Agent Harness Construction

If you're building agents that call tools and you're tired of watching them retry the same failed action or pick the wrong tool entirely, this helps you redesign the harness. It's focused on the mechanics: how you structure tool schemas, what you put in observations, how you signal recovery paths, and when to use micro-tools versus macro-tools. The guidance is opinionated about things like status enums, next_actions hints, and context budgeting strategies. Most useful when you're past the prototype stage and debugging why your completion rate is stuck at 60% instead of 90%. Think of it as refactoring advice for the layer between your agent's reasoning loop and the actual tools it calls.

Install to Claude Code

npx -y skills add affaan-m/ecc --skill agent-harness-construction --agent claude-code