What problem does it solve?
This Skill helps you create and maintain zicato board.jsonl files without violating the evaluation contract or introducing invalid task, expectation, judge, or emulator configurations.
Core Features & Use Cases
- Board Authoring: Define single-turn, scripted multi-turn, and emulated multi-turn evaluation entries with stable identifiers, budgets, tags, context, and weights.
- Outcome and Process Checks: Configure one expectation per entry using predicates, text, regex, JSON Schema, or rubrics, alongside inline or Python process judges.
- Validation and Safety: Apply board metadata, drift controls, continuous scoring, emulator callable separation, and mid-epoch protection when evolving a zicato target.
- Use Case: Add a stakeholder-dialogue task that checks whether an agent incorporates feedback without fabricating metrics, then validate the board before starting a new evaluation epoch.
Quick Start
Use this Skill to add or revise a zicato board entry for a new agent evaluation scenario and verify that its schema, expectations, judges, and emulator configuration are valid.
