The Console distinguishes instructions, selected context and effectively authorized capabilities. This helps diagnose a missing tool or an answer using the wrong evidence.
Prerequisites
Open an agent you may edit. Choose a reading mission and a question with a known result. Before querying Client and Commande, inspect their instances in the same workspace.
1. Choose a capability profile
Select an appropriate Capability profile. A reading profile retrieves data. Approved actions require a compatible profile and capability; availability depends on execution context.
Tools are filtered by the profile and permissions. The product's overall tool count does not describe what your agent can use.
2. Select tools and APIs
Enable useful capabilities in Tools and API. For a business API, select a published operation and check its authorized use. Reading and preparing an action have different effects. The full internal catalogue is not the public contract.
| Need | Check before testing |
|---|---|
| Read orders | Target type, filters and reading capability |
| Consult a contract | Processed document, search scope and returned evidence |
| Prepare an action | Compatible published operation and verifiable inputs |
| Use a skill | Relevant instructions and already authorized capabilities |
The effective selection is limited by authorized capabilities. Selecting an option does not grant permission.
3. Choose context and sources
In Context and sources, choose relevant model elements and documents. A narrower context improves targeting; access boundaries depend on workspace permissions and enforcement. “Only consult Client” in a prompt is not a security mechanism.
Specify the customer identifier and period for the Client–Commande mission. Include the contract only when the question requires documentary terms.
4. Verify the applied configuration
Save and open Verification. Compare the draft, saved configuration and effectively applied configuration. Distinguish enforced constraints from advisory settings. Permissions are checked again when a run starts.
Run a reading question and inspect the tool call: capability name, arguments, result and errors. If the expected tool is missing, check the profile, selection, API publication and permissions before changing instructions.
Expected result
You can explain which capabilities the agent can use and have observed them on a known case. Continue with Test and evaluate and Prepare a controlled action.
