[
https://issues.apache.org/jira/browse/CAMEL-25441?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18127029#comment-18127029
]
Claus Ibsen commented on CAMEL-25441:
-------------------------------------
Won't fix as proposed. Camel TUI is a developer tool: it runs on the
developer's own machine, for their own integrations, and the AI panel (F8) is
built for a fast loop where the model shows and tells at once. A permission
popup before every tool call that changes something defeats that, so the
panel's own model does not get one. The ACP path now follows the same rule for
the TUI's own tools, which gives the parity this issue asked for, the other way
round.
Instead the TUI asks the user itself, for every caller alike (the panel's
model, an ACP agent, an MCP client), before the few calls that reach beyond the
integration, may destroy data or run code:
* {{tui_send_message}} to an endpoint the integration does not have; internal
endpoints such as {{log:}}, {{direct:}}, {{seda:}} and {{mock:}}, and the
integration's own endpoints, go ahead
* {{tui_execute_sql}} with a statement other than a single SELECT, SHOW or
DESCRIBE, and {{tui_update_row}}
* {{camel_eval_expression}} in a language that runs code, such as Groovy
(CAMEL-25443)
* file writes, through the write mode as before, only inside the integration's
source directory
An ACP agent's own tools (its shell commands, file changes and fetches) still
go to the Permission popup. The security model has a new "Developer tools"
section with this contract, and CAMEL-25652 proposes an {{internal}} flag in
the component metadata to replace the TUI's own list of internal components.
Commit:
https://github.com/apache/camel/commit/c754310d6f88194d967d78028498840f7a5f3a00
_Claude Code on behalf of Claus Ibsen_
> camel-jbang-plugin-tui - confirm AI panel tool execution for parity with the
> ACP permission flow
> ------------------------------------------------------------------------------------------------
>
> Key: CAMEL-25441
> URL: https://issues.apache.org/jira/browse/CAMEL-25441
> Project: Camel
> Issue Type: Improvement
> Components: camel-jbang
> Reporter: Andrea Cosentino
> Assignee: Claus Ibsen
> Priority: Minor
>
> In the Camel TUI's built-in AI panel (direct LLM providers), the tools the
> model requests are executed immediately (AiPanel.runAgentLoop ->
> executeTuiTool -> TuiToolRegistry.execute). The ACP agent path, by contrast,
> routes every tool that is not in TuiToolRegistry.READ_ONLY_TOOLS through a
> user permission popup (AiPanel.AcpPanelPermissionHandler).
> For consistency, and to keep the user in control of the actions the assistant
> performs (writing files, controlling integrations, sending messages, etc.),
> the built-in panel should apply the same policy: prompt for confirmation on
> tools that are not read-only, reusing the existing READ_ONLY_TOOLS set and
> the permission popup. This makes the two AI entry points behave the same way.
> Module: dsl/camel-jbang/camel-jbang-plugin-tui (AiPanel).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
