AI

Assistant Governed Writes

Open write tools to the in-app assistant in a way that is enforced where the data is (Core approvals at the model), confirmed by the person in the conversation, bounded by the acting user's permissions, and hard to talk out of its rules. Remove the unfed ActionRequest approval path that this replaces.

State
Done
Progress
100%
Steps
42 / 42
Started
2026-10-07
Finished
2026-10-07
Module
LaraAI

Tasks.

  • Task 1: remove the ActionRequest path4/4
  • Task 2: wildcard tool names in policy rule sets3/3
  • Task 3: the governed_writes capability, in-app only4/4
  • Task 4: what the provider offers (opt-in, no voting, budget)4/4
  • Task 5: write proposals and the person's confirmation4/4
  • Task 6: wire respond(), metadata and a guarded claim4/4
  • Task 7: who is acting, and what may be asked4/4
  • Task 8: confinement and injection hardening5/5
  • Task 9: audit3/3
  • Task 10: documentation and closing3/3
  • Final verification4/4

Read the full plan on GitHub