Company Manager Skill

Multi-agent project execution is coordinated through file-based state tracking, separating Manager planning and approval from Worker execution to manage dependencies, concurrency layers, and delivery handoffs without hidden state.

File-based state tracking uses board states, task cards, employee logs, and manager inbox files in shared project folders to monitor task claims, delivery handoffs, and progress updates for collaborative AI work.

Task concurrency layers are managed by breaking project goals into dependency-aware task cards, allowing multiple workers to claim ready tasks in parallel while the Manager controls final acceptance workflows.

You do not need external databases; workflow coordination is handled entirely using shared markdown and JSON files for task cards, board states, and employee logs to track progress and approvals.

To set up roles, instruct the model to read the company-kit SKILL.md file, assign either the Manager or Worker role, and follow the listed startup protocol to initialize the project workflow.

You should avoid file-driven workflow coordination if your project requires real-time concurrent editing with strict transactional consistency rather than asynchronous task claiming and markdown-based approval gates.