agent-company-claude-skill

This Skill addresses the lack of division of labor and persistence in multi-agent collaboration. In practical development, AI agents often only execute tasks in a single session and cannot collaborate with specialized roles or accumulate context like a real team. As project requirements grow, previous work is easily lost, leading to redundant effort and reduced efficiency.

Usage is simple: just tell the AI "/agent-company your goal", and it will automatically assume the Leader role, break down your goal into DAG-based tasks, and assign them to corresponding domain-specific Workers. Workers execute independently, and all state is persistently saved in the .mycompany/ directory, allowing progress to resume even if the session is interrupted and restarted. You can later use "/agent-company resume" to continue, or commands like "status" and "review" to check progress.

It is well-suited for teams or individuals who need to coordinate multiple agents for complex development, especially developers already using Claude for coding but who want to assign tasks and track progress like managing a real team. Project managers, agent developers, and full-stack engineers can all benefit, upgrading from "working alone" to "team collaboration."

It is recommended to use this tool when starting a new project or breaking down complex requirements. Note that it primarily relies on Claude's Agent capabilities, and the Leader must strictly follow the rule of "only assigning tasks, not writing code," otherwise it may disrupt collaboration discipline. Long-term use requires attention to the disk space occupied by the .mycompany/ directory.