What this is not
What is the difference between an agent skill and an enforcement runtime?
This is not a claim that installing a skill can stop an unauthorized action. A skill can describe permit, deny, escalate, and error outcomes, but an agent or process that ignores those outcomes can still act.
This is not a registration claim, a security certification, or a promise that an agent becomes safe by reading or using a skill. HANRIA is a coined name with no expansion published here.
The distinction
An agent skill is an instruction and checking layer. It can define vocabulary, inspect a structured mandate, apply ordered clauses, identify an expiry or required human decision, and return and record a local result. Those checks advise an agent about what it should attempt. They do not stop, block, prevent, control, or mediate any path by which software can perform the action.
An enforcement runtime is an execution layer. For a protected operation, the runtime would sit on the path to the enrolled tool, credential, transaction, or other action, evaluate the applicable authority, and permit, deny, or escalate before the operation proceeds. Its value depends on the protected surface actually being reachable only through that boundary.
The difference is therefore about where the decision takes effect. A skill advises an agent. A runtime is designed to mediate execution. A successful skill check is a local decision record. It is not evidence that an action was authorized or that an external system could not be reached another way.
A practical test
Ask what happens when the caller ignores the result. If the caller can continue by choosing another tool, credential, process, or path, you are looking at advisory guidance. If the protected operation cannot proceed except through the checked boundary, you are looking at an enforcement design.
Keep the two layers separate in documentation and review. The skill can make a mandate explicit and evaluate one proposed action against it. It keeps no cross-action state and does not detect aggregate authority drift. It cannot turn a convention into a control boundary by itself.
Entry action
Use the hosted check through MCP at https://check.hanria.ai/v1/mcp or HTTP at https://check.hanria.ai/v1/check. Read the instructions, schemas, examples, and limitations at https://hanria.ai/developers/. A local result does not replace an execution boundary.
Honest verdict
Use the skill when you need explicit instruction vocabulary and local checks around delegated authority. If your requirement is to make an unauthorized operation technically unable to proceed, a skill alone is the wrong layer. That claim requires an independently verified execution boundary on every path to the protected operation.