Test Cases
Three complete agents, each packaged as a single bundle file: the System Prompt, its Skills (SKILL.md), and its Tool Schemas. They are inputs, not results — the same case can be inspected by the lite skill, the CLI, or the platform, under any rule version.
Each bundle is a Fixed Artifacts snapshot: the post-remediation version of an agent whose original defects had already been repaired. That is why the cases themselves no longer carry blocking defects — what differs between tools is what they find on a clean artifact, not whether the artifact is broken.
| Case | Domain | Artifacts | Notes |
|---|---|---|---|
| Energy Management Assistant | Energy | 1 Prompt · 3 Skills · 6 Tools | Read-only over a local dataset; AC-L2 prompt, OR-L1 skills/tools |
| Freight Logistics Assistant | Freight logistics | 1 Prompt · 3 Skills · 9 Tools | Read-only over a local dataset; AC-L2 prompt, OR-L1 skills/tools |
| Medication Information Assistant | Healthcare | 1 Prompt · 1 Skill · 5 Tools | Read-only over a local dataset; AC-L2 prompt, OR-L1 skills/tools |
How to use a case
- With the lite skill — hand the whole bundle to
sanityops-inspect-lite; it segments the file into a Prompt, Skills, and Tool Schemas itself, and reports per rule ID. - With the CLI — split the bundle into loadable per-type files first (System Prompt →
prompt.md, eachSKILL.md, each toolparametersschema), then runsanityops-cliat the desired check level. - As a reference — the bundles show how a Prompt, its Skills, and its Tool Schemas reference each other (tool names, parameter names, return fields), which is exactly what the cross-artifact checks operate on.
Results produced from these cases are under Results.