Artifact database access refuses risky notification-driven edits and unanswerable prompts

What

Artifacts that keep a shared database are read and changed with the read_db and write_db actions. Two kinds of calls are now handled more strictly:

• A write_db edit prompted by the new-comments notification, where the edit reads back rows written by the artifact's viewers, now goes through a check that answers "Nothing was written". The confirmation text names the risk.
• Some read_db and write_db calls that used to ask for permission now fail with an explanation instead. This covers reads triggered by a notification, and reads of an artifact you do not own or whose owner could not be confirmed.
• That failure happens when nobody is there to answer a permission prompt: a Cowork turn you did not start yourself, or plan mode. The message tells Claude to raise the matter with you in the chat.
• The consent checks were combined into one shared helper.

Why

Comments on an artifact are written by other people, so an edit that acts on what viewers wrote could carry instructions you never gave. Such edits are now blocked. Sessions that run without you watching no longer wait on a prompt nobody can answer, and Claude is told to ask you in the chat instead.