Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →OpenAI Codex CLI + Claude Code (Hizir) birlikte kullanim rehberi. Is dagitim pattern'leri, GitHub Actions workflow ornekleri, review dongusu ve iki AI yazilim asistaninin guclu yanlarini birlestiren orchestration stratejileri.
.claude/skills/codex-orchestration/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 64% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 33% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 41% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 56% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 66% | 0% |
| Yetenek | Codex CLI | Claude Code (Hizir) | Kazanan | |---------|-----------|---------------------|---------| | Hiz | gpt-5.6-luna / max | Claude adapter modeline bagli | Codex | | Maliyet | Worker modeline ve kullanimina bagli | Claude pricing | Adapter'a bagli | | Context window | Sinirli | 1M token | Claude Code | | Multi-agent | YOK (tek agent) | 134+ agent swarm | Claude Code | | Hook/self-learning | YOK | Tam destek (74 hook) | Claude Code | | Memory/state | Stateless (her cagri bagimsiz) | Persistent memory | Claude Code | | Code review | Basit lint | Derinlemesine + security | Claude Code | | Bulk refactoring | Cok iyi (hiz+ucuz) | Iyi ama pahali | Codex | | Mimari karar | Zayif | Guclu (architect agent) | Claude Code | | Test yazma | Iyi | Iyi + TDD workflow | Esit | | Security audit | Basit | 3-katman (SAST + review + manual) | Claude Code | | CI/CD entegrasyonu | GitHub native | GitHub Actions + webhook | Esit |
En yaygin ve etkili pattern.
1. Claude Code → plan.md olusturur (architect agent)
2. Codex → plan.md'ye gore implement eder (full-auto mode)
3. Codex → PR acar
4. Claude Code → PR review eder (code-reviewer + security-reviewer)
5. Codex → Review bulgularini fix eder
6. Claude Code → Final verify + merge onayGitHub Actions Workflow:
yaml# .github/workflows/codex-implement-claude-review.yml name: Codex + Claude Code Pipeline on: issues: types: [labeled] jobs: codex-implement: if: contains(github.event.label.name, 'codex-task') runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 - name: Codex Implementation env: OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }} run: | npx codex --approval-mode full-auto \ --quiet \ -q "Implement the task described in issue #${{ github.event.issue.number }}. Follow the plan if provided." - name: Create PR env: GH_TOKEN: ${{ secrets.GITHUB_TOKEN }} run: | git checkout -b codex/issue-${{ github.event.issue.number }} git add -A git commit -m "feat: implement issue #${{ github.event.issue.number }} [codex]" git push -u origin codex/issue-${{ github.event.issue.number }} gh pr create --title "Codex: Issue #${{ github.event.issue.number }}" \ --body "Automated implementation by Codex CLI. Awaiting Claude Code review." claude-review: needs: codex-implement runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 - name: Claude Code Review env: ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }} run: | # claude-review workflow zaten mevcut claude -p "Review the PR changes. Run code-reviewer + security-reviewer. Report findings."
yaml# .github/workflows/dual-review.yml name: Dual AI Review on: pull_request: types: [opened, synchronize] jobs: codex-quick-review: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 - name: Codex Quick Lint Review run: | npx codex --approval-mode full-auto -q \ "Review the diff for obvious issues: lint errors, typos, missing imports, type errors. Be brief." claude-deep-review: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 - name: Claude Code Deep Review run: | claude -p "Deep review: architecture, security, performance, edge cases. Use code-reviewer + security-reviewer agents."
Birden fazla kuuk task'i Codex'e dagit, Claude Code koordine etsin.
bash#!/bin/bash # batch-codex-tasks.sh TASKS=( "Add input validation to all API endpoints in src/api/" "Convert all var declarations to const/let in src/utils/" "Add JSDoc comments to all exported functions in src/lib/" "Fix all TypeScript strict mode errors in src/models/" ) for i in "${!TASKS[@]}"; do echo "Task $((i+1)): ${TASKS[$i]}" npx codex --approval-mode full-auto --quiet -q "${TASKS[$i]}" & done wait echo "All Codex tasks complete. Running Claude Code verification..." claude -p "Verify all changes: build, test, lint. Report issues."
PHASE 1: PLAN (Claude Code)
├── architect agent ile plan olustur
├── Task'lari tanimla
├── Kabul kriterleri belirle
└── plan.md veya GitHub issue olustur
PHASE 2: IMPLEMENT (Codex)
├── codex --approval-mode full-auto
├── Her task icin ayri branch
├── PR ac
└── Codex kendi testlerini de yazabilir
PHASE 3: REVIEW (Claude Code)
├── code-reviewer: Kalite, pattern, best practice
├── security-reviewer: Guvenlik aciklari
├── verifier: Build + test + lint
└── VERDICT: PASS / FAIL + feedback
PHASE 4: FIX (Codex)
├── Claude Code'un feedback'ini al
├── Sadece belirtilen sorunlari duzelt
├── Yeni feature EKLEME
└── Tekrar PR guncelle
PHASE 5: MERGE (Claude Code)
├── Final verify
├── Tum testler geciyor mu?
├── Security temiz mi?
└── Merge onay + deploybash# Interaktif mod (terminal'de calisir) codex # Tek seferlik task (non-interactive) codex -q "task aciklamasi" # Full auto (onay istemeden yapar) codex --approval-mode full-auto -q "task" # Model secimi luna_worker sozlesmesinden gelir; burada ikinci bir router yok codex -q "task" # Quiet mode (CI/CD icin) codex --quiet -q "task"
bash# GitHub Actions icinde CODEX_QUIET=1 npx codex --approval-mode full-auto -q "$TASK" # Cikti kontrolu if [ $? -eq 0 ]; then echo "Codex task basarili" else echo "Codex task basarisiz, Claude Code'a devret" claude -p "Fix the failed Codex task: $TASK" fi
| Senaryo | Sadece Claude Code | Sadece Codex | Hybrid | |---------|-------------------|-------------|--------| | 10 dosya refactoring | ~$2-5 | ~$0.30-0.50 | ~$0.80 | | Security review | ~$1-3 (derinlemesine) | ~$0.20 (yuzeysel) | ~$1.20 | | Test yazma (20 test) | ~$3-5 | ~$0.50-1.00 | ~$1.50 | | Mimari planlama | ~$2-4 (detayli) | ~$0.30 (yuzeysel) | ~$2.30 | | Toplam | ~$8-17 | ~$1.30-2.00 | ~$5.80 |
Hybrid strateji %40-60 maliyet tasarrufu saglar.
Formul: Codex'i "volume work" icin, Claude Code'u "judgment work" icin kullan.
| Ozellik | Durum | Not | |---------|-------|-----| | Multi-agent | YOK | Tek agent, swarm mumkun degil | | Hook sistemi | YOK | PreToolUse/PostToolUse yok | | Self-learning | YOK | Hatalardan ders cikarma yok | | Persistent memory | YOK | Her cagri bagimsiz | | Agent personas | YOK | Tek persona | | Skill auto-discovery | KISMI | SKILL.md okuyabilir ama agent.md'leri kullanamaz | | Context window | SINIRLI | 1M degil, daha kucuk | | Tool calling | SINIRLI | Bash + file read/write |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | fail→pass | 14,621 | 9,821 | -33% | 1 | 1 | 0% | 2,496 | 4,087 | +64% | 0 | 0 | — |
case-05 | fail→fail | 26,425 | 13,827 | -48% | 1 | 1 | 0% | 4,155 | 5,405 | +30% | 0 | 0 | — |
case-03 | fail→pass | 35,146 | 30,982 | -12% | 1 | 1 | 0% | 5,746 | 7,655 | +33% | 0 | 0 | — |
case-01 | fail→pass | 17,583 | 12,823 | -27% | 1 | 1 | 0% | 3,580 | 5,048 | +41% | 0 | 0 | — |
case-02 | fail→pass | 22,698 | 19,976 | -12% | 1 | 1 | 0% | 4,251 | 6,647 | +56% | 0 | 0 | — |
case-06 | fail→pass | 11,932 | 6,277 | -47% | 1 | 1 | 0% | 2,113 | 3,505 | +66% | 0 | 0 | — |
case-07 | pass→pass | 8,321 | 9,642 | +16% | 1 | 1 | 0% | 1,190 | 3,897 | +227% | 0 | 0 | — |
case-08 | fail→fail | 19,311 | 11,723 | -39% | 1 | 1 | 0% | 2,849 | 4,454 | +56% | 0 | 0 | — |
case-09 | fail→pass | 17,598 | 13,807 | -22% | 1 | 1 | 0% | 2,813 | 4,982 | +77% | 0 | 0 | — |
case-10 | pass→pass | 10,525 | 10,325 | -2% | 1 | 1 | 0% | 1,874 | 4,156 | +122% | 0 | 0 | — |
case-11 | pass→fail | 10,839 | 8,291 | -24% | 1 | 1 | 0% | 1,920 | 3,669 | +91% | 0 | 0 | — |
case-12 | pass→pass | 11,901 | 13,305 | +12% | 1 | 1 | 0% | 2,203 | 5,133 | +133% | 0 | 0 | — |
case-13 | fail→pass | 9,714 | 10,520 | +8% | 1 | 1 | 0% | 1,526 | 4,064 | +166% | 0 | 0 | — |
case-14 | fail→pass | 14,407 | 11,029 | -23% | 1 | 1 | 0% | 2,184 | 4,190 | +92% | 0 | 0 | — |
case-15 | fail→fail | 34,660 | 10,280 | -70% | 1 | 1 | 0% | 1,153 | 3,974 | +245% | 0 | 0 | — |
case-16 | fail→pass | 9,068 | 4,151 | -54% | 1 | 1 | 0% | 1,500 | 3,024 | +102% | 0 | 0 | — |
case-17 | fail→pass | 11,669 | 3,027 | -74% | 1 | 1 | 0% | 2,015 | 2,889 | +43% | 0 | 0 | — |
case-18 | pass→pass | 10,087 | 16,685 | +65% | 1 | 1 | 0% | 1,732 | 3,813 | +120% | 0 | 0 | — |
case-19 | fail→fail | 19,568 | 16,051 | -18% | 1 | 1 | 0% | 2,848 | 4,909 | +72% | 0 | 0 | — |
case-20 | fail→pass | 8,425 | 2,577 | -69% | 1 | 1 | 0% | 1,282 | 2,733 | +113% | 0 | 0 | — |
case-21 | fail→pass | 28,548 | 11,759 | -59% | 1 | 1 | 0% | 2,722 | 4,157 | +53% | 0 | 0 | — |
case-22 | fail→pass | 10,839 | 4,087 | -62% | 1 | 1 | 0% | 1,774 | 2,953 | +66% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +55 percentage points is the difference between those two pass rates over the 21 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 7/29/2026 | +64% |
Other measured skills in the registry, with their headline benchmark lift.