Problem
An agent can have useful tools while it runs in the playground, but those tools are not necessarily available when the same configured agent runs from another entry point, such as an automation, schedule, event, or API invocation.
Some operational tools need to remain available across run contexts. Examples include:
- running a test with
test_run
- renaming the current session
- creating or managing automations
- retrieving the canonical URL for the current session
- optional utilities such as web search
Users need to opt specific tools into the agent configuration so the committed agent can use them outside the playground.
Requested behavior
Let users add optional operational tools to an agent's saved configuration. The selected tools should be available whenever that agent runs, including outside a playground session, subject to the configured permission policy.
Tool selection should remain granular. Adding one operational tool should not require enabling every tool in the same collection.
Acceptance criteria
- A user can add an optional operational tool to an agent's saved configuration.
- After the configuration is committed, the agent can call that tool from a non-playground run.
- The same configured tool is available in playground and non-playground runs.
- Tools that were not selected are not added to the run.
- Existing tool permission rules still apply in every run context.
Implementation note
Using the Build Kit and its individual tool controls is one possible implementation. This is a suggestion, not a requirement. The final design may use a separate optional operational-tool mechanism, especially if tools such as web search should follow the same model.
Related work
- AGE-4123 added individual Build Kit tool activation controls in the playground.
- AGE-4029 fixed missing read, shell, edit, and write tools for Pi agents outside the playground.
- AGE-4339 requests a read-only tool that returns the current session URL.
Problem
An agent can have useful tools while it runs in the playground, but those tools are not necessarily available when the same configured agent runs from another entry point, such as an automation, schedule, event, or API invocation.
Some operational tools need to remain available across run contexts. Examples include:
test_runUsers need to opt specific tools into the agent configuration so the committed agent can use them outside the playground.
Requested behavior
Let users add optional operational tools to an agent's saved configuration. The selected tools should be available whenever that agent runs, including outside a playground session, subject to the configured permission policy.
Tool selection should remain granular. Adding one operational tool should not require enabling every tool in the same collection.
Acceptance criteria
Implementation note
Using the Build Kit and its individual tool controls is one possible implementation. This is a suggestion, not a requirement. The final design may use a separate optional operational-tool mechanism, especially if tools such as web search should follow the same model.
Related work