Agent Primitives: Modular Infrastructure for Claude Code Multi-Agent Workflows

The agent-primitives project provides reusable agent definitions and stack configuration schemes for Claude Code multi-agent development. Using a layered architecture design, it addresses the pain points of scattered agent configurations and difficulty in reuse, helping projects evolve in a standardized and scalable way.

With the popularity of the AI programming assistant Claude Code, multi-agent collaborative development models have emerged. However, agent definitions are scattered everywhere, lacking unified standards and reusable components. The open-source agent-primitives project by mumoryan aims to solve this problem by providing reusable basic agent definitions and stack configuration schemes, allowing teams to share core capabilities across multiple projects and avoid reinventing the wheel.

agent-primitives adopts a three-layer architecture:

1. Base Layer (base/): Toolchain is project-independent, including identity definition, capability contracts, and output modes—it is the cornerstone of the architecture;
2. Stack Layer (stacks/): Reusable across projects, including tech stack-specific coding standards, framework best practices, and general tool conventions;
3. Project Layer (.claude/agents/): Personalized customization, including project-specific path constraints, locked architecture decisions, and functional agent definitions.

agent-primitives recommends the following usage patterns:

• Project-level agents extend base definitions via stub files (/.claude/agents/);
• Execute the merge-agent.sh script to merge base+stack+stub into a complete agent definition. Static parts are merged first to support prompt caching optimization.

1. Unified code standards across projects: The base layer defines general review agents, the stack layer adds Python/React-specific standards, and the project layer configures paths;
2. SaaS multi-tenant architecture: The base layer defines tenant isolation check agents, the stack layer configures database access modes, and the project layer customizes tenant verification rules;
3. Enterprise-level development process standardization: Establish an enterprise-level security review agent library, define compliance basic capabilities, and business lines extend department rules.

• Maintainability: The scope of impact of changes is controllable (toolchain upgrades modify the base layer, tech stack migrations change the stack layer, project requirement changes adjust the stub layer);
• Reusability: Core capabilities are shared across projects and have been reused in projects like Loci;
• Extensibility: New agents create definition files by layer and integrate via the merge script;
• Performance Optimization: Static merging supports prompt caching, improving the response speed of multi-turn dialogues.

agent-primitives provides a clear and extensible architecture blueprint for Claude Code multi-agent development, solving the problems of scattered configuration and reuse. It is recommended that teams currently using or planning to use Claude Code for multi-agent development take it as an infrastructure reference. It not only provides ready-made configuration schemes but also demonstrates a sustainable evolution method for multi-agent engineering.