing
+
+### Issue #24 — remote-spawn file transfer
+- **Lines**: 73
+- **Affected files**: `scripts/remote-spawn.mjs` (primary), prompt pipeline, scp path (~3 files)
+- **High-cost keywords**: 0
+- **Size**: L (standard)
+- **Type**: implement
+- **Profile**: `codex53_high`
+- **Flags**: `-c 'model="gpt-5.3-codex"' -c 'model_reasoning_effort="high"'`
+
+### Issue #25 — resize blank-screen investigation
+- **Lines**: 76
+- **Affected files**: `scripts/remote-spawn.mjs`, psmux pane capture, SSH PTY (~4 files)
+- **High-cost keywords**: 0
+- **Size**: L (standard)
+- **Type**: investigate
+- **Profile**: `gpt54_high`
+- **Flags**: `-c 'model="gpt-5.4"' -c 'model_reasoning_effort="high"'`
+
+### Issue #26 — session cleanup
+- **Lines**: 70
+- **Affected files**: `scripts/remote-spawn.mjs`, `hub/server.mjs` (~2-3 files)
+- **High-cost keywords**: 0
+- **Size**: L (standard)
+- **Type**: implement
+- **Profile**: `codex53_high`
+- **Flags**: `-c 'model="gpt-5.3-codex"' -c 'model_reasoning_effort="high"'`
+
+### Issue #27 — HUD dashboard anchor
+- **Lines**: 73
+- **Affected files**: `hub/team/headless.mjs` (primary), WT config (~2 files)
+- **High-cost keywords**: 0
+- **Size**: L (standard)
+- **Type**: implement
+- **Profile**: `codex53_high`
+- **Flags**: `-c 'model="gpt-5.3-codex"' -c 'model_reasoning_effort="high"'`
+
+### Issue #28 — headless-guard spawn deadlock
+- **Lines**: 69
+- **Affected files**: `scripts/headless-guard.mjs`, `tests/unit/headless-guard.test.mjs` (~2 files)
+- **High-cost keywords**: 0
+- **Size**: L (standard)
+- **Type**: implement
+- **Profile**: `codex53_high`
+- **Flags**: `-c 'model="gpt-5.3-codex"' -c 'model_reasoning_effort="high"'`
+
+### Issue #29 — headless cwd propagation
+- **Lines**: 70
+- **Affected files**: `hub/team/cli/commands/start/parse-args.mjs`, `hub/team/cli/commands/start/start-headless.mjs`, `hub/team/headless.mjs`, `hub/team/backend.mjs`, `scripts/tfx-route.sh` (~5 files)
+- **High-cost keywords**: 0
+- **Size**: L (standard)
+- **Type**: implement
+- **Profile**: `codex53_high`
+- **Flags**: `-c 'model="gpt-5.3-codex"' -c 'model_reasoning_effort="high"'`
+
+### Issue #30 — terminal minimize investigation
+- **Lines**: 70
+- **Affected files**: `scripts/remote-spawn.mjs`, `hub/team/headless.mjs`, WT focus (~3 files)
+- **High-cost keywords**: 0
+- **Size**: L (standard)
+- **Type**: investigate
+- **Profile**: `gpt54_high`
+- **Flags**: `-c 'model="gpt-5.4"' -c 'model_reasoning_effort="high"'`
+
+---
+
+## Routing Summary Table
+
+| # | Task | Size | Type | Profile | Model | Effort |
+|---|------|------|------|---------|-------|--------|
+| 24 | file-transfer | L | implement | `codex53_high` | gpt-5.3-codex | high |
+| 25 | resize-blank | L | investigate | `gpt54_high` | gpt-5.4 | high |
+| 26 | session-cleanup | L | implement | `codex53_high` | gpt-5.3-codex | high |
+| 27 | dashboard-anchor | L | implement | `codex53_high` | gpt-5.3-codex | high |
+| 28 | guard-deadlock | L | implement | `codex53_high` | gpt-5.3-codex | high |
+| 29 | cwd-propagation | L | implement | `codex53_high` | gpt-5.3-codex | high |
+| 30 | terminal-minimize | L | investigate | `gpt54_high` | gpt-5.4 | high |
+
+## Execution Config
+
+| Setting | Value | Notes |
+|---------|-------|-------|
+| Execution mode | `--full-auto` | sandbox auto-approve |
+| Worktree | enabled | per-session isolated worktree |
+| psmux | enabled | session management |
+| WT layout | single window, 7 tabs | user requested tabs |

package/skills/tfx-codex-swarm-workspace/iteration-1/full-swarm-all-prds/without_skill/timing.json

@@ -0,0 +1,5 @@
+{
+  "total_tokens": 50555,
+  "duration_ms": 218213,
+  "total_duration_seconds": 218.2
+}

package/skills/tfx-codex-swarm-workspace/iteration-1/implicit-swarm-no-keywords/eval_metadata.json

@@ -0,0 +1,32 @@
+{
+  "eval_id": 3,
+  "eval_name": "implicit-swarm-no-keywords",
+  "prompt": "이 PRD 3개 파일을 각각 독립적으로 코덱스한테 맡겨서 병렬로 구현해줘. docs/prd/auth-refactor.md, docs/prd/api-v2.md, docs/prd/cache-layer.md",
+  "assertions": [
+    {
+      "name": "3-paths-recognized",
+      "description": "사용자 제공 3개 파일 경로 인식",
+      "check": "classification.md에 auth-refactor, api-v2, cache-layer 모두 등장"
+    },
+    {
+      "name": "filename-classification",
+      "description": "파일명 기반 유형 분류 시도",
+      "check": "auth-refactor를 리팩터링으로, api-v2/cache-layer를 구현으로 분류"
+    },
+    {
+      "name": "profile-routing-applied",
+      "description": "자동 프로파일 라우팅 적용",
+      "check": "routing.md에 프로파일 매핑 존재 (codex53 또는 gpt54)"
+    },
+    {
+      "name": "worktree-isolation",
+      "description": "3개 독립 worktree 생성",
+      "check": "commands.sh에 git worktree add가 3개"
+    },
+    {
+      "name": "psmux-and-wt",
+      "description": "psmux 세션 + WT 탭 명령 생성",
+      "check": "commands.sh에 psmux + wt 명령 존재"
+    }
+  ]
+}

package/skills/tfx-codex-swarm-workspace/iteration-1/implicit-swarm-no-keywords/with_skill/grading.json

@@ -0,0 +1,9 @@
+{
+  "expectations": [
+    {"text": "3-paths-recognized", "passed": true, "evidence": "auth-refactor, api-v2, cache-layer 모두 classification.md에 등장"},
+    {"text": "filename-classification", "passed": true, "evidence": "auth-refactor=리팩터링, api-v2=구현, cache-layer=구현"},
+    {"text": "profile-routing-applied", "passed": true, "evidence": "모두 codex53_high 라우팅"},
+    {"text": "worktree-isolation", "passed": true, "evidence": "3개 git worktree add 명령"},
+    {"text": "psmux-and-wt", "passed": true, "evidence": "psmux 세션 + WT 탭 명령 존재"}
+  ]
+}