AI Assistant loses its code-editing tools on follow-up prompts

During an AI Assistant conversation with code editing enabled in Settings, the assistant can edit project files on one prompt and then report that it cannot write files on the next prompt. Nothing changed in the settings between the two prompts.

The cause is that Studio re-selects the assistant's tool list for every prompt by keyword-matching the wording of that single message. The code-editing tools are only included when the message contains an English action verb (add, create, edit, fix, update, ...) or a short approval such as "yes" or "do it". A follow-up such as "continue", "keep going", "the same for the other windows", or any message written in another language matches none of those words, so the editing tools are silently omitted from that request. The model then truthfully reports that it cannot write files, and if it tries anyway the call is rejected. Re-sending the request with an action verb brings the tools back, which makes the failure look intermittent.

Expected behaviour: while code editing is enabled in Settings > AI Assistant, the editing tools stay available for every prompt in the conversation, independent of how each message is worded. Message wording may add specialist tools on top of the enabled set, but must never remove a capability the user has explicitly enabled — the same rule the web-search tools already follow.

Fix delivered in Objo Studio: editing tools are now offered whenever code editing is enabled; the active-document kind (project source versus IDE Script) decides which editing surfaces are offered, with both offered when both apply and wording able to add, not remove, a surface. Keyword matching no longer gates any write capability, so prompting in any language works.