oc-agent-forge

What problem does it solve?

This Skill streamlines the development of Claude Agent SDK applications by providing a comprehensive framework that includes a Planner/Builder/Evaluator loop, subagent topology management, tool budgets, harness loop shapes, and agent evaluation. It addresses the challenges of designing, building, and evaluating multi-agent systems, enabling efficient development of robust and efficient Claude Agent SDK apps.

Core Features & Use Cases

• Planner/Builder/Evaluator Loop: Facilitates the systematic development of Claude Agent SDK apps, from planning the agent topology and tool budget to building the harness and evaluating the agent.
• Subagent Topology Management: Offers the flexibility to design and manage subagent topologies for single-agent, orchestrator-worker, pipeline, and hierarchical architectures.
• Tool Budgets: Allows the definition of tool allowlists, call ceilings, and deferred-loading for efficient tool usage and cost management.
• Harness Loops: Provides options for react, plan-execute, and reflect/critique loop shapes to accommodate various task structures and requirements.
• Agent Evaluation: Evaluates the agent's performance against a fixture suite, considering task success, trajectory, tool efficiency, and cost.
• Use Case: Suppose you need to develop a Claude Agent SDK app for a multi-step task that involves research, data processing, and decision-making. Use this Skill to design the agent's topology, define the tool budget, build the harness, and evaluate the agent's performance.

Quick Start

Use the /oc-agent command to design, build, and evaluate a Claude Agent SDK app.