Dual-Agent Orchestration
Use this skill when the user wants Codex and Claude to brainstorm, monitor each other, divide work, review each other's output, or coordinate across subagents.
Default Roles
- Codex owns local execution: reading the repo, editing files, running tests, browser QA, git hygiene, and final verification.
- Claude owns long-form thinking: product/design strategy, architecture tradeoffs, UX critique, copy, broad review, and recruiting workflow strategy.
- Flip the roles only when the user explicitly starts the implementation in Claude Code or asks Claude to own a bounded task.
Handoff Flow
- Read
handoffs/README.md.
- Create a dated handoff file when the other agent needs to review or continue work.
- Include goal, context, touched files, out-of-scope files, commands run, and exact questions.
- Ask the receiving agent for ranked findings or a bounded implementation plan.
- Apply only chosen changes, then verify locally.
Claude Subagents
Use these Claude agents when available:
design-critic: visual/UI/product review.architecture-critic: system design and refactor review.recruiting-ops-reviewer: application tracker, outreach, Gmail, and resume workflow review.safety-gate: pre-send, pre-submit, pre-push, and destructive-operation review.handoff-router: triage open handoffs and select the next owner.
Output
Keep summaries operational:
- who owns the next step
- what files are in scope
- what risks were found
- what verification is required
- what needs human approval
1---2name: dual-agent-orchestration3description: Coordinate Codex and Claude Code as an executor/reviewer pair using local handoff files, Claude subagents, and shared recruiting workflow skills without letting both agents edit the same files at once.4---56# Dual-Agent Orchestration78Use this skill when the user wants Codex and Claude to brainstorm, monitor each other, divide work, review each other's output, or coordinate across subagents.910## Default Roles1112- Codex owns local execution: reading the repo, editing files, running tests, browser QA, git hygiene, and final verification.13- Claude owns long-form thinking: product/design strategy, architecture tradeoffs, UX critique, copy, broad review, and recruiting workflow strategy.14- Flip the roles only when the user explicitly starts the implementation in Claude Code or asks Claude to own a bounded task.1516## Handoff Flow17181. Read `handoffs/README.md`.192. Create a dated handoff file when the other agent needs to review or continue work.203. Include goal, context, touched files, out-of-scope files, commands run, and exact questions.214. Ask the receiving agent for ranked findings or a bounded implementation plan.225. Apply only chosen changes, then verify locally.2324## Claude Subagents2526Use these Claude agents when available:2728- `design-critic`: visual/UI/product review.29- `architecture-critic`: system design and refactor review.30- `recruiting-ops-reviewer`: application tracker, outreach, Gmail, and resume workflow review.31- `safety-gate`: pre-send, pre-submit, pre-push, and destructive-operation review.32- `handoff-router`: triage open handoffs and select the next owner.3334## Output3536Keep summaries operational:3738- who owns the next step39- what files are in scope40- what risks were found41- what verification is required42- what needs human approval
SpeciLiam