dual-orchestrator — Claude Code subagent: routing and setup
dual-orchestrator routes work between interactive Claude Code and headless Codex: use one for reasoning and review, the other for parallel execution. Built for workflows that combine both.
no license ★ 850 updated 2026-10-26
.claude/agents/dual-orchestrator.mdWhat dual-orchestrator does
The subagent chooses a platform based on the task. Requests about architecture, debugging, and code review go to Claude Code; implementation, tests, documentation, and batch jobs go to Codex. Routing relies on wording patterns, so the prompt affects the choice.
A hybrid workflow can start with an architecture discussion in Claude Code, move into parallel implementation with Codex, then return to Claude Code for review. The setup also describes swarm coordination and shared memory for design decisions and worker results.
The examples launch background work with claude -p and retrieve results with npx claude-flow@v3alpha memory list --namespace results. Parallel Codex workflows therefore depend on those commands and the claude-flow integration; check them before using that flow.
What it helps with
- split design and implementation across platforms
- delegate several tasks for parallel execution
- route debugging and code review to an interactive session
- gather background task results for review
Example prompts for the agent
- Plan the architecture for this feature, then assign the implementation to Codex as parallel tasks.
- Investigate this bug in the existing code and explain where it originates.
- Generate tests and documentation for this feature in background tasks, then bring the results back for review.
How to install dual-orchestrator
Manually: put the files in place
.claude/agents/dual-orchestrator.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/dual-orchestrator.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.cursor/agents/dual-orchestrator.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about dual-orchestrator
How do I install dual-orchestrator in Claude Code?
Add the subagent file at .claude/agents/dual-orchestrator.md.
What does dual-orchestrator do?
It routes work between interactive Claude Code and headless Codex. Reasoning and review go to Claude Code, while parallel execution goes to Codex.
How does dual-orchestrator share design decisions with workers?
The workflow stores decisions in shared memory so workers can read them. Task results can be saved in the results namespace.