workflow-guide

A two-agent workflow between Cursor and Claude Code is coordinated through structured role separation where Cursor acts as PM and Claude Code as Worker, using a shared Plans.md file with task markers and commands like /plan-with-agent and /handoff-to-cursor to manage planning, execution, and reviews.

Managing task handoffs in a CI/CD pipeline with multiple AI agents is best handled through structured Plans.md task markers like pm:依頼中, cc:WIP, and cc:完了, which track task progression and enforce guardrails during the review and staging phases of the workflow.

The /work command executes tasks in a project management workflow by triggering Claude Code to begin work on items defined in Plans.md, transitioning task markers to WIP status, and following established guardrails for testing and reviews before handing off back to Cursor.

Yes, you can use Plans.md task markers for parallel work across different agents; the workflow supports parallel task execution by tracking individual task states independently, allowing Cursor and Claude Code to manage multiple WIP items simultaneously while maintaining clear role separation.

To start a two-agent workflow, you need to create a Plans.md file with your initial tasks, then run the /plan-with-agent command to break them down before executing /work; no specific broader project structure is required beyond this file.

You should not use a two-agent workflow for project management if your tasks require no planning, execution, testing, or review phases, or if a single agent can handle the entire lifecycle without needing structured handoffs, role separation, or CI/CD guardrails.