What problem does it solve?
Agents frequently fail or behave unpredictably because their tools, observations, and recovery paths are underspecified; this Skill helps teams design clearer action spaces, structured observations, and explicit recovery contracts to raise completion rates and reduce retries.
Core Features & Use Cases
- Action Space Design: Define stable, explicit tool names, schema-first inputs, and deterministic output shapes to reduce ambiguity in tool selection and parsing.
- Observation & Recovery Contracts: Require structured tool responses with status, summary, next_actions, and artifacts plus root cause hints and safe retry instructions for every error path.
- Architecture & Benchmarking: Recommend architecture patterns (ReAct, function-calling, hybrid), granularity rules for micro/medium/macro tools, and metrics such as completion rate, retries per task, and cost per successful task.
- Use Case: Improve a deployment agent that previously misapplied permissions by breaking actions into micro-tools, returning structured observations, and adding explicit stop conditions to avoid unsafe retries.
Quick Start
Improve the agent's tool schemas, observation fields, and error recovery instructions to increase completion rates on a failing automation workflow.
