Capability-of-Entry (CoE) Gate Dataset: An Initial 30-Scenario Calibration Set for AI Agent Action Readiness

This release introduces an initial 30-scenario calibration dataset for the Capability-of-Entry (CoE) Gate, a proposed decision layer for evaluating whether an AI agent should enter, limit, defer, block, or seek approval before executing an action. The dataset covers agentic decision points across baseline organizational tasks, customer support, sales and commercial commitments, legal/document workflows, privacy/data-leakage cases, and one hybrid access-control scenario. Each scenario includes a user request, action type, execution mode, contextual metadata, an expected gate decision, and explanatory behavioral contrasts between naive and gated agent behavior. This public v0.1 release includes a sanitized scenario dataset and a public validator script for schema integrity and label-distribution checks. It intentionally excludes private scoring fields, internal thresholds, risk-class logic, policy-class logic, and the private decision engine. The purpose of this release is to provide a citable, inspectable research/demo artifact for studying action-readiness gating in agentic AI workflows, not a production safety system.