Orgs is a skillset for AI agents that simulates an engineering organization. The goal is to guide agents through a process of software development similar to humans, exploiting agent capabilities like forking to do certain things humans can’t, so a human user can apply intelligence with discretion across the breadth of a problem.

Orgs includes several skills that interact together towards this end:

• spec: A CTO agent that talks to you to figure out the high level thing you are trying to build and orchestrate further work.
• decompose: A skill to break a high-level design doc into smaller workstreams.
• implement: A skill to have an agent implement a specific change within a certain scope, simulating how teams will attempt to not edit areas of the codebase owned by other teams.
• council: A skill to have more thorough review from a panel of different models, currently supporting the codex, claude, and agy harnesses.
• integrate: A skill to have a team lead supervise merging and maintain the quality of the codebase.
• standup: A skill that has agents check in with the work that they’ve done and receive notes from a “standup meeting” via forking.
• crystal: A skill to duplicate the Crystal system and notify agents when their changes have merge conflicts with parallel implementation agents.
• retro: A skill to examine what went right and wrong in implementation and update agent instructions.
• sprint: A skill that acts as a runbook for all of the other skills, composing them as needed for the given task scope.