Orchestrate parallel Claude subagents from Codex using background terminal processes
Invert the traditional AI orchestration paradigm: Codex conducts, Claude agents perform.
This project enables Codex to spawn and coordinate multiple Claude subagents via background terminal processes, creating a cross-model collaboration pattern for parallel AI-assisted development.
Codex (Orchestrator)
│
├─→ Background Terminal 1 ──→ Claude Subagent (Research)
│
├─→ Background Terminal 2 ──→ Claude Subagent (Implementation)
│
├─→ Background Terminal 3 ──→ Claude Subagent (Testing)
│
└─→ Result Aggregation ←──── stdout/files from all terminals
- Cross-model synergy - Combine Codex's strengths with Claude's capabilities
- Parallel execution - Multiple Claude instances working concurrently
- Unified orchestration - Single Codex session managing the swarm
- No context fragmentation - Coordinated multi-agent assistance
Phase: Research & Discovery
The GPT-5 Pro / Claw Dev team is investigating the optimal integration approach. See the Product Research Proposal for details.
| # | Question | Status |
|---|---|---|
| Q1 | Architecture Pattern Selection | Investigating |
| Q2 | Background Terminal Mechanics | Investigating |
| Q3 | Context Synchronization | Investigating |
| Q4 | Task Distribution Patterns | Investigating |
| Q5 | Claude Code Integration | Investigating |
| Q6 | Security & Sandboxing | Investigating |
| Q7 | Developer Experience | Investigating |
codex-subagents/
├── _bmad/ # BMAD Framework (methodology)
├── _bmad-output/ # Project artifacts
│ └── prp-codex-subagents-integration.md # Research proposal
├── .github/
│ └── PULL_REQUEST_TEMPLATE.md
└── README.md
Coming after research phase completes
- Review the PRP for context
- Pick a research question (Q1-Q7) to investigate
- Submit findings via PR using our PR template
- Follow BMAD framework conventions
- Orchestrator: OpenAI Codex
- Subagents: Claude (via Claude Code CLI)
- Coordination: Background terminal processes
- Framework: BMAD (BMad Method)
TBD
Keywords: codex, claude, anthropic, openai, multi-agent, orchestration, background-terminal, parallel-execution, swarm, context-synchronization, task-distribution, cross-model, ai-agents, developer-tools, subagents