skillful-subagent-creator

To build a Claude subagent, define its role and select 2-5 core skills for preloading alongside additional skills for a dynamic catalog. The process involves writing a structured 4-section prompt covering identity, skill usage rules, task handling loop, and constraints.

DAG orchestration in agent workflows coordinates specialized subagents using patterns like Single Specialist, Chain, Fan-out/Fan-in, Loop, and Human-in-the-Loop. It relies on defining input/output JSON contracts to wire subagents into directed acyclic graph executions.

Defining input and output contracts for agent orchestration requires specifying structured JSON schemas that subagents consume and produce. This contract integration enables reliable DAG workflow patterns by ensuring data handoffs between orchestrated agents remain strictly typed.

A subagent should preload 2-5 core skills for prompt engineering to maintain operational focus. Additional skills can be placed in a dynamic catalog for on-demand retrieval, preventing scope creep while preserving specialized task handling.

Common anti-patterns when creating subagents include deploying skill-less agents that lack curated capabilities and allowing scope creep where agents exceed their defined role. Identifying these issues during testing protocols ensures subagents remain constrained and effective.

Yes, you can integrate subagents into a CI/CD pipeline by defining their role, selecting appropriate skills, and specifying input/output JSON contracts. This allows specialized agents, such as code reviewers, to operate effectively within automated DAG workflows.