What problem does it solve?
Prompts, instructions, agents, subagents, and skills are hard to verify: authors cannot tell whether an artifact actually behaves as intended at its target reasoning profile. This Skill exercises those artifacts through black-box scenarios, captures containment evidence, and produces an independently graded behavior report.
Core Features & Use Cases
- Black-box behavior testing: Designs scenarios that exercise an artifact only through its documented interface, without leaking internal structure or test framing.
- Fidelity control: Runs contained simulation via the HVE Artifact Tester by default, or approved native execution when read-only or sandbox containment preconditions are met.
- Independent grading and durable reports: Dispatches a separate grading subagent and writes a durable report with verdict, findings by action category, fidelity limitations, and a human-review checkbox.
- Correction-run reuse: Reuses prior passing evidence only when every equivalence dimension is traceable, re-executing affected scenarios.
- Use Case: After authoring a new Copilot agent, invoke this Skill to simulate its behavior at the Medium profile, receive a Pass/Revise/Blocked verdict with severity-ordered findings, and apply corrections before release.
Quick Start
Ask the assistant to run the hve-builder-tester skill against your prompt or agent file at its intended profile and produce a behavior report.
