What problem does it solve? Agents built on the AgentOS platform are invisible to the eval suite until someone writes test cases for them. This Skill turns what an agent promises into concrete, runnable Case entries in evals/cases.py, grounded in real session data rather than guesswork. ## Core Features & Use Cases - Promise Mapping: Reads an agent's source file or published Studio config and converts every "always", "never", and "use X for Y" instruction into a testable case. - Session Mining: Queries Postgres for real user sessions and past eval runs to derive realistic test inputs and scenarios. - Case Authoring and Auditing: Writes Case entries with correct hooks, tags, and judge criteria, then runs them repeatedly to detect flaky or undecided rubrics. - Use Case: After building a customer-support agent in AgentOS Studio, run this Skill to generate smoke-tagged eval cases covering its core refund and escalation behaviors, then verify each case passes consistently before release. ## Quick Start Ask the agent to create eval coverage for your newly built agent, for example: "Create evals for my support agent covering its refund and escalation behavior."
