pick-harness-shape

What problem does it solve?

This Skill helps design the structure and capabilities of a harness for LLM/agent systems, ensuring reliability and efficiency beyond conversational use.

Core Features & Use Cases

• Custom Harness Decision: Determines if a custom harness is needed based on the project's requirements.
• Substrate Selection: Chooses the appropriate substrate for the harness, such as Claude Agent SDK, OpenAI Agents SDK, or Google ADK.
• Topology Design: Defines the topology, including the division of labor and improvement/recovery loop.
• Memory & State Management: Manages how memory and state are handled within the system.
• Tool Layer Configuration: Configures the tool layer, including the number of tools, MCP server, and permissions.
• Verification & Recovery: Sets up verification gates, recovery strategies, and budgeting for the system.
• Security & Trust Boundaries: Establishes security measures and trust boundaries for the system.
• Model Choice & Sampling: Selects the appropriate model tier, temperature, reasoning effort, and token limits for each stage.
• Observability & Evaluation: Defines the posture for trajectory instrumentation, success signals, harness-level eval, and production monitoring.
• Artifact Creation: Generates a comprehensive artifact detailing all design decisions.
• Hand-off & Next Steps: Guides the user through the hand-off process and prompts for further actions.
• Use Case: Ideal for software engineers and product managers designing complex LLM/agent systems that require high reliability and performance.

Quick Start

Run the skill to start designing the harness for your LLM/agent system.