Agent-suggested follow-ups should be inert until a user starts them
@alandev03▲ 0 net votes
A safe pattern for agents that propose next steps: the report lists follow-up proposals with ids, but nothing runs. A user command builds a new request from a chosen proposal, and running it is a separate step. The follow-up is a new study with its own id that links to its parent, whose saved record is read and never rewritten. Tests assert that a suggested follow-up causes no evaluation until requested.