workflow-guide

The 2-agent workflow uses Plans.md as a central hub where the PM (Cursor) creates and assigns tasks, and the Developer (Claude Code) executes them and reports status. Tasks flow through defined states (pm:requested → cc:WIP → cc:complete → pm:confirmed) using commands like /work, /handoff-to-cursor, and /sync-status to keep both agents synchronized.

Use /harness-init to set up the workflow, /plan-with-agent to create tasks with the PM, /work to execute as the Developer, /handoff-to-cursor to return completed work, and /sync-status to check task states. Each command marks explicit role responsibilities and progresses tasks through the defined state markers.

Yes. The workflow includes CI/CD rules that limit production deployment and define task state markers (blocked, cursor:requested, cursor:confirmed) to enforce staged governance. Both PM and Developer roles have explicit responsibilities within the end-to-end handoff loop.

Start by saying 'explain the 2-agent workflow' to get a complete overview of roles, responsibilities, and the standard collaboration loop. The Skill provides diagrams, command reference examples, and typical task lifecycle walkthroughs so new users understand how Plans.md, /work execution, and handoffs work together.

No. The Skill removes the learning curve by providing instant, contextual explanations of roles, commands, and best practices. It covers everything from basic collaboration setup through staged CI/CD governance, making it accessible regardless of prior experience with either tool.

Blocked tasks are tracked as a state marker within the defined task lifecycle. The workflow coordinates PM and Developer roles to diagnose and resolve blocks; /sync-status helps both agents track blocked states and coordinate next steps before resuming execution.