Reference definitions for a 50-role agentic AI organization, grouped by operating area and designed for adaptation into Claude, Codex, or other agent runtimes.
Status: reference design. This repo is not an installable
.claude/agentspackage yet. The tracked files group related agent definitions into readable department documents so the model can be reviewed before being split into runtime-specific agent files.
- A three-level organization model: leadership, departments, and specialists.
- 50 agent role definitions with identity, mission, responsibilities, authority, interfaces, success criteria, and boundaries.
- Design rationale for hierarchy, department grouping, role overlap, and governance choices.
agents/
leadership/README.md # 6 strategic orchestration roles
engineering-data/README.md # engineering, infrastructure, security, data, and ML roles
product-marketing-design/README.md # product, research, marketing, growth, and design roles
pm-ops-qa/README.md # delivery, operations, compliance, support, and QA roles
docs/
design-rationale.md # structure, tradeoffs, and source-backed rationale
| Group | File | Roles |
|---|---|---|
| Leadership | agents/leadership/README.md | CEO-style orchestration, technology, product, marketing, operations, and transformation leadership |
| Engineering and data | agents/engineering-data/README.md | frontend, backend, mobile, AI, DevOps, rapid prototyping, security, data engineering, analytics, and ML |
| Product, marketing, and design | agents/product-marketing-design/README.md | product strategy, research, feedback, sprint planning, marketing channels, growth, content, analytics, UI, brand, and visual design |
| PM, operations, and QA | agents/pm-ops-qa/README.md | experiments, shipping, production coordination, support, infrastructure, finance, compliance, people operations, testing, performance, and accessibility |
Use this repo as a source model, not as a drop-in runtime package:
- Pick the operating areas that match your project.
- Copy only the roles you need.
- Adapt tool access, escalation triggers, and decision authority to your environment.
- Split grouped definitions into runtime-specific agent files only after the role boundaries are clear.
- Keep humans responsible for strategy, safety, spending, legal, security, and external commitments.
The model uses a 3-tier hierarchy:
Human stakeholder
-> leadership agents
-> department/specialist agents
The hierarchy is intentionally shallow. Leadership roles coordinate direction and escalation. Specialist roles own concrete work surfaces. The design avoids deep chains of delegation because long agent chains make accountability, context, and verification harder to maintain.
See docs/design-rationale.md for the detailed rationale.
- The repo is a reference design, not a packaged Claude Code or Codex skill distribution.
- Agent definitions are grouped by department rather than emitted as one file per runtime agent.
- The roles are generic starting points; real deployments should tighten tools, permissions, context boundaries, and verification rules.
Do not paste private customer data, internal strategy, secrets, credentials, or production incident details into public issues. See SECURITY.md for reporting guidance.
Focused improvements are welcome: clearer boundaries, safer escalation rules, better role names, and source-backed design rationale. See CONTRIBUTING.md.
Apache License 2.0 - Copyright 2026 Jarel Remick.