org-agent-architecturelisted

Scaffold, decompose, validate, and maintain organizational agent hierarchies under the AGENT → ROLE → SKILL → TOOL convention (one primitive, four kinds, authority blocks, agents.lock). Use this skill whenever the user wants to create or edit an AGENT.md, ROLE.md, SKILL.md, TOOL.md, or NOUN.md; define what an agent owns, decides, escalates, or must never do; set up or reorganize a repository of agents, roles, skills, or tools; wrap an MCP server or API as a tool; split a role into sub-roles or extract a shared skill; review an agent architecture for ownership conflicts or authority gaps; or generate or update agents.lock entries — even if they don't name the convention explicitly and just say things like "give this agent a job description," "what is this agent allowed to do on its own," or "add a new tool for X."
Chuck3PointZero/OAA · ★ 1 · AI & Automation · score 70

Install: claude install-skill Chuck3PointZero/OAA

# Organizational Agent Architecture This skill applies the Organizational Agent Architecture (OAA) convention for structuring multi-agent systems the way companies are structured: agents fill roles, roles require skills, skills require tools, and bounded authority composes down the chain. The full rationale lives in the convention's proposal document (PROPOSAL.md, where the repository includes it); this skill is the working procedure. ## Core Model - **MUST treat the typed filename** (`AGENT.md`, `ROLE.md`, `SKILL.md`, `TOOL.md`) **as the node's identity** — never the directory path or directory name alone. Every node shares one frontmatter schema: `kind`, `name`, `description`, `requires`, `authority`. Two valid forms per kind; MUST NOT invent a third: - **Directory form** (default): typed file inside a named directory — use when the node needs sub-folders (`memory/`, `decisions/`, `server/`) or supporting assets. - **Flat-file form**: `<name>.agent.md` / `<name>.role.md` / `<name>.skill.md` / `<name>.tool.md` — use when the node is self-contained with no sub-folders. - **MUST use `requires` as the only edge**, pointing downward only; MUST NOT allow any edge that points upward (tool→skill, skill→role) or laterally. Kind aliases: agents may spell it `fills`; skills may spell it `allowed-tools` (agentskills.io compatibility). - **MUST apply union composition for authority**: `never` accumulates by union (deny wins everywhere — no layer can remove a `never` declared belo