POMASA
A pattern language and generation toolkit for building multi-agent systems, installed as a skill for Claude Code, Cursor and similar AI coding agents, which guides the agent through constructing a new system from a catalog of reusable architectural patterns.
POMASA is a pattern language and code-generation toolkit for building declarative multi-agent systems. It packages reusable architectural patterns, extracted from real systems, into a pattern catalog together with a generator prompt that guides an AI coding agent through applying those patterns to a new system.
Patterns are organized across three architectural layers: a definition layer, where blueprints describe agent behavior and reference data supplies domain knowledge; an execution layer, where a runtime executes blueprints and an orchestrator coordinates workers through staged pipelines; and a data layer, where a file system acts as the data bus as material moves from drafts to final output. Patterns carry necessity levels of required, recommended or optional, so a system can be composed flexibly from them.
It installs as a skill for Claude Code, Codex, Cursor, Cline and similar agents.