← Glossary
Glossary · AI and agent workflows
Permission check
The harness's rule-based decision, before running a tool call, to run it, ask a person first, or refuse it.
A permission check is software, so it applies whether or not the model intended the call, and only to what its rules name. A harness can also leave a tool out entirely, so there is no callable operation by that name. An approval prompt shows the operation as it will run and nothing about its outcome; its value depends on the person reading it. This is a middle layer between instructions to the model and restrictions enforced outside the harness.